| { |
| "paper_id": "2021", |
| "header": { |
| "generated_with": "S2ORC 1.0.0", |
| "date_generated": "2023-01-19T03:35:46.598132Z" |
| }, |
| "title": "A COVID-19 news coverage mood map of Europe", |
| "authors": [ |
| { |
| "first": "Frankie", |
| "middle": [], |
| "last": "Robertson", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Jyv\u00e4skyl\u00e4", |
| "location": {} |
| }, |
| "email": "frankie@robertson.name" |
| }, |
| { |
| "first": "Jarkko", |
| "middle": [], |
| "last": "Lagus", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Helsinki", |
| "location": {} |
| }, |
| "email": "jalagus@cs.helsinki.fi" |
| }, |
| { |
| "first": "Kaisla", |
| "middle": [], |
| "last": "Kajava", |
| "suffix": "", |
| "affiliation": { |
| "laboratory": "", |
| "institution": "University of Helsinki", |
| "location": {} |
| }, |
| "email": "kaisla.kajava@helsinki.fi" |
| } |
| ], |
| "year": "", |
| "venue": null, |
| "identifiers": {}, |
| "abstract": "We present a COVID-19 news dashboard which visualizes sentiment in pandemic news coverage in different languages across Europe. The dashboard shows analyses for positive/neutral/negative sentiment and moral sentiment for news articles across countries and languages. First we extract news articles from news-crawl. Then we use a pre-trained multilingual BERT model for sentiment analysis of news article headlines and a dictionary and word vectors-based method for moral sentiment analysis of news articles. The resulting dashboard gives a unified overview of news events on COVID-19 news overall sentiment, and the region and language of publication from the period starting from the beginning of January 2020 to the end of January 2021.", |
| "pdf_parse": { |
| "paper_id": "2021", |
| "_pdf_hash": "", |
| "abstract": [ |
| { |
| "text": "We present a COVID-19 news dashboard which visualizes sentiment in pandemic news coverage in different languages across Europe. The dashboard shows analyses for positive/neutral/negative sentiment and moral sentiment for news articles across countries and languages. First we extract news articles from news-crawl. Then we use a pre-trained multilingual BERT model for sentiment analysis of news article headlines and a dictionary and word vectors-based method for moral sentiment analysis of news articles. The resulting dashboard gives a unified overview of news events on COVID-19 news overall sentiment, and the region and language of publication from the period starting from the beginning of January 2020 to the end of January 2021.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Abstract", |
| "sec_num": null |
| } |
| ], |
| "body_text": [ |
| { |
| "text": "The response to the COVID-19 pandemic and its news coverage worldwide has been marked by tension between state-level actions, and those of regional organisations such as the European Union, and the essentially borderless nature of the virus itself. This paper presents the COVID-19 news coverage mood map of Europe which visualizes sentiment in news coverage in different European languages. Within the European context, with its many small languages, this multi-lingual approach is vital for working beyond the state-level.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "In order to evaluate the possibilities of automatic sentiment analysis models in this context, we first created a multilingual COVID-19 news corpus from the websites of state broadcasters. We then applied two types of automatic sentiment analysis to the articles. Finally, we created a dashboard containing a number of interactive plots based on the analysed corpus 1 . This report details work undertaken at the EMBEDDIA Hackashop.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "1 http://covidmoodmap.rahtiapp.fi/ European languages av az ba be bg bs ca ce cs cv cy da de el en es et eu fi fo fr fy ga gl gv hr hu is it kl kv la lb lt lv mk mt nb nl nn no oc os pl pt rm ro ru sk sl sq sr sv tr tt uk yi Langdetect bg ca cs cy da de el en es et fi fr hr hu it lt lv mk nl no pl pt ro ru sk sl sq sv tr uk mBERT bg ca cs cy da de el en es et fi fr hr hu it lt lv mk nl no pl pt ro ru sk sl sq sv tr uk MUSE bg ca cs da de el en es et fi fr hr hu it mk nl no pl pt ro ru sk sl sv tr uk >20 items bg ca cs cy de el en es et fi fr it lt mk nl pl pt ro ru sv tr uk ", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Introduction", |
| "sec_num": "1" |
| }, |
| { |
| "text": "We used the news-please extractor (Hamborg et al., 2017) on news-crawl dumps to obtain a multilingual corpus of European COVID-19 news. Newscrawl is a web crawl provided by the Common Crawl organisation which is updated more frequently and contains only data from news websites 2 . In order to keep the size of the corpus manageable and the extraction time reasonable, a list of internet domain names of European state broadcasters was first obtained from Wikidata, since filtering at the domain level allows for faster processing of Common Crawl dumps. Articles without a language detected by the langdetect language detector 3 were discarded. COVID-19 keyword filtering was also applied, detailed in Section 3.1. Table 1 includes a summary of the considered European languages, their support by langdetect, and the set for which 20 or more items were ultimately extracted.", |
| "cite_spans": [ |
| { |
| "start": 34, |
| "end": 56, |
| "text": "(Hamborg et al., 2017)", |
| "ref_id": "BIBREF7" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 715, |
| "end": 722, |
| "text": "Table 1", |
| "ref_id": "TABREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Corpus extraction", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The resulting corpus contains 468 thousand articles. It is thus just over a quarter of the size of the comparable AYLIEN Coronavirus News Dataset 4 corpus which has 1 673 thousand news articles. Our corpus contains news from a longer period of just over a year versus AYLIEN's, which contains just over a half a year. Most importantly, our corpus has at least 20 items in 22 languages, versus AYLIEN's corpus which is English only.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Corpus extraction", |
| "sec_num": "2" |
| }, |
| { |
| "text": "The full corpus does not include news from all European countries. Table 2 gives a coverage of the countries included in the corpus, while Table 1 gives the coverage of languages in the corpus. There are two possible reasons for the missing countries. The first is that news-crawl uses a fixed set of seeds, so one possibility is that the the websites of the state broadcaster for these countries was not on the seed list. Another possibility is that the article extraction of news-please was not able to deal with these countries. The recall of newsplease is estimated at 71%. Possible future work here would be to obtain and audit the seed list from news-crawl and try other article extraction software such as Trafilatura (Barbaresi, 2019) which has an estimated recall of 88% with higher precision. ", |
| "cite_spans": [ |
| { |
| "start": 725, |
| "end": 742, |
| "text": "(Barbaresi, 2019)", |
| "ref_id": "BIBREF0" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 67, |
| "end": 74, |
| "text": "Table 2", |
| "ref_id": "TABREF2" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Corpus extraction", |
| "sec_num": "2" |
| }, |
| { |
| "text": "All analyses were multilingual, and their coverage of different languages is compared in Table 1 . During development, the full list of tools and resources given by Pollak et al. (2021) was considered.", |
| "cite_spans": [ |
| { |
| "start": 165, |
| "end": 185, |
| "text": "Pollak et al. (2021)", |
| "ref_id": "BIBREF14" |
| } |
| ], |
| "ref_spans": [ |
| { |
| "start": 89, |
| "end": 96, |
| "text": "Table 1", |
| "ref_id": "TABREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Analyses", |
| "sec_num": "3" |
| }, |
| { |
| "text": "In order to detect keywords from fixed lists across languages, including those with inflectional endings which cause changes to the citation form itself, either lemmatisation or stemming must be performed. Since keyword search is also performed as a filtering step for creating the corpus, it should be fast. To achieve this goal, we applied a simple highrecall stemming-like scheme based on BPE. First we obtained the pretrained BPE (Sennrich et al., 2016) model from XLM-RoBERTa-large (Conneau et al., 2019) 5 . As a next step, all BPE tokens with length >= 5 are discarded from the BPE model. The full XLM tokenisation pipeline is then run as normal. The hope is that this segments the word into commonly repeating units, which are likely to include common inflectional endings. The decision to discard longer BPE tokens was made so that common longer words would still be segmented and to bound the maximum number of characters removed from the word, since removing too many is more likely to cause false positives. In cases where at least 3 BPE segments were generated, the last one is discarded. In all cases, the resulting stem or full token is at least 2 BPE characters and 5 characters, a wildcard appended to the end of the token. Matching was performed case-insensitively using the fast pyre2 6 library, which uses deterministic automata for matching.", |
| "cite_spans": [ |
| { |
| "start": 434, |
| "end": 457, |
| "text": "(Sennrich et al., 2016)", |
| "ref_id": "BIBREF16" |
| }, |
| { |
| "start": 487, |
| "end": 511, |
| "text": "(Conneau et al., 2019) 5", |
| "ref_id": null |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Keyword matching", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "While this scheme is certainly likely have lower precision than using a high quality lemmatiser, it is not language dependent beyond its central assumption: that sounds changes -if they occur -are limited towards the end of the word. The exact same procedure is applied to all languages. For comparison the Snowball stemmer 7 supports 15 languages, leaving 16 of those supported by langdetect unhandled. On the other hand, a state-of-the-art mulitlingual lemmatisers such as the Universal Lemmatiser of Kanerva et al. (2020) , which supports over 50 langauges is likely to be slower. Additionally, due to Universal Lemmatiser's architecture being adapted to batch scenarios, this implies adding and extra stage to the pipeline. That said, performing keyword matching based upon Universal Lemmatiser would be a good next step for the keyword matching, and the current scheme could be kept only for the few languages not supported by Universal Lemmatiser.", |
| "cite_spans": [ |
| { |
| "start": 503, |
| "end": 524, |
| "text": "Kanerva et al. (2020)", |
| "ref_id": "BIBREF10" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Keyword matching", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "Keyword lists for both COVID-19 keywords and names of European countries in all considered languages were obtained from Wikidata. For matching the topic of COVID-19, labels from the entities Q84263196 (the COVID-19 disease), Q82069695 (the SARS-CoV-2 virus), and Q89469904 (the hypernym of SARS-CoV-2; all corona viruses) were used. For non-English languages, the set of keywords was extended with the English keywords in case they have been used as loans, especially for example in the early stages of the epidemic. In addition, the commonly occurring trans-lingual patterns: corona * , korona * , covid * and \u043a\u043e\u0440\u043e\u043d\u0430 * were added to all lists. The lists of country names were used as-is.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Keyword matching", |
| "sec_num": "3.1" |
| }, |
| { |
| "text": "In recent years, deep pre-trained language models have produced state-of-the-art results in various natural language processing tasks. BERT models use self-attention layers from the transformer model (Vaswani et al., 2017) , which makes them useful for detecting contextual information in text: this BERT does by looking at the neighboring words on both sides of each word in the text.", |
| "cite_spans": [ |
| { |
| "start": 200, |
| "end": 222, |
| "text": "(Vaswani et al., 2017)", |
| "ref_id": "BIBREF17" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual sentiment analysis of news headlines", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "We used a multilingual BERT model (Devlin et al., 2018) fine-tuned by Pelicon et al. (2020) to classify news article headlines into the polar sentiment categories positive, neutral, and negative. The model was trained on a Slovenian news dataset and evaluated for zero-shot cross-lingual tasks on Croatian news. As the model was originally developed for classification of full news article texts, it is mostly trained on slices of these using the whole of BERT's maximum sequence length of 512. Here we are typically supplying much shorter sequences from the related, but distinct distribution of news headlines, which is likely to affect performance.", |
| "cite_spans": [ |
| { |
| "start": 34, |
| "end": 55, |
| "text": "(Devlin et al., 2018)", |
| "ref_id": "BIBREF4" |
| }, |
| { |
| "start": 70, |
| "end": 91, |
| "text": "Pelicon et al. (2020)", |
| "ref_id": "BIBREF13" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual sentiment analysis of news headlines", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "The results of our experiments show a nearconsistent peak in news coverage of the COVID-19 pandemic in the spring of 2020 across news sources, with the exception of some languages which are less represented in our overall news dataset. Similarly, the results show a peak in news coverage in the fall of 2020, which coincides with the second wave of the pandemic. Negative sentiment is most prevalent in spring 2020. Overall, negative was the most commonly predicted label, with neutral as the next common, and positive the least common. Table 3 shows the five countries with the most news article headlines classified as negative. News articles from the national broadcasters of those countries also constitute a significant portion of the overall data. ", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 537, |
| "end": 544, |
| "text": "Table 3", |
| "ref_id": "TABREF4" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Multilingual sentiment analysis of news headlines", |
| "sec_num": "3.2" |
| }, |
| { |
| "text": "Research shows (Kalimeri et al., 2019a; Curry et al., 2019; Mooijman et al., 2018) that accounting for moral sentiment in addition to other personality traits in natural language can give insights into many different societal phenomena. Methods based on moral sentiment have been previously used, for example, to predict street riots based on Twitter data (Mooijman et al., 2018) and analyze moral narratives in social media conversation about vaccination (Kalimeri et al., 2019b) . The basis for moral sentiment analysis is the moral foundations theory by Graham et al. (2013) and especially in this case the moral foundations dictionary (MFD) 2.0 8 . MFD is a word list that contains words categorized into five different sentiments: care, fairness, loyalty, authority, and sanctity.", |
| "cite_spans": [ |
| { |
| "start": 15, |
| "end": 39, |
| "text": "(Kalimeri et al., 2019a;", |
| "ref_id": "BIBREF8" |
| }, |
| { |
| "start": 40, |
| "end": 59, |
| "text": "Curry et al., 2019;", |
| "ref_id": "BIBREF3" |
| }, |
| { |
| "start": 60, |
| "end": 82, |
| "text": "Mooijman et al., 2018)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 356, |
| "end": 379, |
| "text": "(Mooijman et al., 2018)", |
| "ref_id": "BIBREF12" |
| }, |
| { |
| "start": 456, |
| "end": 480, |
| "text": "(Kalimeri et al., 2019b)", |
| "ref_id": "BIBREF9" |
| }, |
| { |
| "start": 557, |
| "end": 577, |
| "text": "Graham et al. (2013)", |
| "ref_id": "BIBREF6" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "The method we apply here follows the idea described by Kozlowski et al. (2019) in order to extract cultural dimensions from word embeddings. First, we extract antonym pairs using Word-Net (Fellbaum, 1998) per sentiment by searching synsets word-by-word and fetching the list of antonyms. This list of antonyms is then filtered based on the list containing words of opposite sentiment polarity. As an example, we search antonyms for word the \"peace\" from the list of positive polarity words related to care sentiment. We acquire the antonym \"war\" based on the WordNet synset search. As this word is also found from the negative polarity list of care dimension, we add this pair to the list of antonym pairs for the care sentiment.", |
| "cite_spans": [ |
| { |
| "start": 55, |
| "end": 78, |
| "text": "Kozlowski et al. (2019)", |
| "ref_id": "BIBREF11" |
| }, |
| { |
| "start": 188, |
| "end": 204, |
| "text": "(Fellbaum, 1998)", |
| "ref_id": "BIBREF5" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "These pairs we use to compute vectors representing moral sentiment dimension. This is done via encoding the words as word embeddings, doing simple subtraction of these antonym pairs, and computing the mean of these vectors per sentiment. This gives us one vector per sentiment, onto which we can then project any other word to measure the strength of that specific sentiment. As large number of words do not have antonyms that are found from the opposite polarity list, this leads to a noisy estimate of the dimension.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "To compute the sentiment of a document, each word of a document is first encoded into word embedding and then projected onto each moral sentiment vector. This gives a list of scores for each word that we can then just sum to obtain the final score for the document.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "This way of summing everything up creates an effect where longer documents will have stronger sentiment scores if most words are towards the same polarity per sentiment. This is a somewhat desirable effect since we do not wish to have onesentence articles to have the same weight as a full article.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "Multilinguality creates issues since the original MFD is only officially available in English. The issue is solved by using aligned word embeddings and doing approximate translation via distancebased search in the aligned word embedding space. For this, we use embeddings by Conneau et al. (2017) . An alternative option would be to translate the words exactly, but since all languages might have very culture-specific semantics that do not directly translate, it might remove or even change the moral sentiment of the word. Doing approximate translation directly in word embedding space does not suffer from this, as we only search for the word embedding with the closest meaning. This should also preserve the sentiment information.", |
| "cite_spans": [ |
| { |
| "start": 275, |
| "end": 296, |
| "text": "Conneau et al. (2017)", |
| "ref_id": "BIBREF2" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "The distance-based search method does not guarantee a good translation, but should in most cases work better than exact translation. If the exact translation is good and the embedding is close to the actual meaning, the distance-based method will approximately retrieve the same embedding. If the exact translation is good, but because of cultural differences the semantic meaning has shifted, the distance-based method should retrieve an embedding that is closer to original semantics, even though the exact word might be different from the exact translation. So in both of these cases, the distance-based method should yield better results.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "The results show that the strongest sentiment in the positive direction was sanctity and in the negative direction loyalty, both being clearly distinct from the other three sentiments in magnitude. Different sentiments showed fluctuation over time and countries, but overall the sentiment seemed to stay in the same polarity, suggesting no drastic changes in the way COVID-19 was covered in the news over time from the moral sentiment point of view.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Multilingual moral sentiment analysis of news articles", |
| "sec_num": "3.3" |
| }, |
| { |
| "text": "In order to visualise the results of the analyses, we created a dashboard using the Dash framework 9 . The resulting application makes heavy use of analytical queries which tend to feature range selections and grouping based on dates as well as numerical aggregates such as value summations and counts. To run these queries efficiently across the whole data set of 468 thousand articles, we used the DuckDB column database (Raasveldt and M\u00fchleisen, 2019) .", |
| "cite_spans": [ |
| { |
| "start": 423, |
| "end": 454, |
| "text": "(Raasveldt and M\u00fchleisen, 2019)", |
| "ref_id": "BIBREF15" |
| } |
| ], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Visualisation", |
| "sec_num": "4" |
| }, |
| { |
| "text": "The chief dimensions visualised as independent grouping variables in space were date and either the country of production or the country mentioned in articles. For plots in which these variables could not be shown spatially, the user was given the option of filtering using them. The language used in the articles and type of sentiment were also available as filters. The main dependent variables were are either raw article counts or lumped measures. For visualising only time as a visual grouping, a bar chart was used, while for visualising only country, a choropleth map, an example of which is given in Figure 1 . Finally, an animated choropleth map showing consecutive time slices corresponding to weeks taken Monday to Sunday groups by both time and country.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 608, |
| "end": 616, |
| "text": "Figure 1", |
| "ref_id": "FIGREF0" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Visualisation", |
| "sec_num": "4" |
| }, |
| { |
| "text": "The lumped measure used for polar sentiment is a simple ratio, with the neutral class included to pull the measure towards zero: positive + 1 2 neutral positive + neutral + negative For moral sentiments, we have already obtained a measure for each document and each moral sentiment. These are simply summed to create a single aggregate bipolar measure of sentiment strength per moral sentiment (see Figure 2 for an example). The sentiment estimate is rather noisy, so looking at the absolute values is not recommended. A better way to look at these numbers, is to look at them in relation to other countries or time spans. This tells how different countries differ in the way they represent this information and how is the overall trend progressing over time.", |
| "cite_spans": [], |
| "ref_spans": [ |
| { |
| "start": 399, |
| "end": 407, |
| "text": "Figure 2", |
| "ref_id": "FIGREF1" |
| } |
| ], |
| "eq_spans": [], |
| "section": "Visualisation", |
| "sec_num": "4" |
| }, |
| { |
| "text": "We have presented a COVID-19 news dashboard for exploration of reporting across time and from and about different countries during the COVID-19 epidemic. The dashboard demonstrates the potential of automated multilingual text analysis for understanding reporting on complex phenomena such as the COVID-19 crisis beyond the state-level. This type of tool could be integrated into a system used by news agencies to track news trends. Beyond COVID-19, it could be used to plan coverage of other national or global news events such as elections, international summits, or sports events.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "The visualizations in the dashboard do seem to line up with the authors' preconceived ideas about sentiments during COVID-19 and their evolution.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "However, all analyses in the dashboard were produced automatically and have not undergone evaluation within this context. Since the analyses themselves may not be entirely accurate, the resulting plots may be misleading, and thus should not be used as a basis for decision making. Evaluation of the underlying techniques is a clear next step for this work.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Conclusion", |
| "sec_num": "5" |
| }, |
| { |
| "text": "https://commoncrawl.org/2016/10/ news-dataset-available/ 3 https://github.com/Mimino666/ langdetect", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://aylien.com/blog/ free-coronavirus-news-dataset", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://huggingface.co/ xlm-roberta-large 6 https://github.com/andreasvc/pyre2 7 As described at https://snowballstem.org/", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://osf.io/ezn37/", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| }, |
| { |
| "text": "https://plotly.com/dash/", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "", |
| "sec_num": null |
| } |
| ], |
| "back_matter": [ |
| { |
| "text": "The authors wish to thank the organisers of the EMBEDDIA Hackashop 2021 for organising the event and CSC -IT Center for Science, Finland, for computational resources.", |
| "cite_spans": [], |
| "ref_spans": [], |
| "eq_spans": [], |
| "section": "Acknowledgements", |
| "sec_num": null |
| } |
| ], |
| "bib_entries": { |
| "BIBREF0": { |
| "ref_id": "b0", |
| "title": "Generic web content extraction with open-source software", |
| "authors": [ |
| { |
| "first": "Adrien", |
| "middle": [], |
| "last": "Barbaresi", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "KONVENS 2019", |
| "volume": "", |
| "issue": "", |
| "pages": "267--268", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Adrien Barbaresi. 2019. Generic web content extrac- tion with open-source software. In KONVENS 2019, pages 267-268. GSCL.", |
| "links": null |
| }, |
| "BIBREF1": { |
| "ref_id": "b1", |
| "title": "Unsupervised cross-lingual representation learning at scale", |
| "authors": [ |
| { |
| "first": "Alexis", |
| "middle": [], |
| "last": "Conneau", |
| "suffix": "" |
| }, |
| { |
| "first": "Kartikay", |
| "middle": [], |
| "last": "Khandelwal", |
| "suffix": "" |
| }, |
| { |
| "first": "Naman", |
| "middle": [], |
| "last": "Goyal", |
| "suffix": "" |
| }, |
| { |
| "first": "Vishrav", |
| "middle": [], |
| "last": "Chaudhary", |
| "suffix": "" |
| }, |
| { |
| "first": "Guillaume", |
| "middle": [], |
| "last": "Wenzek", |
| "suffix": "" |
| }, |
| { |
| "first": "Francisco", |
| "middle": [], |
| "last": "Guzm\u00e1n", |
| "suffix": "" |
| }, |
| { |
| "first": "Edouard", |
| "middle": [], |
| "last": "Grave", |
| "suffix": "" |
| }, |
| { |
| "first": "Myle", |
| "middle": [], |
| "last": "Ott", |
| "suffix": "" |
| }, |
| { |
| "first": "Luke", |
| "middle": [], |
| "last": "Zettlemoyer", |
| "suffix": "" |
| }, |
| { |
| "first": "Veselin", |
| "middle": [], |
| "last": "Stoyanov", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": { |
| "arXiv": [ |
| "arXiv:1911.02116" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzm\u00e1n, Edouard Grave, Myle Ott, Luke Zettle- moyer, and Veselin Stoyanov. 2019. Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116.", |
| "links": null |
| }, |
| "BIBREF2": { |
| "ref_id": "b2", |
| "title": "Word translation without parallel data", |
| "authors": [ |
| { |
| "first": "Alexis", |
| "middle": [], |
| "last": "Conneau", |
| "suffix": "" |
| }, |
| { |
| "first": "Guillaume", |
| "middle": [], |
| "last": "Lample", |
| "suffix": "" |
| }, |
| { |
| "first": "Marc'aurelio", |
| "middle": [], |
| "last": "Ranzato", |
| "suffix": "" |
| }, |
| { |
| "first": "Ludovic", |
| "middle": [], |
| "last": "Denoyer", |
| "suffix": "" |
| }, |
| { |
| "first": "Herv\u00e9", |
| "middle": [], |
| "last": "J\u00e9gou", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": { |
| "arXiv": [ |
| "arXiv:1710.04087" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Alexis Conneau, Guillaume Lample, Marc'Aurelio Ranzato, Ludovic Denoyer, and Herv\u00e9 J\u00e9gou. 2017. Word translation without parallel data. arXiv preprint arXiv:1710.04087.", |
| "links": null |
| }, |
| "BIBREF3": { |
| "ref_id": "b3", |
| "title": "Is it good to cooperate? testing the theory of morality-ascooperation in 60 societies", |
| "authors": [ |
| { |
| "first": "O", |
| "middle": [], |
| "last": "Curry", |
| "suffix": "" |
| }, |
| { |
| "first": "D", |
| "middle": [], |
| "last": "Whitehouse", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Mullins", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Current Anthropology", |
| "volume": "", |
| "issue": "1", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "O Curry, H Whitehouse, and D Mullins. 2019. Is it good to cooperate? testing the theory of morality-as- cooperation in 60 societies. Current Anthropology, 60(1).", |
| "links": null |
| }, |
| "BIBREF4": { |
| "ref_id": "b4", |
| "title": "Bert: Pre-training of deep bidirectional transformers for language understanding", |
| "authors": [ |
| { |
| "first": "Jacob", |
| "middle": [], |
| "last": "Devlin", |
| "suffix": "" |
| }, |
| { |
| "first": "Ming-Wei", |
| "middle": [], |
| "last": "Chang", |
| "suffix": "" |
| }, |
| { |
| "first": "Kenton", |
| "middle": [], |
| "last": "Lee", |
| "suffix": "" |
| }, |
| { |
| "first": "Kristina", |
| "middle": [], |
| "last": "Toutanova", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": { |
| "arXiv": [ |
| "arXiv:1810.04805" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2018. Bert: Pre-training of deep bidirectional transformers for language understand- ing. arXiv preprint arXiv:1810.04805.", |
| "links": null |
| }, |
| "BIBREF5": { |
| "ref_id": "b5", |
| "title": "WordNet: An Electronic Lexical Database", |
| "authors": [ |
| { |
| "first": "Christiane", |
| "middle": [], |
| "last": "Fellbaum", |
| "suffix": "" |
| } |
| ], |
| "year": 1998, |
| "venue": "", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Christiane Fellbaum. 1998. WordNet: An Electronic Lexical Database. Bradford Books.", |
| "links": null |
| }, |
| "BIBREF6": { |
| "ref_id": "b6", |
| "title": "Chapter two -moral foundations theory: The pragmatic validity of moral pluralism", |
| "authors": [ |
| { |
| "first": "Jesse", |
| "middle": [], |
| "last": "Graham", |
| "suffix": "" |
| }, |
| { |
| "first": "Jonathan", |
| "middle": [], |
| "last": "Haidt", |
| "suffix": "" |
| }, |
| { |
| "first": "Sena", |
| "middle": [], |
| "last": "Koleva", |
| "suffix": "" |
| }, |
| { |
| "first": "Matt", |
| "middle": [], |
| "last": "Motyl", |
| "suffix": "" |
| }, |
| { |
| "first": "Ravi", |
| "middle": [], |
| "last": "Iyer", |
| "suffix": "" |
| }, |
| { |
| "first": "Sean", |
| "middle": [ |
| "P" |
| ], |
| "last": "Wojcik", |
| "suffix": "" |
| }, |
| { |
| "first": "Peter", |
| "middle": [ |
| "H" |
| ], |
| "last": "Ditto", |
| "suffix": "" |
| } |
| ], |
| "year": 2013, |
| "venue": "Advances in Experimental Social Psychology", |
| "volume": "47", |
| "issue": "", |
| "pages": "55--130", |
| "other_ids": { |
| "DOI": [ |
| "10.1016/B978-0-12-407236-7.00002-4" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jesse Graham, Jonathan Haidt, Sena Koleva, Matt Motyl, Ravi Iyer, Sean P. Wojcik, and Peter H. Ditto. 2013. Chapter two -moral foundations theory: The pragmatic validity of moral pluralism. volume 47 of Advances in Experimental Social Psychology, pages 55-130. Academic Press.", |
| "links": null |
| }, |
| "BIBREF7": { |
| "ref_id": "b7", |
| "title": "news-please: A generic news crawler and extractor", |
| "authors": [ |
| { |
| "first": "Felix", |
| "middle": [], |
| "last": "Hamborg", |
| "suffix": "" |
| }, |
| { |
| "first": "Norman", |
| "middle": [], |
| "last": "Meuschke", |
| "suffix": "" |
| }, |
| { |
| "first": "Corinna", |
| "middle": [], |
| "last": "Breitinger", |
| "suffix": "" |
| }, |
| { |
| "first": "Bela", |
| "middle": [], |
| "last": "Gipp", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Proceedings of the 15th International Symposium of Information Science", |
| "volume": "", |
| "issue": "", |
| "pages": "218--223", |
| "other_ids": { |
| "DOI": [ |
| "10.5281/zenodo.4120316" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Felix Hamborg, Norman Meuschke, Corinna Bre- itinger, and Bela Gipp. 2017. news-please: A generic news crawler and extractor. In Proceedings of the 15th International Symposium of Information Science, pages 218-223.", |
| "links": null |
| }, |
| "BIBREF8": { |
| "ref_id": "b8", |
| "title": "Predicting demographics, moral foundations, and human values from digital behaviours", |
| "authors": [ |
| { |
| "first": "Kyriaki", |
| "middle": [], |
| "last": "Kalimeri", |
| "suffix": "" |
| }, |
| { |
| "first": "G", |
| "middle": [], |
| "last": "Mariano", |
| "suffix": "" |
| }, |
| { |
| "first": "Matteo", |
| "middle": [], |
| "last": "Beir\u00f3", |
| "suffix": "" |
| }, |
| { |
| "first": "Robert", |
| "middle": [], |
| "last": "Delfino", |
| "suffix": "" |
| }, |
| { |
| "first": "Ciro", |
| "middle": [], |
| "last": "Raleigh", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Cattuto", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Computers in Human Behavior", |
| "volume": "92", |
| "issue": "", |
| "pages": "428--445", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kyriaki Kalimeri, Mariano G Beir\u00f3, Matteo Delfino, Robert Raleigh, and Ciro Cattuto. 2019a. Predicting demographics, moral foundations, and human val- ues from digital behaviours. Computers in Human Behavior, 92:428-445.", |
| "links": null |
| }, |
| "BIBREF9": { |
| "ref_id": "b9", |
| "title": "Human values and attitudes towards vaccination in social media", |
| "authors": [ |
| { |
| "first": "Kyriaki", |
| "middle": [], |
| "last": "Kalimeri", |
| "suffix": "" |
| }, |
| { |
| "first": "Mariano", |
| "middle": [ |
| "G" |
| ], |
| "last": "Beir\u00f3", |
| "suffix": "" |
| }, |
| { |
| "first": "Alessandra", |
| "middle": [], |
| "last": "Urbinati", |
| "suffix": "" |
| }, |
| { |
| "first": "Andrea", |
| "middle": [], |
| "last": "Bonanomi", |
| "suffix": "" |
| }, |
| { |
| "first": "Alessandro", |
| "middle": [], |
| "last": "Rosina", |
| "suffix": "" |
| }, |
| { |
| "first": "Ciro", |
| "middle": [], |
| "last": "Cattuto", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Companion Proceedings of The 2019 World Wide Web Conference", |
| "volume": "", |
| "issue": "", |
| "pages": "248--254", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Kyriaki Kalimeri, Mariano G. Beir\u00f3, Alessandra Urbinati, Andrea Bonanomi, Alessandro Rosina, and Ciro Cattuto. 2019b. Human values and atti- tudes towards vaccination in social media. In Com- panion Proceedings of The 2019 World Wide Web Conference, pages 248-254.", |
| "links": null |
| }, |
| "BIBREF10": { |
| "ref_id": "b10", |
| "title": "Universal lemmatizer: A sequence to sequence model for lemmatizing universal dependencies treebanks", |
| "authors": [ |
| { |
| "first": "Jenna", |
| "middle": [], |
| "last": "Kanerva", |
| "suffix": "" |
| }, |
| { |
| "first": "Filip", |
| "middle": [], |
| "last": "Ginter", |
| "suffix": "" |
| }, |
| { |
| "first": "Tapio", |
| "middle": [], |
| "last": "Salakoski", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "Natural Language Engineering", |
| "volume": "", |
| "issue": "", |
| "pages": "1--30", |
| "other_ids": { |
| "DOI": [ |
| "10.1017/S1351324920000224" |
| ] |
| }, |
| "num": null, |
| "urls": [], |
| "raw_text": "Jenna Kanerva, Filip Ginter, and Tapio Salakoski. 2020. Universal lemmatizer: A sequence to sequence model for lemmatizing universal dependencies tree- banks. Natural Language Engineering, pages 1-30.", |
| "links": null |
| }, |
| "BIBREF11": { |
| "ref_id": "b11", |
| "title": "The geometry of culture: Analyzing the meanings of class through word embeddings", |
| "authors": [ |
| { |
| "first": "C", |
| "middle": [], |
| "last": "Austin", |
| "suffix": "" |
| }, |
| { |
| "first": "Matt", |
| "middle": [], |
| "last": "Kozlowski", |
| "suffix": "" |
| }, |
| { |
| "first": "James A", |
| "middle": [], |
| "last": "Taddy", |
| "suffix": "" |
| }, |
| { |
| "first": "", |
| "middle": [], |
| "last": "Evans", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "American Sociological Review", |
| "volume": "84", |
| "issue": "5", |
| "pages": "905--949", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Austin C Kozlowski, Matt Taddy, and James A Evans. 2019. The geometry of culture: Analyzing the mean- ings of class through word embeddings. American Sociological Review, 84(5):905-949.", |
| "links": null |
| }, |
| "BIBREF12": { |
| "ref_id": "b12", |
| "title": "Moralization in social networks and the emergence of violence during protests", |
| "authors": [ |
| { |
| "first": "Marlon", |
| "middle": [], |
| "last": "Mooijman", |
| "suffix": "" |
| }, |
| { |
| "first": "Joe", |
| "middle": [], |
| "last": "Hoover", |
| "suffix": "" |
| }, |
| { |
| "first": "Ying", |
| "middle": [], |
| "last": "Lin", |
| "suffix": "" |
| }, |
| { |
| "first": "Ji", |
| "middle": [], |
| "last": "Heng", |
| "suffix": "" |
| }, |
| { |
| "first": "Morteza", |
| "middle": [], |
| "last": "Dehghani", |
| "suffix": "" |
| } |
| ], |
| "year": 2018, |
| "venue": "Nature human behaviour", |
| "volume": "2", |
| "issue": "6", |
| "pages": "389--396", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Marlon Mooijman, Joe Hoover, Ying Lin, Heng Ji, and Morteza Dehghani. 2018. Moralization in social net- works and the emergence of violence during protests. Nature human behaviour, 2(6):389-396.", |
| "links": null |
| }, |
| "BIBREF13": { |
| "ref_id": "b13", |
| "title": "Zero-shot learning for cross-lingual news sentiment classification", |
| "authors": [ |
| { |
| "first": "Andra\u017e", |
| "middle": [], |
| "last": "Pelicon", |
| "suffix": "" |
| }, |
| { |
| "first": "Marko", |
| "middle": [], |
| "last": "Pranji'c", |
| "suffix": "" |
| }, |
| { |
| "first": "Dragana", |
| "middle": [], |
| "last": "Miljkovi'c", |
| "suffix": "" |
| }, |
| { |
| "first": "Bla\u017e", |
| "middle": [], |
| "last": "\u0160krlj", |
| "suffix": "" |
| }, |
| { |
| "first": "Senja", |
| "middle": [], |
| "last": "Pollak", |
| "suffix": "" |
| } |
| ], |
| "year": 2020, |
| "venue": "Applied Sciences", |
| "volume": "10", |
| "issue": "17", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Andra\u017e Pelicon, Marko Pranji'c, Dragana Miljkovi'c, Bla\u017e \u0160krlj, and Senja Pollak. 2020. Zero-shot learn- ing for cross-lingual news sentiment classification. Applied Sciences, 10(17):5993.", |
| "links": null |
| }, |
| "BIBREF14": { |
| "ref_id": "b14", |
| "title": "EMBEDDIA tools, datasets and challenges: Resources and hackathon contributions", |
| "authors": [ |
| { |
| "first": "Senja", |
| "middle": [], |
| "last": "Pollak", |
| "suffix": "" |
| }, |
| { |
| "first": "Marko", |
| "middle": [ |
| "Robnik" |
| ], |
| "last": "\u0160ikonja", |
| "suffix": "" |
| }, |
| { |
| "first": "Matthew", |
| "middle": [], |
| "last": "Purver", |
| "suffix": "" |
| }, |
| { |
| "first": "Michele", |
| "middle": [], |
| "last": "Boggia", |
| "suffix": "" |
| }, |
| { |
| "first": "Ravi", |
| "middle": [], |
| "last": "Shekhar", |
| "suffix": "" |
| }, |
| { |
| "first": "Marko", |
| "middle": [], |
| "last": "Pranji\u0107", |
| "suffix": "" |
| }, |
| { |
| "first": "Salla", |
| "middle": [], |
| "last": "Salmela", |
| "suffix": "" |
| }, |
| { |
| "first": "Ivar", |
| "middle": [], |
| "last": "Krustok", |
| "suffix": "" |
| }, |
| { |
| "first": "Tarmo", |
| "middle": [], |
| "last": "Paju", |
| "suffix": "" |
| }, |
| { |
| "first": "Carl-Gustav", |
| "middle": [], |
| "last": "Linden", |
| "suffix": "" |
| }, |
| { |
| "first": "Leo", |
| "middle": [], |
| "last": "Lepp\u00e4nen", |
| "suffix": "" |
| }, |
| { |
| "first": "Elaine", |
| "middle": [], |
| "last": "Zosa", |
| "suffix": "" |
| }, |
| { |
| "first": "Matej", |
| "middle": [], |
| "last": "Ul\u010dar", |
| "suffix": "" |
| }, |
| { |
| "first": "Linda", |
| "middle": [], |
| "last": "Freienthal", |
| "suffix": "" |
| }, |
| { |
| "first": "Silver", |
| "middle": [], |
| "last": "Traat", |
| "suffix": "" |
| }, |
| { |
| "first": "Luis", |
| "middle": [ |
| "Adri\u00e1n" |
| ], |
| "last": "Cabrera-Diego", |
| "suffix": "" |
| }, |
| { |
| "first": "Matej", |
| "middle": [], |
| "last": "Martinc", |
| "suffix": "" |
| }, |
| { |
| "first": "Nada", |
| "middle": [], |
| "last": "Lavra\u010d", |
| "suffix": "" |
| }, |
| { |
| "first": "Bla\u017e", |
| "middle": [], |
| "last": "\u0160krlj", |
| "suffix": "" |
| }, |
| { |
| "first": "Martin", |
| "middle": [], |
| "last": "\u017dnidar\u0161i\u010d", |
| "suffix": "" |
| }, |
| { |
| "first": "Andra\u017e", |
| "middle": [], |
| "last": "Pelicon", |
| "suffix": "" |
| }, |
| { |
| "first": "Boshko", |
| "middle": [], |
| "last": "Koloski", |
| "suffix": "" |
| }, |
| { |
| "first": "Vid", |
| "middle": [], |
| "last": "Podpe\u010dan", |
| "suffix": "" |
| }, |
| { |
| "first": "Janez", |
| "middle": [], |
| "last": "Kranjc", |
| "suffix": "" |
| }, |
| { |
| "first": "Shane", |
| "middle": [], |
| "last": "Sheehan", |
| "suffix": "" |
| }, |
| { |
| "first": "Emanuela", |
| "middle": [], |
| "last": "Boros", |
| "suffix": "" |
| }, |
| { |
| "first": "Jose", |
| "middle": [], |
| "last": "Moreno", |
| "suffix": "" |
| }, |
| { |
| "first": "Antoine", |
| "middle": [], |
| "last": "Doucet", |
| "suffix": "" |
| }, |
| { |
| "first": "Hannu", |
| "middle": [], |
| "last": "Toivonen", |
| "suffix": "" |
| } |
| ], |
| "year": 2021, |
| "venue": "Proceedings of the EACL Hackashop on News Media Content Analysis and Automated Report Generation", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Senja Pollak, Marko Robnik \u0160ikonja, Matthew Purver, Michele Boggia, Ravi Shekhar, Marko Pranji\u0107, Salla Salmela, Ivar Krustok, Tarmo Paju, Carl-Gustav Linden, Leo Lepp\u00e4nen, Elaine Zosa, Matej Ul\u010dar, Linda Freienthal, Silver Traat, Luis Adri\u00e1n Cabrera- Diego, Matej Martinc, Nada Lavra\u010d, Bla\u017e \u0160krlj, Mar- tin \u017dnidar\u0161i\u010d, Andra\u017e Pelicon, Boshko Koloski, Vid Podpe\u010dan, Janez Kranjc, Shane Sheehan, Emanuela Boros, Jose Moreno, Antoine Doucet, and Hannu Toivonen. 2021. EMBEDDIA tools, datasets and challenges: Resources and hackathon contributions. In Proceedings of the EACL Hackashop on News Me- dia Content Analysis and Automated Report Gener- ation. Association for Computational Linguistics.", |
| "links": null |
| }, |
| "BIBREF15": { |
| "ref_id": "b15", |
| "title": "Duckdb: an embeddable analytical database", |
| "authors": [ |
| { |
| "first": "Mark", |
| "middle": [], |
| "last": "Raasveldt", |
| "suffix": "" |
| }, |
| { |
| "first": "Hannes", |
| "middle": [], |
| "last": "M\u00fchleisen", |
| "suffix": "" |
| } |
| ], |
| "year": 2019, |
| "venue": "Proceedings of the 2019 International Conference on Management of Data", |
| "volume": "", |
| "issue": "", |
| "pages": "1981--1984", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Mark Raasveldt and Hannes M\u00fchleisen. 2019. Duckdb: an embeddable analytical database. In Proceedings of the 2019 International Conference on Manage- ment of Data, pages 1981-1984.", |
| "links": null |
| }, |
| "BIBREF16": { |
| "ref_id": "b16", |
| "title": "Neural machine translation of rare words with subword units", |
| "authors": [ |
| { |
| "first": "Rico", |
| "middle": [], |
| "last": "Sennrich", |
| "suffix": "" |
| }, |
| { |
| "first": "B", |
| "middle": [], |
| "last": "Haddow", |
| "suffix": "" |
| }, |
| { |
| "first": "Alexandra", |
| "middle": [], |
| "last": "Birch", |
| "suffix": "" |
| } |
| ], |
| "year": 2016, |
| "venue": "ArXiv", |
| "volume": "", |
| "issue": "", |
| "pages": "", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Rico Sennrich, B. Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with sub- word units. ArXiv, abs/1508.07909.", |
| "links": null |
| }, |
| "BIBREF17": { |
| "ref_id": "b17", |
| "title": "Attention is all you need", |
| "authors": [ |
| { |
| "first": "Ashish", |
| "middle": [], |
| "last": "Vaswani", |
| "suffix": "" |
| }, |
| { |
| "first": "Noam", |
| "middle": [], |
| "last": "Shazeer", |
| "suffix": "" |
| }, |
| { |
| "first": "Niki", |
| "middle": [], |
| "last": "Parmar", |
| "suffix": "" |
| }, |
| { |
| "first": "Jakob", |
| "middle": [], |
| "last": "Uszkoreit", |
| "suffix": "" |
| }, |
| { |
| "first": "Llion", |
| "middle": [], |
| "last": "Jones", |
| "suffix": "" |
| }, |
| { |
| "first": "Aidan", |
| "middle": [ |
| "N" |
| ], |
| "last": "Gomez", |
| "suffix": "" |
| }, |
| { |
| "first": "\u0141ukasz", |
| "middle": [], |
| "last": "Kaiser", |
| "suffix": "" |
| }, |
| { |
| "first": "Illia", |
| "middle": [], |
| "last": "Polosukhin", |
| "suffix": "" |
| } |
| ], |
| "year": 2017, |
| "venue": "Advances in neural information processing systems", |
| "volume": "", |
| "issue": "", |
| "pages": "5998--6008", |
| "other_ids": {}, |
| "num": null, |
| "urls": [], |
| "raw_text": "Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, \u0141ukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. In Advances in neural information pro- cessing systems, pages 5998-6008.", |
| "links": null |
| } |
| }, |
| "ref_entries": { |
| "FIGREF0": { |
| "num": null, |
| "uris": null, |
| "type_str": "figure", |
| "text": "An example choropleth map showing the positive sentiment distribution over countries mentioned. Note that the purple color at the lower end of the spectrum does not indicate a high amount of negative news, but just the lack of positive news." |
| }, |
| "FIGREF1": { |
| "num": null, |
| "uris": null, |
| "type_str": "figure", |
| "text": "An example bar chart of one week of moral sentiment scores. Height indicates the strength of the sentiment component and polarity tells if it is positive or negative sentiment along that dimension." |
| }, |
| "TABREF0": { |
| "num": null, |
| "html": null, |
| "content": "<table/>", |
| "type_str": "table", |
| "text": "Pictogram with ISO 639-1 language codes summarising language coverage of multilingual techniques along with our COVID-19 news corpus. Rows are subsetted from the row above the previous rule." |
| }, |
| "TABREF2": { |
| "num": null, |
| "html": null, |
| "content": "<table><tr><td>: Pictogram with ISO 3166-1 alpha-2 country</td></tr><tr><td>codes summarising European countries covered (>20</td></tr><tr><td>items extracted) and missed in our corpus.</td></tr></table>", |
| "type_str": "table", |
| "text": "" |
| }, |
| "TABREF4": { |
| "num": null, |
| "html": null, |
| "content": "<table/>", |
| "type_str": "table", |
| "text": "Five countries with the highest absolute number of news headlines predicted as negative." |
| } |
| } |
| } |
| } |
