Datasets:

WINGNUS
/

ACL-OCL

	{
	"paper_id": "2020",
	"header": {
	"generated_with": "S2ORC 1.0.0",
	"date_generated": "2023-01-19T13:28:55.593665Z"
	},
	"title": "In Media Res: A Corpus for Evaluating Named Entity Linking with Creative Works",
	"authors": [
	{
	"first": "Adrian",
	"middle": [
	"M P"
	],
	"last": "Bra\u015foveanu",
	"suffix": "",
	"affiliation": {
	"laboratory": "",
	"institution": "MODUL Technology GmbH Vienna",
	"location": {
	"country": "Austria"
	}
	},
	"email": "adrian.brasoveanu@modul.ac.at"
	},
	{
	"first": "Albert",
	"middle": [],
	"last": "Weichselbraun",
	"suffix": "",
	"affiliation": {},
	"email": "albert.weichselbraun@fhgr.ch"
	},
	{
	"first": "Lyndon",
	"middle": [
	"J B"
	],
	"last": "Nixon",
	"suffix": "",
	"affiliation": {
	"laboratory": "",
	"institution": "MODUL Technology GmbH",
	"location": {
	"settlement": "Vienna",
	"country": "Austria"
	}
	},
	"email": "lyndon.nixon@modul.ac.at"
	}
	],
	"year": "",
	"venue": null,
	"identifiers": {},
	"abstract": "Annotation styles express guidelines that direct human annotators by explicitly stating the rules to follow when creating gold standard annotations of text corpora. These guidelines not only shape the gold standards they help create, but also influence the training and evaluation of Named Entity Linking (NEL) tools, since different annotation styles correspond to divergent views on the entities present in a document. Such divergence is particularly relevant for texts from the media domain containing references to creative works. This paper presents a corpus of 1000 annotated documents from sources such as Wikipedia, TVTropes and WikiNews that are organized in ten partitions. Each document contains multiple gold standard annotations representing various annotation styles. The corpus is used to evaluate a series of Named Entity Linking tools in order to understand the impact of the differences in annotation styles on the reported accuracy when processing highly ambiguous entities such as names of creative works. Relaxed annotation guidelines that include overlap styles, for instance, lead to better results across all tools.",
	"pdf_parse": {
	"paper_id": "2020",
	"_pdf_hash": "",
	"abstract": [
	{
	"text": "Annotation styles express guidelines that direct human annotators by explicitly stating the rules to follow when creating gold standard annotations of text corpora. These guidelines not only shape the gold standards they help create, but also influence the training and evaluation of Named Entity Linking (NEL) tools, since different annotation styles correspond to divergent views on the entities present in a document. Such divergence is particularly relevant for texts from the media domain containing references to creative works. This paper presents a corpus of 1000 annotated documents from sources such as Wikipedia, TVTropes and WikiNews that are organized in ten partitions. Each document contains multiple gold standard annotations representing various annotation styles. The corpus is used to evaluate a series of Named Entity Linking tools in order to understand the impact of the differences in annotation styles on the reported accuracy when processing highly ambiguous entities such as names of creative works. Relaxed annotation guidelines that include overlap styles, for instance, lead to better results across all tools.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Abstract",
	"sec_num": null
	}
	],
	"body_text": [
	{
	"text": "Identifying all entities from texts and linking them to modern Knowledge Graphs (KGs) like DBpedia and Wikidata is the core task of Named Entity Linking (NEL) (Rosales-M\u00e9ndez et al., 2018b) . A particular challenge in NEL is disambiguating the entity references in the surface form (original text). For example, mentions such as NYT or NY Times or even Times could all be surface forms referencing the entity The New York Times and they can then be linked to the associated entries from wellknown public KGs like DBPedia and Wikidata (i.e., dbr:The_New_York_Times or wd:Q9684 1 ). In ad-dition to entities, the relations between them, as defined in the texts themselves or extracted from a KG, can also help enlighten the narratives. All of the NEL related tasks are supported by large communities which have a vested interest in advancing their Knowledge Extraction (KE) capabilities. The various good results for general NEL benchmarks (e.g., F1 scores of 0.80 (Rosales-M\u00e9ndez et al., 2018a) ) are encouraging and suggest that the tasks tested for in these benchmarks may be solved soon. However, the main focus of most reported evaluations to date has been on the correct annotation of several specific types of entities like Persons (PER), Locations (LOC) or Organizations (ORG). Sometimes fine-grained typing was also allowed for locations, therefore allowing us to use types like natural locations (LOC) to depict naturally occurring locations like mountains or rivers, facility (FAC) to describe buildings or infrastructure like bridges or airports and Geo-Political Entities (GPE) to describe cities, region or countries (Ji et al., 2017) . Only in the last half-decade the focus has slowly shifted towards expanding the typing to encompass a wider array of types, from dates and products to games, sports, books or movies. A large subset of the fine-grained types of entities found naturally in texts, especially in the media domain, are focused on what can generically be called Works or sometimes Creative Works and which would encompass all creative works (manifestations of creative effort by a creator or group of creators), from books or songs, to movies, television episodes and video games, for example. Collective works with unknown authors (e.g., religious works, folklore), festivals, concerts or sports events can also be considered creative works, even though they may be labelled differently (e.g., a concert might be labelled as an Event, whereas its audio or video recording could be labelled as a Creative Work).",
	"cite_spans": [
	{
	"start": 159,
	"end": 189,
	"text": "(Rosales-M\u00e9ndez et al., 2018b)",
	"ref_id": "BIBREF32"
	},
	{
	"start": 963,
	"end": 993,
	"text": "(Rosales-M\u00e9ndez et al., 2018a)",
	"ref_id": "BIBREF29"
	},
	{
	"start": 1629,
	"end": 1646,
	"text": "(Ji et al., 2017)",
	"ref_id": null
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Introduction",
	"sec_num": "1"
	},
	{
	"text": "The current generation of automated annotators provide a rather mixed set of results for mediarelated annotations due to inconsistent annotation guidelines related to this domain in the past. For example, a TV show like Star Trek: Deep Space Nine will either get no result or two different results, one marking the franchise (Star Trek) and the second the titular space station (Deep Space Nine), but will probably not be annotated with the full correct entity (Star Trek: Deep Space Nine) by many automated annotator tools. Such an annotation could alternatively be considered correct in our view, if instead of the space station, the second annotation would point to the TV show itself. Of course, if we examine the characters from the respective TV shows, we can run into similar problems, as some characters might be described by multiple resources, each of them highlighting a particular performance (e.g., James Bond is described both as literary characterdbr:James_Bond_(literary_character) -and Person dbr:James_Bond). In order to counteract such issues, we have developed a corpus focused around annotating such media-related entities. We share this corpus, together with the corresponding annotation guidelines, as well as a set of alternative annotations, especially for nested entities 2 . Since we plan to continuously update this corpus with new partitions every few months, one of its core concepts is the fact that for each ten partitions created there will be a summary partition, a partition that will combine annotations similar to those from the previous nine partitions. If users aim at performing a quick evaluation, they will be able to use this summary partition which we call 'core set' in order to have a first set of results and decide upon their preferred fine-tuning procedure.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Introduction",
	"sec_num": "1"
	},
	{
	"text": "The expression media annotations is typically used to refer to video or audio annotations, therefore in the context of this paper we have selected the expression media entity annotations to signify the annotation of texts created and/or published by various media conglomerates which contain entities of type Creative Work.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Introduction",
	"sec_num": "1"
	},
	{
	"text": "We consider our work as an important step in the road towards building culturally-aware AIs since a significant part of human culture is represented by the creative works humanity has produced, AIs are needed that can correctly annotate and interpret information about the referenced creative works.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Introduction",
	"sec_num": "1"
	},
	{
	"text": "The rest of the paper is organized as follows: Section 2 showcases somewhat similar work and articles that discuss some of the ideas that we followed during our design process; Section 3 describes the design of the corpora; Section 4 evaluates several tools on some of the partitions included in our corpora; whereas the final section provides a reflection upon the insights gained from designing a gold standard that focuses on creative works.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Introduction",
	"sec_num": "1"
	},
	{
	"text": "Since this dataset uses several concepts that are not widely used in NEL evaluation datasets (e.g., lenses, core sets, multiple annotation styles), this section also includes explanations for those concepts.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "After examining a rich set of media annotation papers we have discovered that very few papers actually discuss how to correctly annotate media assets with creative work entities. Early results were focused on the correct semantic annotation of text extracted from the multimedia content (e.g., transcription or subtitles), as presented in a set of papers that originated from the LinkedTV project like (Li et al., 2013; Nixon and Troncy, 2014) , but using the classical entity types well supported by NEL tools (Person, Organization, Location). More recent media annotation work discussed cross-modal annotations and story identification (Nixon et al., 2019) . None of these projects or papers focused on the correct identification of creative works or the various naming variants involved in creative works referencing. An episode name like The Trouble with Tribbles can be an episode of Star Trek, but also an instance of a chapter, comic book or book in the same franchise. Even though the existing research would help us annotate the respective episode's video, it will not really help us contextualize it, if mentioned in a press release, for example.",
	"cite_spans": [
	{
	"start": 402,
	"end": 419,
	"text": "(Li et al., 2013;",
	"ref_id": "BIBREF21"
	},
	{
	"start": 420,
	"end": 443,
	"text": "Nixon and Troncy, 2014)",
	"ref_id": "BIBREF25"
	},
	{
	"start": 638,
	"end": 658,
	"text": "(Nixon et al., 2019)",
	"ref_id": "BIBREF24"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "Due to the shortcomings of current approaches, we have also examined several other avenues, including corpora that contain creative works or historical entities that might otherwise be difficult to identify by the general public and were created for different purposes but in somewhat similar situations (e.g., no corpora available, specific domain, etc). The digital humanities domain, and in particular historical documents and literary criticism, are perhaps the closest scenarios to our use case that come to mind. Several works in this domain like the historical documents from the Impresso collection (Ehrmann et al., 2020) , the multilingual news corpora MeanTime (Minard et al., 2016 ) and Dekker's work on extracting small snippets of literary criticism from social media (Dekker et al., 2018) have served as a starting point in our journey, helping us to define our annotation guidelines. Some early ideas about naming variations (in particular (Ehrmann et al., 2017) , (Rosales-M\u00e9ndez et al., 2019) and (Weichselbraun et al., 2019b) ), and nested annotations ((Ju et al., 2018) and (Ji et al., 2017) ) have also shaped our understanding of the difficulties of correctly capturing names referencing entities, regardless of the domain, and have led us to multiple works on annotation styles and lenses. The idea of data lenses comes from multiple places, but it was generally inspired by photography where different lenses are used in order to get different views on an object. In NEL, and in NLP in general, the idea is to enable different views onto the same dataset. During the last few years, lenses have increasingly been used to transform data between different views (Rajkumar et al., 2013) . In the NLP domain, lenses have traditionally been implemented as annotation sets that reflect the view of a human annotator upon the specific data, being an important feature of annotation packages like GATE (Maynard, 2009) , as well as for implementation of large parallel corpora typically used for multilingual settings (e.g., (Iranzo-S\u00e1nchez et al., 2019) . In the Semantic Web (SW), ontologies have traditionally been used to act as lenses over data eventually leading to an entire field of study: Ontology-Based Data Access (Calvanese et al., 2015) . More recently, lenses have been used to create multiple views over chemistry data (Batchelor et al., 2014) , help manage large data sets (Lenzerini, 2018) , or understand big data and AI workloads (Gao et al., 2018) .",
	"cite_spans": [
	{
	"start": 607,
	"end": 629,
	"text": "(Ehrmann et al., 2020)",
	"ref_id": "BIBREF9"
	},
	{
	"start": 671,
	"end": 691,
	"text": "(Minard et al., 2016",
	"ref_id": "BIBREF23"
	},
	{
	"start": 781,
	"end": 802,
	"text": "(Dekker et al., 2018)",
	"ref_id": "BIBREF7"
	},
	{
	"start": 955,
	"end": 977,
	"text": "(Ehrmann et al., 2017)",
	"ref_id": "BIBREF8"
	},
	{
	"start": 980,
	"end": 1009,
	"text": "(Rosales-M\u00e9ndez et al., 2019)",
	"ref_id": "BIBREF30"
	},
	{
	"start": 1014,
	"end": 1043,
	"text": "(Weichselbraun et al., 2019b)",
	"ref_id": "BIBREF35"
	},
	{
	"start": 1070,
	"end": 1088,
	"text": "((Ju et al., 2018)",
	"ref_id": "BIBREF19"
	},
	{
	"start": 1093,
	"end": 1110,
	"text": "(Ji et al., 2017)",
	"ref_id": null
	},
	{
	"start": 1683,
	"end": 1706,
	"text": "(Rajkumar et al., 2013)",
	"ref_id": "BIBREF27"
	},
	{
	"start": 1917,
	"end": 1932,
	"text": "(Maynard, 2009)",
	"ref_id": "BIBREF22"
	},
	{
	"start": 2039,
	"end": 2068,
	"text": "(Iranzo-S\u00e1nchez et al., 2019)",
	"ref_id": "BIBREF16"
	},
	{
	"start": 2239,
	"end": 2263,
	"text": "(Calvanese et al., 2015)",
	"ref_id": "BIBREF5"
	},
	{
	"start": 2348,
	"end": 2372,
	"text": "(Batchelor et al., 2014)",
	"ref_id": null
	},
	{
	"start": 2403,
	"end": 2420,
	"text": "(Lenzerini, 2018)",
	"ref_id": "BIBREF20"
	},
	{
	"start": 2463,
	"end": 2481,
	"text": "(Gao et al., 2018)",
	"ref_id": "BIBREF12"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "With respect to entity linking, an early example of lenses was the concept of approximate matching implemented in tools like Neleval (Hachey et al., 2014) and Gerbil (R\u00f6der et al., 2018) . The computation of partial matches was done based on the number of overlapping characters or entity types with respect to the measures that can be optimized (e.g., precision or recall). Many optimization strategies will fail to correctly identify either connections between entities (e.g., parent) or nested entities. What is missing and why, can generally be discovered during the error analysis phase of the evaluations, as shown in (Bra\u015foveanu et al., 2018b) and (Stanislawek et al., 2019) . Understanding error classification schemes like the ones described in these publications can be the key towards moving the field forward.",
	"cite_spans": [
	{
	"start": 133,
	"end": 154,
	"text": "(Hachey et al., 2014)",
	"ref_id": "BIBREF13"
	},
	{
	"start": 166,
	"end": 186,
	"text": "(R\u00f6der et al., 2018)",
	"ref_id": "BIBREF28"
	},
	{
	"start": 624,
	"end": 650,
	"text": "(Bra\u015foveanu et al., 2018b)",
	"ref_id": "BIBREF4"
	},
	{
	"start": 655,
	"end": 681,
	"text": "(Stanislawek et al., 2019)",
	"ref_id": "BIBREF33"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "More recently, lenses have been used to understand entity annotations (Bra\u015foveanu et al., 2018a) and to study the effect of relaxed annotation schemas that include more types than usual (Rosales-M\u00e9ndez et al., 2018a) . A theoretical treatment of automatic procedures for building lenses is presented in (Weichselbraun et al., 2019a) . The paper showcases how to automatically build lenses by following simple rules (e.g., expand the surface form to the maximum length -the longest mention -or reduce the surface form to the minimal possible length -the shortest mention). Using the longest mention leads to an annotation schema with less entities, whereas using the shortest mention will lead to more entities being annotated. Intermediary schemas are also possible, if we take into account the possibilities of combining the previous schemas or if we add additional rules. We test some of the lenses discussed in this paper, as it can be seen in Section 3.",
	"cite_spans": [
	{
	"start": 186,
	"end": 216,
	"text": "(Rosales-M\u00e9ndez et al., 2018a)",
	"ref_id": "BIBREF29"
	},
	{
	"start": 303,
	"end": 332,
	"text": "(Weichselbraun et al., 2019a)",
	"ref_id": "BIBREF34"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "Our In Media Res corpora contains partitions from multiple domains. Due to this aspect, it was decided to introduce core sets partitions at every nine partitions. By adding these core sets, evaluators can later build a summary dataset for the whole corpus, regardless of the number of partitions contained in it. The concept of core sets was borrowed from computational geometry and robotics. A core set is a small engineered subset of a very large dataset that retains its properties as accurately as possible (e.g., a set of points that approximate the shape of a figure). Its size typically depends on the desired accuracy, not necessarily on the size of the original data set. Core sets have been successfully applied to a variety of problems, from dimensionality reduction of massive data sets (Feldman et al., 2016) , to vector summarization (Feldman et al., 2017) or compression of neural networks (Baykal et al., 2018) .",
	"cite_spans": [
	{
	"start": 799,
	"end": 821,
	"text": "(Feldman et al., 2016)",
	"ref_id": "BIBREF11"
	},
	{
	"start": 848,
	"end": 870,
	"text": "(Feldman et al., 2017)",
	"ref_id": "BIBREF10"
	},
	{
	"start": 905,
	"end": 926,
	"text": "(Baykal et al., 2018)",
	"ref_id": "BIBREF2"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "We, therefore, started our corpus design by testing various annotation styles and creating an annotation guideline for Creative Work entities. After we identified a solution, we created multiple partitions from various domains.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "The name of the corpus comes from the Latin expression in media res that refers to narratives (e.g., books or movies) that start in the middle of the story. Similarly, the created corpus has been designed with the aim of understanding different annotation styles, and rather than starting from zero builds upon prior research in this area. Also, the fact that most of the annotated documents are related to media (e.g., franchises, books, TV shows) has been one of the reasons why we have selected this name.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Related Work",
	"sec_num": "2"
	},
	{
	"text": "While classic corpora are quite good for identifying people, organizations or locations, there are less adequate corpora to help with evaluating works (e.g., books, TV shows, music, etc) or events. This has been the main reason why we have decided to create a corpus focused mostly on creative works to help us fine-tune media domain document annotations.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Annotation Styles",
	"sec_num": "3.1"
	},
	{
	"text": "As opposed to the classic entity types in NEL evaluations (e.g., Person, Organization, Location), the annotation of documents from the media domain (TV, radio, film etc.) raises specific challenges. In addition to the core entity types like Person, Organization and Location, media domain document annotation also needs to support a fourth large class of entities: Creative Work or Work 3 . This class encompasses a large selection of entities that might be classified as creative works, from books and songs, through to games, movies, TV Shows and entire media franchises. It is important to note that, if we leave the temporal attributes aside (e.g., new positions for a person, key people for a company, new episodes for a TV show), there is a lot of variation when it comes to the main attributes of this entity type as opposed to the three core types (Person, Organization, Location). A TV show might have some executive producers, a production company, some actors starring in it, as well as a set of episodes, each with their own list of directors, writers or stars; a book will have some author(s), publisher, and awards or links to a book series; and a song might have an interpret, author, music producer, and so on. As it can be seen, it is difficult to find common attributes between the various sub-classes, except for the fact that they are all types of creative works that were published in some format or medium in a certain period of time. It can even be argued that all these creative works should be modeled as their own entity types, but in order to perform such a fine-grained extraction, it is important to first identify the large class to which the entities belong. Adding works can also lead to a high number of false positives, as often fictional characters (e.g., James Bond, Harry Potter) might share names with real people, as well as with their own media franchises which can encompass different sets of series (e.g., books, tv shows, movies, comics, etc); fictional characters might be based on real people (e.g., see the recent trend of music biopics based on N.W.A., Queen or Elton John or TV shows like Narcos who often fictionalize real characters by changing their names or changing the events in which they participate) or the name of a work is later used for a different work (e.g., again the example of music biopics is relevant). Table 1 showcases some of the issues encountered by annotators such as AIDA (Hoffart et al., 2011) , DBpedia Spotlight (Daiber et al., 2013) and Recognyze (Weichselbraun et al., 2019b) when performing named entity linking on works. As illustrated in the table, each annotator tends to return a different set of results based on its settings. Unfortunately, the best settings for annotators are not published, therefore, even when NEL annotators are integrated into benchmarking systems like Gerbil (R\u00f6der et al., 2018) or (Odoni et al., 2018) it is still difficult to understand if the obtained results really represent the best possible outcome. When examining these differences between tools, we decided to use lenses as a method for further investigating overlaps and partial matches.",
	"cite_spans": [
	{
	"start": 2446,
	"end": 2468,
	"text": "(Hoffart et al., 2011)",
	"ref_id": "BIBREF15"
	},
	{
	"start": 2489,
	"end": 2510,
	"text": "(Daiber et al., 2013)",
	"ref_id": "BIBREF6"
	},
	{
	"start": 2868,
	"end": 2888,
	"text": "(R\u00f6der et al., 2018)",
	"ref_id": "BIBREF28"
	},
	{
	"start": 2892,
	"end": 2912,
	"text": "(Odoni et al., 2018)",
	"ref_id": "BIBREF26"
	}
	],
	"ref_spans": [
	{
	"start": 2370,
	"end": 2377,
	"text": "Table 1",
	"ref_id": "TABREF1"
	}
	],
	"eq_spans": [],
	"section": "Annotation Styles",
	"sec_num": "3.1"
	},
	{
	"text": "For the current evaluation, we have considered the following annotation styles based on (Weichselbraun et al., 2019a) (here illustrated on the annotation of the text snippet .",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Annotation Styles",
	"sec_num": "3.1"
	},
	{
	"text": "It has to be noted that two of the styles ( OMAX and OMAX) can also extract the person rather than the TV show Picard as a separate entity, but since this result would have a different type (Person instead of Work, as it points to the Jean-Luc Picard character from the same franchise) and different link (e.g., dbr : Jean \u2212 Luc_Picard instead of dbr : S tar_T rek : _Picard), it would be an incorrect result that is automatically removed in our implementation. The presented rules only consider borderline cases, even though combinations of them can also be used within a corpus. A corpus which would not apply the OMAX rule, for example, might lose the extended reference to Sir Patrick Stewart OBE and only return Patrick Stewart or end up removing the references to the actor's titles (e.g., Sir, OBE). We consider OMAX annotation rule to be the best, as it essentially merges the other annotation styles. The advantage of using these rules comes from the fact that they can easily be automated. By using them we have generated alternative annotations (also known as lenses) for all of our dataset partitions. One such example is provided for the core set of our dataset in the evaluation from Section 4. The annotations in Table 2 illustrate the gold standard results for the different lenses described in this section. The OMIN results are somewhat closer to the expected full annotation (e.g., the one presented in the example column) and only introduce minor variations. One of our assumptions was that lenses such as OMAX should even the playing field by reducing the penalty of overlaps in terms of false positives, while also offering some clues on what kind of surface forms are picked up more frequently by the various annotators.",
	"cite_spans": [],
	"ref_spans": [
	{
	"start": 1228,
	"end": 1235,
	"text": "Table 2",
	"ref_id": "TABREF3"
	}
	],
	"eq_spans": [],
	"section": "Annotation Styles",
	"sec_num": "3.1"
	},
	{
	"text": "The corpora currently has 10 sections each with 100 documents. We plan to add more partitions in time to cover different domains and therefore test different algorithms for domain adaptation in Named Entity Linking.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "At every 900 documents, we include a numbered partition called general (e.g., general-1, general-2, etc) which will contain some documents from each of the domains covered in these 900 documents, therefore representing a summary or a core set of the previous set of partitions. This is done in order to create a large core set of the entire corpora. The General core sets can be used both as smaller standalone corpora, as well as small test beds in order to decide if certain partitions are useful.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "We have started by collecting several sentences from the Wikipedia abstracts of 100 articles about creative works (as classified by their respective DB-Pedia resource). The initial set of entities contained books, TV shows, media companies, YouTube influencers and media franchises. Several entity types were annotated, include Person (PER), Orga- nization (ORG), Location (LOC), Work (WORK), Event (EVENT) or Other (OTHER). The corpus was annotated by two annotators following our annotation guideline. The human annotators were asked to use the OMIN lens for creating the initial annotations, therefore disregarding overlapping entities and selecting the minimum possible number of matches. A judge was available for questions during the whole annotation process and helped solve disagreements after the annotation process has been completed. The rest of the lenses (e.g., OMAX, OMAX) have later been automatically generated using a Python script. The judge has then verified the resulting annotations in order to eliminate mistakes. This process was iterative, therefore most of the errors reported being eliminated from the script until the end of the process. The resulting corpus was exported into multiple formats, including CSV and NIF. The remaining texts were collected from the open source repositories TVTropes 4 and WikiNews 5 . Currently the following partitions are available (we indicate the sources in parentheses):",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Franchises (TV Tropes) set is focused on big multimedia franchises like Marvel Cinematic Universe or Star Wars and the creative works in various formats (movies, TV shows, books, video games) that support them. Due to the fact that many NEL tools are not trained to correctly recognize creative works and due to the popularity prior settings used by various 4 https://tvtropes.org/ 5 https://en.wikinews.org/ algorithms, this partition is generally considered difficult for current NEL tools.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 RegionalTV (TV Tropes) set contains texts about European TV Shows.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 EuroFilm (TV Tropes) is focused on classic and modern European films. We have typically included five to fifteen movies for the selected countries (France, Germany, Austria, Switzerland, U.K., Netherlands, Italy, Denmark, Sweden).",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 The WebMedia (TV Tropes) set was built around YouTube influencers. Due to the dynamic nature of the YouTube platform, the channels of some of the annotated entities may not exist in the future. Therefore, the content from this partition should be considered time-sensitive and relatively difficult. For the current version of the corpus the various types of YouTube subcultures 6 were not annotated, but we consider including such types in future versions.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 News (Wiki News) collects general interest News on a variety of topics. As expected, the level of difficulty for this partition is medium, since most tools were trained for such content.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Politics (Wiki News) encompasses general politics News related to elections, political events (e.g., Syrian Conflict, Arab Spring) and war-related News. In some cases there might be overlaps between this partition and the News partition, but this is simply due to the fact that they include the same types of entities.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Business (Wikipedia) includes documents on corporations from the tech, medical and media domains. The difficulty level is medium.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Climate (Wikipedia) contains coverage on Climate Change, sustainability and related entities (e.g., Greenpeace, Al Gore, Greta Thunberg).",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Entertainment News (Wikipedia) is a partition related to celebrity News during early 2020.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "\u2022 Core set (General) (Wikipedia) partition contains the core set of the first 9 partitions, including short texts from domains like general news, politics, franchises, TV or movies. The level of difficulty is generally medium.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "Some basic statistics about this corpus can be found in Table 3 . Most of the documents have one to three sentences and can be considered equivalent to DBpedia abstracts, even though they were collected from various sources. Due to the nature of the collected information (e.g., franchises, TV shows, books) the early partitions often draw upon abstracts. Later partitions, in contrast, are randomly selected from the actual content of the articles. This assures that the collection is heterogeneous and that it can later be used for testing multiple use cases (e.g., media-related entities, news media, policies, etc).",
	"cite_spans": [],
	"ref_spans": [
	{
	"start": 56,
	"end": 63,
	"text": "Table 3",
	"ref_id": "TABREF5"
	}
	],
	"eq_spans": [],
	"section": "Partitions and Statistics",
	"sec_num": "3.2"
	},
	{
	"text": "This section describes the tools used during the evaluation, its design and a discussion around results.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation",
	"sec_num": "4"
	},
	{
	"text": "The following tools have been used during our general-purpose evaluation:",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "\u2022 DBpedia Spotlight (Daiber et al., 2013 ) is a statistical NEL engine that was originally built as a demo for showcasing DBpedia's capabilities and has been ported to multiple languages. The statistical models from Spotlight are really good for larger Knowledge Extraction or WSD challenges where all words need to be linked to their respective KG entities, but they are not necessarily fine-tuned for typed NEL tasks.",
	"cite_spans": [
	{
	"start": 20,
	"end": 40,
	"text": "(Daiber et al., 2013",
	"ref_id": "BIBREF6"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "\u2022 AIDA (Hoffart et al., 2011) uses graph-based disambiguation algorithms and is considered one of the best NEL engines focused around Wikipedia linking.",
	"cite_spans": [
	{
	"start": 7,
	"end": 29,
	"text": "(Hoffart et al., 2011)",
	"ref_id": "BIBREF15"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "\u2022 Recognyze (Weichselbraun et al., 2019b) is a multi-KG (e.g., DBpedia, Wikidata, Wikipedia) graph-based disambiguation engine focused on the issue of name variance.",
	"cite_spans": [
	{
	"start": 12,
	"end": 41,
	"text": "(Weichselbraun et al., 2019b)",
	"ref_id": "BIBREF35"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "We have created two different sets of evaluations. The first one (see Table 4 ) contains the results for the core set partition of the corpus. This evaluation also lists the results for different annotation styles (lenses) as they were presented in this paper. The second evaluation (see Table 5 ) compares the results of several tools on the entire corpora.",
	"cite_spans": [],
	"ref_spans": [
	{
	"start": 70,
	"end": 77,
	"text": "Table 4",
	"ref_id": "TABREF7"
	},
	{
	"start": 288,
	"end": 295,
	"text": "Table 5",
	"ref_id": null
	}
	],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "As it can be seen in Table 4 , the multiple annotation styles had an impact on almost all of the evaluated tools. This suggests that most of the tools do seem to perform better when considering these annotation styles in succession, with Spotlight and Recognyze gaining up to 4%. It is interesting to note that the rules seem to improve the recall of DBpedia, Spotlight and Recognyze in all cases, whereas precision is not impacted by OMAX styles for Spotlight. There might be a need for multiple evaluations in a future publication to establish the full impact of these guidelines, but since such annotation styles can automatically be generated from any dataset following the outlined rules, they are definitely worth investigating.",
	"cite_spans": [],
	"ref_spans": [
	{
	"start": 21,
	"end": 28,
	"text": "Table 4",
	"ref_id": "TABREF7"
	}
	],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "As expected the results on the core set (Table 4 ) and the entire corpus (Table 5) are similar. They are between 2% and 3% lower for each tool, which due to the higher number of entities should be considered a good results. This also indicates that the content of the core set was indeed carefully chosen Table 5 : Results on the entire corpora -with overlaps -OMAX lens (m -micro; M -macro; p -precision; r -recall; F1 -F1).",
	"cite_spans": [],
	"ref_spans": [
	{
	"start": 40,
	"end": 48,
	"text": "(Table 4",
	"ref_id": "TABREF7"
	},
	{
	"start": 73,
	"end": 82,
	"text": "(Table 5)",
	"ref_id": null
	},
	{
	"start": 305,
	"end": 312,
	"text": "Table 5",
	"ref_id": null
	}
	],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "to reflect the content of the whole dataset up to this point in time.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Evaluation Design and Results",
	"sec_num": "4.1"
	},
	{
	"text": "NEL performance on the In Media Res corpus is considerably lower than the results obtained on traditional data sets. This was expected due to the large amount of errors introduced by adding creative works to the corpora. Also as expected, the tools were not able to distinguish well between a character and the franchise that bears its name or offer good results on the YouTube influencer partition of the corpus. The influencer partition is especially difficult due to the fact that some of the works mentioned there (e.g., YouTube channels that were shut down or early gigs for famous influencers) are NIL (i.e. entities that have not been included in Wikipedia or related KGs such as DBpedia and Wikidata).",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Discussion",
	"sec_num": "4.2"
	},
	{
	"text": "While some media franchises are well-covered by Wikipedia (e.g., Harry Potter) and related KGs, others are not. In such cases a good approach towards improving coverage and results might be leveraging Linked Data extracted from dedicated wikis such as Memory Alpha (covering the Star Trek franchise), Wookieepedia (covering Star Wars) or Marvel Database (covering both Marvel Comics and the Marvel Cinematic Universe). Most of the fandoms organize around such wikis and also many of them are published through Fandom 7 and similar wiki engines. Some of the information from these wikis is also collected in Linked Data form through DBkwik (Hertling and Paulheim, 2018 ).",
	"cite_spans": [
	{
	"start": 639,
	"end": 667,
	"text": "(Hertling and Paulheim, 2018",
	"ref_id": "BIBREF14"
	}
	],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Discussion",
	"sec_num": "4.2"
	},
	{
	"text": "7 https://www.fandom.com/",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Discussion",
	"sec_num": "4.2"
	},
	{
	"text": "The road towards designing culturally-aware AIs has started with the expansion of fields like digital humanities during the last decade. While some steps towards this goal were made, the selected topics still depend on funding and on the researcher's own goals, as this is a relatively new field. Even so, we find it surprising that there was a lack of guidance in the NEL community related to name variation and nested entities for creative works. The corpus and concepts introduced in this paper are a first attempt to address this issue.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Conclusion and Future Work",
	"sec_num": "5"
	},
	{
	"text": "The results of the various annotators on the In Media Res corpus are understandably lower than on corpora from traditional domains such as news articles and social media. While applying different annotation lenses improves the results for some annotators, it is clear that there is a need for more progress in this area.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Conclusion and Future Work",
	"sec_num": "5"
	},
	{
	"text": "We plan to continue maintaining the corpus and provide updates, new partitions, or lenses. We hope that these efforts will contribute to increased accuracy in the annotation of creative work entities and, therefore, aid annotation systems in taking a further step towards culturally-aware AIs.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Conclusion and Future Work",
	"sec_num": "5"
	},
	{
	"text": "dbr represents the namespace abbreviation of DBpedia resources, wd for Wikidata resources",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "",
	"sec_num": null
	},
	{
	"text": "The corpus is available online at: https://github. com/modultechnology/in_media_res.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "",
	"sec_num": null
	},
	{
	"text": "The In Media Res CorpusSince most corpora used in ground truth annotation are focused on news media or tweets that describe current events, there is no wide agreement on annotation styles for entities of type Creative Work.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "",
	"sec_num": null
	},
	{
	"text": "represented by http://dbpedia.org/ontology/ Work (abbreviated as dbo:Work) in DBpedia or https://schema.org/CreativeWork in the schema.org vocabulary",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "",
	"sec_num": null
	},
	{
	"text": "https://tvtropes.org/pmwiki/pmwiki.php/ UsefulNotes/Subcultures",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "",
	"sec_num": null
	}
	],
	"back_matter": [
	{
	"text": "This research has been partially funded through the following projects: the ReTV project (www.retvproject.eu) funded by the European Union's Horizon 2020 Research and Innovation Programme (No. 780656), and MedMon (www.fhgr.ch/medmon) funded by the Swiss Innovation Agency Innosuisse.",
	"cite_spans": [],
	"ref_spans": [],
	"eq_spans": [],
	"section": "Acknowledgments",
	"sec_num": null
	}
	],
	"bib_entries": {
	"BIBREF1": {
	"ref_id": "b1",
	"title": "Scientific Lenses to Support Multiple Views over Linked Chemistry Data",
	"authors": [
	{
	"first": "Lee",
	"middle": [],
	"last": "Groth",
	"suffix": ""
	},
	{
	"first": "Karen",
	"middle": [],
	"last": "Harland",
	"suffix": ""
	},
	{
	"first": "Antonis",
	"middle": [],
	"last": "Karapetyan",
	"suffix": ""
	},
	{
	"first": "John",
	"middle": [
	"P"
	],
	"last": "Loizou",
	"suffix": ""
	},
	{
	"first": "Steve",
	"middle": [],
	"last": "Overington",
	"suffix": ""
	},
	{
	"first": "Jon",
	"middle": [],
	"last": "Pettifer",
	"suffix": ""
	},
	{
	"first": "Robert",
	"middle": [],
	"last": "Steele",
	"suffix": ""
	},
	{
	"first": "Valery",
	"middle": [],
	"last": "Stevens",
	"suffix": ""
	},
	{
	"first": "Andra",
	"middle": [],
	"last": "Tkachenko",
	"suffix": ""
	},
	{
	"first": "Antony",
	"middle": [
	"J"
	],
	"last": "Waagmeester",
	"suffix": ""
	},
	{
	"first": "Egon",
	"middle": [
	"L"
	],
	"last": "Williams",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Willighagen",
	"suffix": ""
	}
	],
	"year": 2014,
	"venue": "The Semantic Web -ISWC 2014 -13th International Semantic Web Conference",
	"volume": "8796",
	"issue": "",
	"pages": "98--113",
	"other_ids": {
	"DOI": [
	"10.1007/978-3-319-11964-9_7"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Groth, Lee Harland, Karen Karapetyan, Antonis Loizou, John P. Overington, Steve Pettifer, Jon Steele, Robert Stevens, Valery Tkachenko, Andra Waagmeester, Antony J. Williams, and Egon L. Wil- lighagen. 2014. Scientific Lenses to Support Mul- tiple Views over Linked Chemistry Data. In The Semantic Web -ISWC 2014 -13th International Se- mantic Web Conference, Riva del Garda, Italy, Octo- ber 19-23, 2014. Proceedings, Part I, volume 8796 of Lecture Notes in Computer Science, pages 98- 113. Springer.",
	"links": null
	},
	"BIBREF2": {
	"ref_id": "b2",
	"title": "Data-Dependent Coresets for Compressing Neural Networks with Applications to Generalization Bounds",
	"authors": [
	{
	"first": "Cenk",
	"middle": [],
	"last": "Baykal",
	"suffix": ""
	},
	{
	"first": "Lucas",
	"middle": [],
	"last": "Liebenwein",
	"suffix": ""
	},
	{
	"first": "Igor",
	"middle": [],
	"last": "Gilitschenski",
	"suffix": ""
	},
	{
	"first": "Dan",
	"middle": [],
	"last": "Feldman",
	"suffix": ""
	},
	{
	"first": "Daniela",
	"middle": [],
	"last": "Rus",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Cenk Baykal, Lucas Liebenwein, Igor Gilitschenski, Dan Feldman, and Daniela Rus. 2018. Data- Dependent Coresets for Compressing Neural Net- works with Applications to Generalization Bounds. CoRR, abs/1804.05345.",
	"links": null
	},
	"BIBREF3": {
	"ref_id": "b3",
	"title": "StoryLens: A Multiple Views Corpus for Location and Event Detection",
	"authors": [
	{
	"first": "M",
	"middle": [
	"P"
	],
	"last": "Adrian",
	"suffix": ""
	},
	{
	"first": "Lyndon",
	"middle": [
	"J B"
	],
	"last": "Bra\u015foveanu",
	"suffix": ""
	},
	{
	"first": "Albert",
	"middle": [],
	"last": "Nixon",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Weichselbraun",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the 8th International Conference on Web Intelligence, Mining and Semantics (WIMS 2018)",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {
	"DOI": [
	"10.1145/3227609.3227674"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Adrian M. P. Bra\u015foveanu, Lyndon J.B. Nixon, and Al- bert Weichselbraun. 2018a. StoryLens: A Multi- ple Views Corpus for Location and Event Detection. In Proceedings of the 8th International Conference on Web Intelligence, Mining and Semantics (WIMS 2018), Novi Sad, Serbia. ACM.",
	"links": null
	},
	"BIBREF4": {
	"ref_id": "b4",
	"title": "Framing Named Entity Linking Error Types",
	"authors": [
	{
	"first": "M",
	"middle": [
	"P"
	],
	"last": "Adrian",
	"suffix": ""
	},
	{
	"first": "Giuseppe",
	"middle": [],
	"last": "Bra\u015foveanu",
	"suffix": ""
	},
	{
	"first": "Philipp",
	"middle": [],
	"last": "Rizzo",
	"suffix": ""
	},
	{
	"first": "Albert",
	"middle": [],
	"last": "Kuntschick",
	"suffix": ""
	},
	{
	"first": "Lyndon",
	"middle": [
	"J B"
	],
	"last": "Weichselbraun",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Nixon",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)",
	"volume": "",
	"issue": "",
	"pages": "266--271",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Adrian M. P. Bra\u015foveanu, Giuseppe Rizzo, Philipp Kuntschick, Albert Weichselbraun, and Lyndon J.B. Nixon. 2018b. Framing Named Entity Linking Error Types. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), pages 266-271, Paris, France. Euro- pean Language Resources Association (ELRA).",
	"links": null
	},
	"BIBREF5": {
	"ref_id": "b5",
	"title": "A 'historical case' of Ontology-Based Data Access",
	"authors": [
	{
	"first": "Diego",
	"middle": [],
	"last": "Calvanese",
	"suffix": ""
	},
	{
	"first": "Alessandro",
	"middle": [],
	"last": "Mosca",
	"suffix": ""
	},
	{
	"first": "Jos\u00e9",
	"middle": [],
	"last": "Remesal",
	"suffix": ""
	},
	{
	"first": "Mart\u00edn",
	"middle": [],
	"last": "Rezk",
	"suffix": ""
	},
	{
	"first": "Guillem",
	"middle": [],
	"last": "Rull",
	"suffix": ""
	}
	],
	"year": 2015,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "291--298",
	"other_ids": {
	"DOI": [
	"10.1109/DigitalHeritage.2015.7419510"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Diego Calvanese, Alessandro Mosca, Jos\u00e9 Remesal, Mart\u00edn Rezk, and Guillem Rull. 2015. A 'histori- cal case' of Ontology-Based Data Access. In 2015 Digital Heritage, Granada, Spain, September 28 - October 2, 2015, pages 291-298. IEEE.",
	"links": null
	},
	"BIBREF6": {
	"ref_id": "b6",
	"title": "Improving Efficiency and Accuracy in Multilingual Entity Extraction",
	"authors": [
	{
	"first": "Joachim",
	"middle": [],
	"last": "Daiber",
	"suffix": ""
	},
	{
	"first": "Max",
	"middle": [],
	"last": "Jakob",
	"suffix": ""
	},
	{
	"first": "Chris",
	"middle": [],
	"last": "Hokamp",
	"suffix": ""
	},
	{
	"first": "Pablo",
	"middle": [
	"N"
	],
	"last": "Mendes",
	"suffix": ""
	}
	],
	"year": 2013,
	"venue": "I-SEMANTICS 2013 -9th International Conference on Semantic Systems, ISEM '13",
	"volume": "",
	"issue": "",
	"pages": "121--124",
	"other_ids": {
	"DOI": [
	"10.1145/2506182.2506198"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Joachim Daiber, Max Jakob, Chris Hokamp, and Pablo N. Mendes. 2013. Improving Efficiency and Accuracy in Multilingual Entity Extraction. In I- SEMANTICS 2013 -9th International Conference on Semantic Systems, ISEM '13, Graz, Austria, September 4-6, 2013, pages 121-124. ACM.",
	"links": null
	},
	"BIBREF7": {
	"ref_id": "b7",
	"title": "Evaluating Social Network Extraction for Classic and Modern Fiction Literature",
	"authors": [
	{
	"first": "Niels",
	"middle": [],
	"last": "Dekker",
	"suffix": ""
	},
	{
	"first": "Tobias",
	"middle": [],
	"last": "Kuhn",
	"suffix": ""
	},
	{
	"first": "Marieke",
	"middle": [],
	"last": "Van Erp",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "PeerJ Prepr",
	"volume": "6",
	"issue": "",
	"pages": "",
	"other_ids": {
	"DOI": [
	"10.7287/peerj.preprints.27263v1"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Niels Dekker, Tobias Kuhn, and Marieke van Erp. 2018. Evaluating Social Network Extraction for Classic and Modern Fiction Literature. PeerJ Prepr., 6:e27263.",
	"links": null
	},
	"BIBREF8": {
	"ref_id": "b8",
	"title": "Jrc-names: Multilingual entity name variants and titles as linked data",
	"authors": [
	{
	"first": "Maud",
	"middle": [],
	"last": "Ehrmann",
	"suffix": ""
	},
	{
	"first": "Guillaume",
	"middle": [],
	"last": "Jacquet",
	"suffix": ""
	},
	{
	"first": "Ralf",
	"middle": [],
	"last": "Steinberger",
	"suffix": ""
	}
	],
	"year": 2017,
	"venue": "Semantic Web",
	"volume": "8",
	"issue": "",
	"pages": "283--295",
	"other_ids": {
	"DOI": [
	"10.3233/SW-160228"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Maud Ehrmann, Guillaume Jacquet, and Ralf Stein- berger. 2017. Jrc-names: Multilingual entity name variants and titles as linked data. Semantic Web, 8(2):283-295.",
	"links": null
	},
	"BIBREF9": {
	"ref_id": "b9",
	"title": "Language Resources for Historical Newspapers: The Impresso Collection",
	"authors": [
	{
	"first": "Maud",
	"middle": [],
	"last": "Ehrmann",
	"suffix": ""
	},
	{
	"first": "Matteo",
	"middle": [],
	"last": "Romanello",
	"suffix": ""
	},
	{
	"first": "Simon",
	"middle": [],
	"last": "Clematide",
	"suffix": ""
	},
	{
	"first": "Phillip",
	"middle": [],
	"last": "Str\u00f6bel",
	"suffix": ""
	},
	{
	"first": "Rapha\u00ebl",
	"middle": [],
	"last": "Barman",
	"suffix": ""
	}
	],
	"year": 2020,
	"venue": "Proceedings of The 12th Language Resources and Evaluation Conference",
	"volume": "2020",
	"issue": "",
	"pages": "958--968",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Maud Ehrmann, Matteo Romanello, Simon Clematide, Phillip Str\u00f6bel, and Rapha\u00ebl Barman. 2020. Lan- guage Resources for Historical Newspapers: The Im- presso Collection. In Proceedings of The 12th Lan- guage Resources and Evaluation Conference, LREC 2020, Marseille, France, May 11-16, 2020, pages 958-968. European Language Resources Associa- tion.",
	"links": null
	},
	"BIBREF10": {
	"ref_id": "b10",
	"title": "Coresets for Vector Summarization with Applications to Network Graphs",
	"authors": [
	{
	"first": "Dan",
	"middle": [],
	"last": "Feldman",
	"suffix": ""
	},
	{
	"first": "Sedat",
	"middle": [],
	"last": "Ozer",
	"suffix": ""
	},
	{
	"first": "Daniela",
	"middle": [],
	"last": "Rus",
	"suffix": ""
	}
	],
	"year": 2017,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Dan Feldman, Sedat Ozer, and Daniela Rus. 2017. Coresets for Vector Summarization with Applica- tions to Network Graphs. CoRR, abs/1706.05554.",
	"links": null
	},
	"BIBREF11": {
	"ref_id": "b11",
	"title": "Dimensionality Reduction of Massive Sparse Datasets Using Coresets",
	"authors": [
	{
	"first": "Dan",
	"middle": [],
	"last": "Feldman",
	"suffix": ""
	},
	{
	"first": "Mikhail",
	"middle": [],
	"last": "Volkov",
	"suffix": ""
	},
	{
	"first": "Daniela",
	"middle": [],
	"last": "Rus",
	"suffix": ""
	}
	],
	"year": 2016,
	"venue": "Advances in Neural Information Processing Systems 29: Annual Conference on Neural Information Processing Systems 2016",
	"volume": "",
	"issue": "",
	"pages": "2766--2774",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Dan Feldman, Mikhail Volkov, and Daniela Rus. 2016. Dimensionality Reduction of Massive Sparse Datasets Using Coresets. In Advances in Neu- ral Information Processing Systems 29: Annual Conference on Neural Information Processing Sys- tems 2016, December 5-10, 2016, Barcelona, Spain, pages 2766-2774.",
	"links": null
	},
	"BIBREF12": {
	"ref_id": "b12",
	"title": "Data Motifs: A Lens Towards Fully Understanding Big Data and AI Workloads",
	"authors": [
	{
	"first": "Wanling",
	"middle": [],
	"last": "Gao",
	"suffix": ""
	},
	{
	"first": "Jianfeng",
	"middle": [],
	"last": "Zhan",
	"suffix": ""
	},
	{
	"first": "Lei",
	"middle": [],
	"last": "Wang",
	"suffix": ""
	},
	{
	"first": "Chunjie",
	"middle": [],
	"last": "Luo",
	"suffix": ""
	},
	{
	"first": "Daoyi",
	"middle": [],
	"last": "Zheng",
	"suffix": ""
	},
	{
	"first": "Fei",
	"middle": [],
	"last": "Tang",
	"suffix": ""
	},
	{
	"first": "Biwei",
	"middle": [],
	"last": "Xie",
	"suffix": ""
	},
	{
	"first": "Chen",
	"middle": [],
	"last": "Zheng",
	"suffix": ""
	},
	{
	"first": "Xu",
	"middle": [],
	"last": "Wen",
	"suffix": ""
	},
	{
	"first": "Xiwen",
	"middle": [],
	"last": "He",
	"suffix": ""
	},
	{
	"first": "Hainan",
	"middle": [],
	"last": "Ye",
	"suffix": ""
	},
	{
	"first": "Rui",
	"middle": [],
	"last": "Ren",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the 27th International Conference on Parallel Architectures and Compilation Techniques, PACT 2018",
	"volume": "2",
	"issue": "",
	"pages": "1--2",
	"other_ids": {
	"DOI": [
	"10.1145/3243176.3243190"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Wanling Gao, Jianfeng Zhan, Lei Wang, Chunjie Luo, Daoyi Zheng, Fei Tang, Biwei Xie, Chen Zheng, Xu Wen, Xiwen He, Hainan Ye, and Rui Ren. 2018. Data Motifs: A Lens Towards Fully Understanding Big Data and AI Workloads. In Proceedings of the 27th International Conference on Parallel Architec- tures and Compilation Techniques, PACT 2018, Li- massol, Cyprus, November 01-04, 2018, pages 2:1- 2:14. ACM.",
	"links": null
	},
	"BIBREF13": {
	"ref_id": "b13",
	"title": "Cheap and Easy Entity Evaluation",
	"authors": [
	{
	"first": "Ben",
	"middle": [],
	"last": "Hachey",
	"suffix": ""
	},
	{
	"first": "Joel",
	"middle": [],
	"last": "Nothman",
	"suffix": ""
	},
	{
	"first": "Will",
	"middle": [],
	"last": "Radford",
	"suffix": ""
	}
	],
	"year": 2014,
	"venue": "Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics",
	"volume": "2",
	"issue": "",
	"pages": "464--469",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Ben Hachey, Joel Nothman, and Will Radford. 2014. Cheap and Easy Entity Evaluation. In Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, ACL 2014, June 22-27, 2014, Baltimore, MD, USA, Volume 2: Short Papers, pages 464-469. The Association for Computer Lin- guistics.",
	"links": null
	},
	"BIBREF14": {
	"ref_id": "b14",
	"title": "DBkWik: A Consolidated Knowledge Graph from Thousands of Wikis",
	"authors": [
	{
	"first": "Sven",
	"middle": [],
	"last": "Hertling",
	"suffix": ""
	},
	{
	"first": "Heiko",
	"middle": [],
	"last": "Paulheim",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "2018 IEEE International Conference on Big Knowledge",
	"volume": "",
	"issue": "",
	"pages": "17--24",
	"other_ids": {
	"DOI": [
	"10.1109/ICBK.2018.00011"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Sven Hertling and Heiko Paulheim. 2018. DBkWik: A Consolidated Knowledge Graph from Thousands of Wikis. In 2018 IEEE International Conference on Big Knowledge, ICBK 2018, Singapore, November 17-18, 2018, pages 17-24. IEEE Computer Society.",
	"links": null
	},
	"BIBREF15": {
	"ref_id": "b15",
	"title": "Robust Disambiguation of Named Entities in Text",
	"authors": [
	{
	"first": "Johannes",
	"middle": [],
	"last": "Hoffart",
	"suffix": ""
	},
	{
	"first": "Mohamed",
	"middle": [
	"Amir"
	],
	"last": "Yosef",
	"suffix": ""
	},
	{
	"first": "Ilaria",
	"middle": [],
	"last": "Bordino",
	"suffix": ""
	},
	{
	"first": "Hagen",
	"middle": [],
	"last": "F\u00fcrstenau",
	"suffix": ""
	},
	{
	"first": "Manfred",
	"middle": [],
	"last": "Pinkal",
	"suffix": ""
	},
	{
	"first": "Marc",
	"middle": [],
	"last": "Spaniol",
	"suffix": ""
	},
	{
	"first": "Bilyana",
	"middle": [],
	"last": "Taneva",
	"suffix": ""
	},
	{
	"first": "Stefan",
	"middle": [],
	"last": "Thater",
	"suffix": ""
	},
	{
	"first": "Gerhard",
	"middle": [],
	"last": "Weikum",
	"suffix": ""
	}
	],
	"year": 2011,
	"venue": "Proceedings of the 2011 Conference on Empirical Methods in Natural Language Processing",
	"volume": "",
	"issue": "",
	"pages": "782--792",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Johannes Hoffart, Mohamed Amir Yosef, Ilaria Bor- dino, Hagen F\u00fcrstenau, Manfred Pinkal, Marc Span- iol, Bilyana Taneva, Stefan Thater, and Gerhard Weikum. 2011. Robust Disambiguation of Named Entities in Text. In Proceedings of the 2011 Con- ference on Empirical Methods in Natural Language Processing, EMNLP 2011, 27-31 July 2011, John McIntyre Conference Centre, Edinburgh, UK, A meeting of SIGDAT, a Special Interest Group of the ACL, pages 782-792.",
	"links": null
	},
	"BIBREF16": {
	"ref_id": "b16",
	"title": "Europarl-ST: A Multilingual Corpus For Speech Translation Of Parliamentary Debates",
	"authors": [
	{
	"first": "Javier",
	"middle": [],
	"last": "Iranzo-S\u00e1nchez",
	"suffix": ""
	},
	{
	"first": "Joan",
	"middle": [
	"Albert"
	],
	"last": "Silvestre-Cerd\u00e0",
	"suffix": ""
	},
	{
	"first": "Javier",
	"middle": [],
	"last": "Jorge",
	"suffix": ""
	},
	{
	"first": "Nahuel",
	"middle": [],
	"last": "Rosell\u00f3",
	"suffix": ""
	},
	{
	"first": "Adri\u00e0",
	"middle": [],
	"last": "Gim\u00e9nez",
	"suffix": ""
	},
	{
	"first": "Albert",
	"middle": [],
	"last": "Sanch\u00eds",
	"suffix": ""
	},
	{
	"first": "Jorge",
	"middle": [],
	"last": "Civera",
	"suffix": ""
	},
	{
	"first": "Alfons",
	"middle": [],
	"last": "Juan",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Javier Iranzo-S\u00e1nchez, Joan Albert Silvestre-Cerd\u00e0, Javier Jorge, Nahuel Rosell\u00f3, Adri\u00e0 Gim\u00e9nez, Al- bert Sanch\u00eds, Jorge Civera, and Alfons Juan. 2019. Europarl-ST: A Multilingual Corpus For Speech Translation Of Parliamentary Debates. CoRR, abs/1911.03167.",
	"links": null
	},
	"BIBREF18": {
	"ref_id": "b18",
	"title": "Proceedings of the 2017 Text Analysis Conference",
	"authors": [],
	"year": 2017,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Languages Entity Discovery and Linking. In Pro- ceedings of the 2017 Text Analysis Conference, TAC 2017, Gaithersburg, Maryland, USA, November 13- 14, 2017, page 4. NIST.",
	"links": null
	},
	"BIBREF19": {
	"ref_id": "b19",
	"title": "A Neural Layered Model for Nested Named Entity Recognition",
	"authors": [
	{
	"first": "Meizhi",
	"middle": [],
	"last": "Ju",
	"suffix": ""
	},
	{
	"first": "Makoto",
	"middle": [],
	"last": "Miwa",
	"suffix": ""
	},
	{
	"first": "Sophia",
	"middle": [],
	"last": "Ananiadou",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies",
	"volume": "1",
	"issue": "",
	"pages": "1446--1459",
	"other_ids": {
	"DOI": [
	"10.18653/v1/n18-1131"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Meizhi Ju, Makoto Miwa, and Sophia Ananiadou. 2018. A Neural Layered Model for Nested Named Entity Recognition. In Proceedings of the 2018 Con- ference of the North American Chapter of the Asso- ciation for Computational Linguistics: Human Lan- guage Technologies, NAACL-HLT 2018, New Or- leans, Louisiana, USA, June 1-6, 2018, Volume 1 (Long Papers), pages 1446-1459. Association for Computational Linguistics.",
	"links": null
	},
	"BIBREF20": {
	"ref_id": "b20",
	"title": "Managing Data through the Lens of an Ontology",
	"authors": [
	{
	"first": "Maurizio",
	"middle": [],
	"last": "Lenzerini",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "",
	"volume": "39",
	"issue": "",
	"pages": "65--74",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Maurizio Lenzerini. 2018. Managing Data through the Lens of an Ontology. AI Magazine, 39(2):65-74.",
	"links": null
	},
	"BIBREF21": {
	"ref_id": "b21",
	"title": "Enriching Media Fragments with Named Entities for Video Classification",
	"authors": [
	{
	"first": "Yunjia",
	"middle": [],
	"last": "Li",
	"suffix": ""
	},
	{
	"first": "Giuseppe",
	"middle": [],
	"last": "Rizzo",
	"suffix": ""
	},
	{
	"first": "Jos\u00e9",
	"middle": [],
	"last": "Luis Redondo",
	"suffix": ""
	},
	{
	"first": "Rapha\u00ebl",
	"middle": [],
	"last": "Garc\u00eda",
	"suffix": ""
	},
	{
	"first": "Mike",
	"middle": [],
	"last": "Troncy",
	"suffix": ""
	},
	{
	"first": "Gary",
	"middle": [],
	"last": "Wald",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Wills",
	"suffix": ""
	}
	],
	"year": 2013,
	"venue": "22nd International World Wide Web Conference, WWW '13",
	"volume": "",
	"issue": "",
	"pages": "469--476",
	"other_ids": {
	"DOI": [
	"10.1145/2487788.2487970"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Yunjia Li, Giuseppe Rizzo, Jos\u00e9 Luis Redondo Gar- c\u00eda, Rapha\u00ebl Troncy, Mike Wald, and Gary Wills. 2013. Enriching Media Fragments with Named En- tities for Video Classification. In 22nd International World Wide Web Conference, WWW '13, Rio de Janeiro, Brazil, May 13-17, 2013, Companion Vol- ume, pages 469-476. International World Wide Web Conferences Steering Committee / ACM.",
	"links": null
	},
	"BIBREF22": {
	"ref_id": "b22",
	"title": "GATE: Bridging the Gap between Terminology and Linguistics",
	"authors": [
	{
	"first": "Diana",
	"middle": [],
	"last": "Maynard",
	"suffix": ""
	}
	],
	"year": 2009,
	"venue": "Proceedings of the 8th International Conference on Terminology and Artificial Intelligence",
	"volume": "578",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Diana Maynard. 2009. GATE: Bridging the Gap be- tween Terminology and Linguistics. In Proceedings of the 8th International Conference on Terminology and Artificial Intelligence, Toulouse, France, Novem- ber 18-20, 2009, volume 578 of CEUR Workshop Proceedings. CEUR-WS.org.",
	"links": null
	},
	"BIBREF23": {
	"ref_id": "b23",
	"title": "MEANTIME, the NewsReader Multilingual Event and Time Corpus",
	"authors": [
	{
	"first": "Anne-Lyse",
	"middle": [],
	"last": "Minard",
	"suffix": ""
	},
	{
	"first": "Manuela",
	"middle": [],
	"last": "Speranza",
	"suffix": ""
	},
	{
	"first": "Ruben",
	"middle": [],
	"last": "Urizar",
	"suffix": ""
	},
	{
	"first": "Bego\u00f1a",
	"middle": [],
	"last": "Altuna",
	"suffix": ""
	},
	{
	"first": "Anneleen",
	"middle": [],
	"last": "Marieke Van Erp",
	"suffix": ""
	},
	{
	"first": "Chantal",
	"middle": [],
	"last": "Schoen",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Van Son",
	"suffix": ""
	}
	],
	"year": 2016,
	"venue": "Proceedings of the Tenth International Conference on Language Resources and Evaluation LREC 2016",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Anne-Lyse Minard, Manuela Speranza, Ruben Urizar, Bego\u00f1a Altuna, Marieke van Erp, Anneleen Schoen, and Chantal van Son. 2016. MEANTIME, the NewsReader Multilingual Event and Time Corpus. In Proceedings of the Tenth International Confer- ence on Language Resources and Evaluation LREC 2016, Portoro\u017e, Slovenia, May 23-28, 2016. Euro- pean Language Resources Association (ELRA).",
	"links": null
	},
	"BIBREF24": {
	"ref_id": "b24",
	"title": "Multimodal Video Annotation for Retrieval and Discovery of Newsworthy Video in a News Verification Scenario",
	"authors": [
	{
	"first": "J",
	"middle": [
	"B"
	],
	"last": "Lyndon",
	"suffix": ""
	},
	{
	"first": "Evlampios",
	"middle": [
	"E"
	],
	"last": "Nixon",
	"suffix": ""
	},
	{
	"first": "Foteini",
	"middle": [],
	"last": "Apostolidis",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Markatopoulou",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "MultiMedia Modeling -25th International Conference, MMM 2019",
	"volume": "11295",
	"issue": "",
	"pages": "143--155",
	"other_ids": {
	"DOI": [
	"10.1007/978-3-030-05710-7_12"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Lyndon J. B. Nixon, Evlampios E. Apostolidis, Foteini Markatopoulou, Ioannis Patras, and Vasileios Mezaris. 2019. Multimodal Video Annotation for Retrieval and Discovery of Newsworthy Video in a News Verification Scenario. In MultiMedia Mod- eling -25th International Conference, MMM 2019, Thessaloniki, Greece, January 8-11, 2019, Proceed- ings, Part I, volume 11295 of Lecture Notes in Com- puter Science, pages 143-155. Springer.",
	"links": null
	},
	"BIBREF25": {
	"ref_id": "b25",
	"title": "Survey of Semantic Media Annotation Tools for the Web: Towards New Media Applications with Linked Media",
	"authors": [
	{
	"first": "J",
	"middle": [
	"B"
	],
	"last": "Lyndon",
	"suffix": ""
	},
	{
	"first": "Rapha\u00ebl",
	"middle": [],
	"last": "Nixon",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Troncy",
	"suffix": ""
	}
	],
	"year": 2014,
	"venue": "",
	"volume": "8798",
	"issue": "",
	"pages": "100--114",
	"other_ids": {
	"DOI": [
	"10.1007/978-3-319-11955-7_9"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Lyndon J. B. Nixon and Rapha\u00ebl Troncy. 2014. Sur- vey of Semantic Media Annotation Tools for the Web: Towards New Media Applications with Linked Media. In The Semantic Web: ESWC 2014 Satel- lite Events -ESWC 2014 Satellite Events, Anissaras, Crete, Greece, May 25-29, 2014, Revised Selected Papers, volume 8798 of Lecture Notes in Computer Science, pages 100-114. Springer.",
	"links": null
	},
	"BIBREF26": {
	"ref_id": "b26",
	"title": "On the importance of drill-down analysis for assessing gold standards and named entity linking performance",
	"authors": [
	{
	"first": "Fabian",
	"middle": [],
	"last": "Odoni",
	"suffix": ""
	},
	{
	"first": "Philipp",
	"middle": [],
	"last": "Kuntschik",
	"suffix": ""
	},
	{
	"first": "Adrian",
	"middle": [
	"M P"
	],
	"last": "Bra\u015foveanu",
	"suffix": ""
	},
	{
	"first": "Albert",
	"middle": [],
	"last": "Weichselbraun",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the 14th International Conference on Semantic Systems, SEMAN-TICS 2018",
	"volume": "137",
	"issue": "",
	"pages": "33--42",
	"other_ids": {
	"DOI": [
	"10.1016/j.procs.2018.09.004"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Fabian Odoni, Philipp Kuntschik, Adrian M. P. Bra\u015foveanu, and Albert Weichselbraun. 2018. On the importance of drill-down analysis for assess- ing gold standards and named entity linking per- formance. In Proceedings of the 14th Interna- tional Conference on Semantic Systems, SEMAN- TICS 2018, Vienna, Austria, September 10-13, 2018, volume 137 of Procedia Computer Science, pages 33-42. Elsevier.",
	"links": null
	},
	"BIBREF27": {
	"ref_id": "b27",
	"title": "Lenses for Web Data",
	"authors": [
	{
	"first": "Raghu",
	"middle": [],
	"last": "Rajkumar",
	"suffix": ""
	},
	{
	"first": "Nate",
	"middle": [],
	"last": "Foster",
	"suffix": ""
	},
	{
	"first": "Sam",
	"middle": [],
	"last": "Lindley",
	"suffix": ""
	},
	{
	"first": "James",
	"middle": [],
	"last": "Cheney",
	"suffix": ""
	}
	],
	"year": 2013,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {
	"DOI": [
	"10.14279/tuj.eceasst.57.879"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Raghu Rajkumar, Nate Foster, Sam Lindley, and James Cheney. 2013. Lenses for Web Data. ECEASST, 57.",
	"links": null
	},
	"BIBREF28": {
	"ref_id": "b28",
	"title": "GERBIL -Benchmarking Named Entity Recognition and Linking Consistently. Semantic Web",
	"authors": [
	{
	"first": "Michael",
	"middle": [],
	"last": "R\u00f6der",
	"suffix": ""
	},
	{
	"first": "Ricardo",
	"middle": [],
	"last": "Usbeck",
	"suffix": ""
	},
	{
	"first": "Axel-Cyrille Ngonga",
	"middle": [],
	"last": "Ngomo",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "",
	"volume": "9",
	"issue": "",
	"pages": "605--625",
	"other_ids": {
	"DOI": [
	"10.3233/SW-170286"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Michael R\u00f6der, Ricardo Usbeck, and Axel- Cyrille Ngonga Ngomo. 2018. GERBIL - Benchmarking Named Entity Recognition and Linking Consistently. Semantic Web, 9(5):605-625.",
	"links": null
	},
	"BIBREF29": {
	"ref_id": "b29",
	"title": "VoxEL: A Benchmark Dataset for Multilingual Entity Linking",
	"authors": [
	{
	"first": "Henry",
	"middle": [],
	"last": "Rosales-M\u00e9ndez",
	"suffix": ""
	},
	{
	"first": "Aidan",
	"middle": [],
	"last": "Hogan",
	"suffix": ""
	},
	{
	"first": "Barbara",
	"middle": [],
	"last": "Poblete",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "The Semantic Web -ISWC 2018 -17th International Semantic Web Conference",
	"volume": "11137",
	"issue": "",
	"pages": "170--186",
	"other_ids": {
	"DOI": [
	"10.1007/978-3-030-00668-6_11"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Henry Rosales-M\u00e9ndez, Aidan Hogan, and Barbara Poblete. 2018a. VoxEL: A Benchmark Dataset for Multilingual Entity Linking. In The Semantic Web - ISWC 2018 -17th International Semantic Web Con- ference, Monterey, CA, USA, October 8-12, 2018, Proceedings, Part II, volume 11137 of Lecture Notes in Computer Science, pages 170-186. Springer.",
	"links": null
	},
	"BIBREF30": {
	"ref_id": "b30",
	"title": "NIFify: Towards Better Quality Entity Linking Datasets",
	"authors": [
	{
	"first": "Henry",
	"middle": [],
	"last": "Rosales-M\u00e9ndez",
	"suffix": ""
	},
	{
	"first": "Aidan",
	"middle": [],
	"last": "Hogan",
	"suffix": ""
	},
	{
	"first": "Barbara",
	"middle": [],
	"last": "Poblete",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "Companion of The",
	"volume": "",
	"issue": "",
	"pages": "",
	"other_ids": {
	"DOI": [
	"10.1145/3308560.3316465"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Henry Rosales-M\u00e9ndez, Aidan Hogan, and Barbara Poblete. 2019. NIFify: Towards Better Quality En- tity Linking Datasets. In Companion of The 2019",
	"links": null
	},
	"BIBREF31": {
	"ref_id": "b31",
	"title": "World Wide Web Conference",
	"authors": [],
	"year": 2019,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "815--818",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "World Wide Web Conference, WWW 2019, San Fran- cisco, CA, USA, May 13-17, 2019., pages 815-818. ACM.",
	"links": null
	},
	"BIBREF32": {
	"ref_id": "b32",
	"title": "What Should Entity Linking link?",
	"authors": [
	{
	"first": "Henry",
	"middle": [],
	"last": "Rosales-M\u00e9ndez",
	"suffix": ""
	},
	{
	"first": "Barbara",
	"middle": [],
	"last": "Poblete",
	"suffix": ""
	},
	{
	"first": "Aidan",
	"middle": [],
	"last": "Hogan",
	"suffix": ""
	}
	],
	"year": 2018,
	"venue": "Proceedings of the 12th Alberto Mendelzon International Workshop on Foundations of Data Management",
	"volume": "15",
	"issue": "",
	"pages": "",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Henry Rosales-M\u00e9ndez, Barbara Poblete, and Aidan Hogan. 2018b. What Should Entity Linking link? In Proceedings of the 12th Alberto Mendelzon In- ternational Workshop on Foundations of Data Man- agement, Cali, Colombia, May 21-25, 2018., vol- ume 2100 of CEUR Workshop Proceedings, page 15. CEUR-WS.org.",
	"links": null
	},
	"BIBREF33": {
	"ref_id": "b33",
	"title": "Named Entity Recognition -Is There a Glass Ceiling?",
	"authors": [
	{
	"first": "Tomasz",
	"middle": [],
	"last": "Stanislawek",
	"suffix": ""
	},
	{
	"first": "Anna",
	"middle": [],
	"last": "Wr\u00f3blewska",
	"suffix": ""
	},
	{
	"first": "Alicja",
	"middle": [],
	"last": "W\u00f3jcicka",
	"suffix": ""
	},
	{
	"first": "Daniel",
	"middle": [],
	"last": "Ziembicki",
	"suffix": ""
	},
	{
	"first": "Przemyslaw",
	"middle": [],
	"last": "Biecek",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "Proceedings of the 23rd Conference on Computational Natural Language Learning",
	"volume": "",
	"issue": "",
	"pages": "624--633",
	"other_ids": {
	"DOI": [
	"10.18653/v1/K19-1058"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Tomasz Stanislawek, Anna Wr\u00f3blewska, Alicja W\u00f3j- cicka, Daniel Ziembicki, and Przemyslaw Biecek. 2019. Named Entity Recognition -Is There a Glass Ceiling? In Proceedings of the 23rd Conference on Computational Natural Language Learning, CoNLL 2019, Hong Kong, China, November 3-4, 2019, pages 624-633. Association for Computational Lin- guistics.",
	"links": null
	},
	"BIBREF34": {
	"ref_id": "b34",
	"title": "Improving Named Entity Linking Corpora Quality. RANLP 2019",
	"authors": [
	{
	"first": "Albert",
	"middle": [],
	"last": "Weichselbraun",
	"suffix": ""
	},
	{
	"first": "M",
	"middle": [
	"P"
	],
	"last": "Adrian",
	"suffix": ""
	},
	{
	"first": "Philipp",
	"middle": [],
	"last": "Brasoveanu",
	"suffix": ""
	},
	{
	"first": "Lyndon Jb",
	"middle": [],
	"last": "Kuntschik",
	"suffix": ""
	},
	{
	"first": "",
	"middle": [],
	"last": "Nixon",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "",
	"volume": "",
	"issue": "",
	"pages": "1328--1337",
	"other_ids": {},
	"num": null,
	"urls": [],
	"raw_text": "Albert Weichselbraun, Adrian MP Brasoveanu, Philipp Kuntschik, and Lyndon JB Nixon. 2019a. Improv- ing Named Entity Linking Corpora Quality. RANLP 2019, pages 1328-1337.",
	"links": null
	},
	"BIBREF35": {
	"ref_id": "b35",
	"title": "Name Variants for Improving Entity Discovery and Linking",
	"authors": [
	{
	"first": "Albert",
	"middle": [],
	"last": "Weichselbraun",
	"suffix": ""
	},
	{
	"first": "Philipp",
	"middle": [],
	"last": "Kuntschik",
	"suffix": ""
	},
	{
	"first": "Adrian",
	"middle": [
	"M P"
	],
	"last": "Brasoveanu",
	"suffix": ""
	}
	],
	"year": 2019,
	"venue": "2nd Conference on Language, Data and Knowledge, LDK 2019",
	"volume": "70",
	"issue": "",
	"pages": "1--14",
	"other_ids": {
	"DOI": [
	"10.4230/OASIcs.LDK.2019.14"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Albert Weichselbraun, Philipp Kuntschik, and Adrian M. P. Brasoveanu. 2019b. Name Variants for Im- proving Entity Discovery and Linking. In 2nd Con- ference on Language, Data and Knowledge, LDK 2019, May 20-23, 2019, Leipzig, Germany., vol- ume 70 of OASICS, pages 14:1-14:15. Schloss Dagstuhl -Leibniz-Zentrum fuer Informatik.",
	"links": null
	},
	"BIBREF36": {
	"ref_id": "b36",
	"title": "The xLiMe system: Cross-lingual and Cross-modal Semantic Annotation, Search and Recommendation over Live-TV, News and Social Media Streams",
	"authors": [
	{
	"first": "Lei",
	"middle": [],
	"last": "Zhang",
	"suffix": ""
	},
	{
	"first": "Andreas",
	"middle": [],
	"last": "Thalhammer",
	"suffix": ""
	},
	{
	"first": "Achim",
	"middle": [],
	"last": "Rettinger",
	"suffix": ""
	},
	{
	"first": "Michael",
	"middle": [],
	"last": "F\u00e4rber",
	"suffix": ""
	},
	{
	"first": "Aditya",
	"middle": [],
	"last": "Mogadala",
	"suffix": ""
	},
	{
	"first": "Ronald",
	"middle": [],
	"last": "Denaux",
	"suffix": ""
	}
	],
	"year": 2017,
	"venue": "J. Web Semant",
	"volume": "",
	"issue": "",
	"pages": "20--30",
	"other_ids": {
	"DOI": [
	"10.1016/j.websem.2017.03.002"
	]
	},
	"num": null,
	"urls": [],
	"raw_text": "Lei Zhang, Andreas Thalhammer, Achim Rettinger, Michael F\u00e4rber, Aditya Mogadala, and Ronald De- naux. 2017. The xLiMe system: Cross-lingual and Cross-modal Semantic Annotation, Search and Rec- ommendation over Live-TV, News and Social Media Streams. J. Web Semant., 46-47:20-30.",
	"links": null
	}
	},
	"ref_entries": {
	"TABREF1": {
	"text": "Understanding differences between annotator results. Numbers were added in order to clarify which entities were retrieved.",
	"type_str": "table",
	"content": "<table><tr><td>pedia entity.</td><td/><td/></tr><tr><td colspan=\"3\">2. The annotation style OMAX also ignores over-</td></tr><tr><td colspan=\"3\">lapping entities but extracts the maximum</td></tr><tr><td colspan=\"3\">number of entities from a given text snippet:</td></tr><tr><td>m dbr:S tar_T rek [Star Trek]</td><td>, m dbr:S tar_T rek:_Picard [Picard]</td><td>.</td></tr></table>",
	"html": null,
	"num": null
	},
	"TABREF3": {
	"text": "Understanding differences between lenses. Numbers were added in order to clarify which entities were retrieved.",
	"type_str": "table",
	"content": "<table/>",
	"html": null,
	"num": null
	},
	"TABREF5": {
	"text": "Basic statistics.",
	"type_str": "table",
	"content": "<table/>",
	"html": null,
	"num": null
	},
	"TABREF7": {
	"text": "Core set experiments with multiple lenses (m -micro; M -macro; p -precision; r -recall; F1 -F1).",
	"type_str": "table",
	"content": "<table><tr><td>Corpus</td><td>System</td><td>mP</td><td>mR</td><td>mF1 MP</td><td>MR</td><td>MF1</td></tr><tr><td/><td>AIDA</td><td colspan=\"5\">0.50 0.49 0.50 0.46 0.47 0.46</td></tr><tr><td>All partitions</td><td>Spotlight</td><td colspan=\"5\">0.65 0.48 0.54 0.66 0.50 0.50</td></tr><tr><td>(3809 entities)</td><td colspan=\"6\">Recognyze 0.69 0.49 0.57 0.70 0.48 0.56</td></tr></table>",
	"html": null,
	"num": null
	}
	}
	}
	}