Base_JSON/prefixY/json/Y12/Y12-1030.json

{
    "paper_id": "Y12-1030",
    "header": {
        "generated_with": "S2ORC 1.0.0",
        "date_generated": "2023-01-19T13:45:37.768566Z"
    },
    "title": "Emotional Tendency Identification for Micro-blog Topics Based on Multiple Characteristics",
    "authors": [
        {
            "first": "Quanchao",
            "middle": [],
            "last": "Liu",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Beijing Institute of Technology",
                "location": {
                    "settlement": "Beijing",
                    "country": "China"
                }
            },
            "email": "liuquanchao@bit.edu.cn"
        },
        {
            "first": "Chong",
            "middle": [],
            "last": "Feng",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Beijing Institute of Technology",
                "location": {
                    "settlement": "Beijing",
                    "country": "China"
                }
            },
            "email": "fengchong@bit.edu.cn"
        },
        {
            "first": "Heyan",
            "middle": [],
            "last": "Huang",
            "suffix": "",
            "affiliation": {
                "laboratory": "",
                "institution": "Beijing Institute of Technology",
                "location": {
                    "settlement": "Beijing",
                    "country": "China"
                }
            },
            "email": ""
        }
    ],
    "year": "",
    "venue": null,
    "identifiers": {},
    "abstract": "Public opinion analysis for micro-blog post is a new trend, and wherein emotional tendency analysis on micro-blog topic is a hot spot in the sentiment analysis. According to the characteristics of contents and the various relations of Chinese micro-blog post, we construct the dictionaries of sentiment words, internet slang and emoticons respectively, and then implement the sentiment analysis algorithms based on phrase path and the multiple characteristics for emotional tendency of micro-blog topics. Using micro-blogs' forwarding, commentaries, sharing and so on, We take a future step to optimize the algorithm based on the multiple characteristics. According to the experimental results, our approach greatly improves the performance of emotional tendency identification on micro-blog topic.",
    "pdf_parse": {
        "paper_id": "Y12-1030",
        "_pdf_hash": "",
        "abstract": [
            {
                "text": "Public opinion analysis for micro-blog post is a new trend, and wherein emotional tendency analysis on micro-blog topic is a hot spot in the sentiment analysis. According to the characteristics of contents and the various relations of Chinese micro-blog post, we construct the dictionaries of sentiment words, internet slang and emoticons respectively, and then implement the sentiment analysis algorithms based on phrase path and the multiple characteristics for emotional tendency of micro-blog topics. Using micro-blogs' forwarding, commentaries, sharing and so on, We take a future step to optimize the algorithm based on the multiple characteristics. According to the experimental results, our approach greatly improves the performance of emotional tendency identification on micro-blog topic.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Abstract",
                "sec_num": null
            }
        ],
        "body_text": [
            {
                "text": "As the fast development of New media, the internet gradually advocates open architecture philosophy that users participate in actively, and has been developed from a simple \"reading webpage\" to \"writing webpage\", \"building webpage together\" for users. So a huge user generated content (UGC) has been developed, especially the emergence of micro-blog post. According to the reports 1 , users applying micro- 1 The Eleventh China Network Media Forum blog have exceeded 300,000,000. As a result of releasing diversely and writing randomly, microblog post is more and more popular in China, and it has brought a tremendous effect on network opinions and human society.",
                "cite_spans": [
                    {
                        "start": 407,
                        "end": 408,
                        "text": "1",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Micro-blog contents formed on internet express people's various emotion and sentiment, such as joy, anger, grief, praise, criticism and so on. More and more people like to use micro-blog post to share their views or experience, which makes people's opinion information expanded rapidly. So it is very difficult to rely on the artificial method to deal with the micro-blog information's collection and processing, there is an urgent need to help user analyse the massive information using computer.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "Chinese micro-blog post has multiple characteristics as following:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "1) Due to the short text message, micro-blog post has terms' sparsity. Feature extraction based on terms is not suitable for micro-blog post. 2) There exist many homophonic words, abbreviated words, internet slang in microblog post, such as \"\u676f\u5177\" standing for \"\u60b2 \u5267 \", \" \u4eb2 \" standing for \" \u4eb2 \u7231 \u7684 \", \"3Q\" standing for \"\u8c22\u8c22\", or using emoticons to express emotion and so on. 3) Many popular new words made by network events would appear in micro-blog post. Taking \"\u76ae\u978b\u5f88\u5fd9\" for example, it appeared because the news report that the materials for producing shoes are processed into edible gelatin. 4) There are a variety of relations between micro-blogs. It is very convenient to forward, comment and share micro-blog post.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "According to the above characteristics of micro-blog post, we analyse emotional tendency on micro-blog topic, and obtain a set of feasible and effective emotional tendency identification algorithm on micro-blog topic.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "The remainder of this paper is structured as follows. In section 2, we briefly summarize related work. Section 3 gives an overview of data construction, including emotional symbol library, emotional dictionary and network slang dictionary. In order to improve the rate of target coverage, the target extended algorithm is described in section 4. We design the different emotional tendency identification algorithms on micro-blog topic in sections 5 and 6 respectively. In section 7, sentiment optimization algorithm is described. Experimental results are reported in section 8 and section 9 concludes our work.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Introduction",
                "sec_num": "1"
            },
            {
                "text": "In the research domain of sentiment analysis, emotional tendency for twitter has been concerned for some time, such as Tweetfeel 1 , Twendz 2 , Twitter Sentiment 3 . In previous related work, (Go et al., 2009) use distant learning to acquire sentiment data. They use tweets ending in positive emoticons like\":)\" as positive and negative emoticons like\":(\" as negative. They build models using Naives Bayes(NB), MaxEnt(ME) and Support Vector Machines(SVM), and they report SVM outperforms other classifiers. In terms of feature space, they try a Unigram, Bigram model in conjunction with parts-of-speech (POS) features. They note that the unigram model outperforms all other models. However, the unigram model isn't suitable for Chinese microblog post, and we make full use of new emoticons which appear frequently in Chinese micro-blog post.",
                "cite_spans": [
                    {
                        "start": 192,
                        "end": 209,
                        "text": "(Go et al., 2009)",
                        "ref_id": "BIBREF9"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Another significant effort for sentiment classification on Twitter data is by (Barbosa and Feng, 2010) . They use polarity predictions from three websites as noisy labels to train a model. They propose the use of syntax features of tweets like retweet, hashtags, link, punctuation and exclamation marks in conjunction with features like prior polarity of words and POS of words. In order to improve target-dependent Twitter sentiment classification, (Long et al., 2011) incorporate target-dependent features and take the relations between twitters into consideration, such as retweet, reply and the twitters published by the same person. We extend their approach by adding a variety of Chinese dictionaries of sentiment, internet slang and emoticons, and then by using syntactic parser 4 and LIBSVM 5 respectively to achieve the sentiment analysis algorithms based on phrase path and the multiple characteristics for emotional tendency of micro-blog topic. Using micro-blogs' forwarding, commentaries and sharing, we take a future step to optimize the algorithm based on the multiple characteristics.",
                "cite_spans": [
                    {
                        "start": 78,
                        "end": 102,
                        "text": "(Barbosa and Feng, 2010)",
                        "ref_id": "BIBREF4"
                    },
                    {
                        "start": 450,
                        "end": 469,
                        "text": "(Long et al., 2011)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "The problem we address in this paper is to determine emotional orientation for micro-blog topic. So the input of our task is a collection of micro-blogs containing the topic and the output is labels assigned to each of the micro-blogs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Related Work",
                "sec_num": "2"
            },
            {
                "text": "Micro-blog post is a social networking that allows users to post real time messages, and its content is restricted to 140 Chinese characters. Our data set comes from Sina 6 and Tencent 7 micro-blog, and micro-blogs are commonly displayed on the Web as shown in figure 1. \"# #\" identifies the microblog topic, \"//\" labels user's forwarding relation, \"@\" specified the user who we speak to, and \"V\" labeling on the user shows the user's information is identified by the Sina or Tencent.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Data Description",
                "sec_num": "3"
            },
            {
                "text": "People usually use sentiment words, internet slang and emoticons to express their opinions and sentiment in micro-blog post. In order to obtain emotional orientation on micro-blog topic, we construct some dictionaries described as follows.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.1 Micro-blog post example",
                "sec_num": null
            },
            {
                "text": "According to (Du et al., 2009) , the sentiment word is one of the best emotional features representations of text, and the rich sentiment words can be conductive to improving emotional tendency identification algorithm. Internet slang that more and more people use in social network is also important factor for emotional orientation. The constructions of them are not only a significant foundation, but also a time-consuming, labor-intensive work.",
                "cite_spans": [
                    {
                        "start": 13,
                        "end": 30,
                        "text": "(Du et al., 2009)",
                        "ref_id": "BIBREF8"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Internet Slang",
                "sec_num": null
            },
            {
                "text": "In order to obtain more abundant sentiment words, we regard these sentiment words provided by HowNet 1 and National Taiwan University Sentiment Dictionary (NTUSD) 2 as the foundation, and then use lexical fusion strategy to enrich the dictionary of sentiment words. HowNet is commonsense knowledge base that describes the concepts and reveals the relation between concepts, including the relation between the attributes of concept. Since October 2007, HowNet has released \"emotional words collection for sentiment analysis\", containing a total of about 17,887 words, of which about 8,942 Chinese words. NTUSD has been summed up by National Taiwan University, including simplified Chinese version and traditional Chinese version, and each version contains 2,812 positive emotion words as well as 8,276 negative emotion words. (Turney and Littman, 2003 ; Wang et al., 2011) use lexical fusion strategy to compute the degree of correlation between test word and seed words that have more obvious emotional orientation, and then obtain emotional orientation of test word. We respectively take 20 words with obvious emotional orientation as seed words in this paper, as shown in Tables 1 and 2. (1) Where pword and nword are positive seed word and negative seed word, Pset and Nset are positive seed words collection and negative seed words collection respectively. PMI(word 1 , word 2 ) is described in formula (2), P(word 1 &word 2 ), P(word 1 ) and P(word 2 ) are probabilities of word 1 and word 2 co-occurring, word 1 appearing, and word 2 appearing in a micro-blog post respectively. When SO(word) is greater than zero, sentiment orientation of word is positive. Otherwise it is negative.",
                "cite_spans": [
                    {
                        "start": 825,
                        "end": 850,
                        "text": "(Turney and Littman, 2003",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Dictionary of Sentiment Words",
                "sec_num": "3.1.1"
            },
            {
                "text": ") ) ( ) ( ) & ( log( ) , ( 2 1 2 1 2 1 word P word P word word P word word PMI =",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Dictionary of Sentiment Words",
                "sec_num": "3.1.1"
            },
            {
                "text": "(2)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Dictionary of Sentiment Words",
                "sec_num": "3.1.1"
            },
            {
                "text": "People usually use homophonic words, abbreviated words and network slang to express their sentiment in social network, and (Agarwal et al., 2011) has analysed the sentiment of twitter data. Sometimes new words, produced by important events or news reports, are used to express their opinions. So we construct the dictionary of internet slang to support emotional tendency identification algorithm on micro-blog topic, containing homophonic words, abbreviated words, network slang and many new words. National Language Resource Monitoring & Research Center (Network Media) 3 has some internet slang, two persons from our lab manually collect more network language through social network, and then integrate these resources together. Finally we achieve this dictionary, containing 861 words with emotional orientation. Table 3 shows part of the dictionary. ",
                "cite_spans": [
                    {
                        "start": 123,
                        "end": 145,
                        "text": "(Agarwal et al., 2011)",
                        "ref_id": "BIBREF0"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 817,
                        "end": 824,
                        "text": "Table 3",
                        "ref_id": "TABREF2"
                    }
                ],
                "eq_spans": [],
                "section": "The Dictionary of Internet Slang",
                "sec_num": "3.1.2"
            },
            {
                "text": "We construct the dictionary of emoticons by combining emotional symbol library in microblog post with other statistical methods. The former is used to select obvious emotion symbols in micro-blog post, such as Sina, Tencent microblog et al. The latter choose emoticons used in other social network, containing user-generated emoticons.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The dictionary of emoticons",
                "sec_num": "3.2"
            },
            {
                "text": "Firstly, two laboratory personnel obtain emotional symbol library, and keep the emoticons with the same emotional orientation after their analysis, and then get rid of emotional symbols with ambiguous orientation, the result is described in Table 4 . Secondly, in order to enrich the dictionary of emoticons, especially user-generated emoticons in social network, two laboratory personnel collect and analyse emotional orientation, and finally obtain the result shown in Table 5 . In order to deal with the content conveniently, we pre-process all the micro-blogs and replace all the emoticons with a their sentiment polarity by looking up the dictionary of emoticons.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 241,
                        "end": 248,
                        "text": "Table 4",
                        "ref_id": "TABREF3"
                    },
                    {
                        "start": 471,
                        "end": 478,
                        "text": "Table 5",
                        "ref_id": "TABREF4"
                    }
                ],
                "eq_spans": [],
                "section": "The dictionary of emoticons",
                "sec_num": "3.2"
            },
            {
                "text": "People usually express their sentiment about object by commenting not on the object itself but on some related things of the object. (Yao et al., 2006) expresses the sentiment about automobile by commenting on the attributes or functionalities of the automobile. As shown in the micro-blog post below, user expresses a positive sentiment about Nokia 5800 by expressing a positive sentiment directly about its screen, keyboard et al.",
                "cite_spans": [
                    {
                        "start": 133,
                        "end": 151,
                        "text": "(Yao et al., 2006)",
                        "ref_id": "BIBREF15"
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extended Topics",
                "sec_num": "4"
            },
            {
                "text": "\"#\u8bfa\u57fa\u4e9a 5800#\u5c4f\u5e55\u5f88\u597d\uff0c\u952e\u76d8\u64cd\u4f5c\u4e5f\u5f88\u65b9\u4fbf\uff0c \u8d28\u91cf\u4e0d\u9519\u54e6~\u4eb2:)\"",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extended Topics",
                "sec_num": "4"
            },
            {
                "text": "It is assumed that user can clearly infer the sentiment about the topic based on those sentiments about the related things. We define those related things as Extended Topics. In order to obtain more micro-blogs' emotional orientation on the topic, we design topic's expansion algorithm, and expand the extended topic set with formula (2). Formula (2) describes the correlation between two words, the greater PMI value, the stronger the correlation. So we identify the top 10 nouns and noun phrases which have the strongest association with the topic in micro-blog set, and then take them as the original topic's extended set.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Extended Topics",
                "sec_num": "4"
            },
            {
                "text": "Chinese syntactic analysis has been considered to be an important technique in the process of Chinese information processing. It can be divided into two methods: one is the phrase structure parsing described in (Zhou and Zhao, 2007) , namely splits the sentence into phrases, and analyses hierarchical relations among phrases. The other is dependency structure analysis described in (Cheng et al., 2005) , namely parse the dependency relations between words. In terms of social network, because the content is brief, and the meaning of the expression is centered relatively, the distance between the sentiment words and evaluation objectives usually is short, and we can use the phrase structure tree to evaluate objectives' emotional orientation. Take the micro-blog post in section 4 for example, its phrase structure tree is described in figure 2, and its original topic is \"\u8bfa\u57fa\u4e9a 5800\", we obtain the topic's extended set {\u5c4f\u5e55, \u952e\u76d8, \u64cd\u4f5c, \u8d28\u91cf} using topic expansion algorithm. So we can confirm the emotional orientation of \"\u8bfa\u57fa \u4e9a 5800\" by analyzing the sentiment of the element of extended set. At first, starting from evaluation objectives, we find out the shortest phrase path between topic (extended topics) and sentiment word. Phrase path defined in (Zhao et al., 2011) is to link any two phrase nodes in the phrase structure tree. Such as \"\u8bfa\u57fa\u4e9a 5800-CD-QP-NP-IP-VP-VP-VA-\u597d\", \"\u5c4f \u5e55-NN-NP-NP-IP-VP-VP-VA-\u597d\" et al. Secondly, according to the dictionary of sentiment words, we confirm emotional orientation of the last word in the shortest phrase path. Phrase structure tree is divided into several sub tree by the punctuations, starting from each sub tree, we find out the shortest phrase path, and determine emotional orientation of the topic. Fig.2 Phrase structure tree of micro-blog example However, the negative word makes the internal sentiment words tend to shift. Such as \"\u6574\u4e2a\u5e97\u9762 \u7684\u88c5\u4fee\u4e0d\u662f\u5f88\u6f02\u4eae\", \"\u8868\u6f14\u6781\u4e0d\u81ea\u4fe1\" et al. \"\u6f02 \u4eae \" and \" \u81ea \u4fe1 \" originally are commendatory terms, but after adding \" \u4e0d \u662f \" and \" \u4e0d \", the whole sentence semantic changes as a pejorative. In this paper, we adopt matching negative rules method, sentiment words matched by the rules adopt opposite emotional orientation to properly reflect the sentiment. At first, we select negative sentences 6,639 from 20,000 micro-blogs which come from the lab 1 , and then accomplish the rule set containing high frequency negative rules 227. Secondly, the rule set is used to match the test sentence, and if sentiment words are the focus of 1 www.nlpir.org negation, we take the opposite emotional orientation.",
                "cite_spans": [
                    {
                        "start": 211,
                        "end": 232,
                        "text": "(Zhou and Zhao, 2007)",
                        "ref_id": "BIBREF16"
                    },
                    {
                        "start": 383,
                        "end": 403,
                        "text": "(Cheng et al., 2005)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 1742,
                        "end": 1747,
                        "text": "Fig.2",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Sentiment Analysis Based on Phrase Path for Micro-blog Topic",
                "sec_num": "5"
            },
            {
                "text": "The negative words used in this algorithm are acquired through HowNet. We select the concepts containing negative sememe, such as {neg|\u5426}, {impossible|\u4e0d\u4f1a}, {OwnNot|\u65e0}, {inferior|\u4e0d\u5982} et al, and obtain 21 negative words after filtering.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Sentiment Analysis Based on Phrase Path for Micro-blog Topic",
                "sec_num": "5"
            },
            {
                "text": "Based on supervised classification is our another method to determine emotional orientation on micro-blog topic. According to previous related studies (Go et al., 2009; Bermingham and Smeaton, 2010; Agrawal et al., 2003; Pang et al., 2002) , NB, ME and SVM are mainly classifiers for text classification, (Go and Barbosa et al.) have ever performed sentiment classification of twitter data, and summarize that SVM is more suitable for short text sentiment classification than other classification models. We use LIBSVM to achieve sentiment classification based on multiple characteristics for micro-blog topic. LIBSVM is an integrated software for support vector classification, and makes everything automatic-from data scaling to parameter selection. We propose the following procedure: 1) Transform data to the format of an SVM package; 2) Conduct simple scaling on the data; 3) Consider the RBF kernel ; SVM needs categorical features, so we take the features of the Chinese micro-blog post into account. At first, take the data features described in section 3 as some categorical features. According to the polarity of sentiment words, internet slang and emoticons, we compute emotion values and then take these values as some attribute values of SVM. Secondly, take the syntax features of micro-blog content as other categorical features. These syntax features are described as following: 1) Verb-Object structure. Sentiment words are verbs and the topic is their object. 2) Adjective-Center structure, namely a combination of adjective and noun. Sentiment words are adjectives and the topic is their attributive center. 3) Adverb-Center structure, namely an adverb phrase. Sentiment words are adverbs and the topic is their adverbial center. 4) Comparative structure. It is suitable for \"A than B + sentiment word\" construction. When A is the topic, its emotional orientation is consistent with the polarity of sentiment word; otherwise B is the topic, we adopt the opposite polarity of sentiment word. 5) Emotional orientation shifting. When the emotional orientation shifting is taken place in 1), 2), 3) and 4), we adopt matching negative rules method described in section 5. According to the polarity of sentiment words, we compute emotion values and then take these values as other attribute values of SVM.",
                "cite_spans": [
                    {
                        "start": 151,
                        "end": 168,
                        "text": "(Go et al., 2009;",
                        "ref_id": "BIBREF9"
                    },
                    {
                        "start": 169,
                        "end": 198,
                        "text": "Bermingham and Smeaton, 2010;",
                        "ref_id": "BIBREF5"
                    },
                    {
                        "start": 199,
                        "end": 220,
                        "text": "Agrawal et al., 2003;",
                        "ref_id": "BIBREF1"
                    },
                    {
                        "start": 221,
                        "end": 239,
                        "text": "Pang et al., 2002)",
                        "ref_id": "BIBREF11"
                    },
                    {
                        "start": 305,
                        "end": 328,
                        "text": "(Go and Barbosa et al.)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Multiple Characteristics-based Sentiment Classification for Micro-blog Topics",
                "sec_num": "6"
            },
            {
                "text": "To some extent, it is limited that only depends on data and syntax features for sentiment analysis of micro-blog topic. In order to improve the accuracy of sentiment analysis, we take relation features between micro-blogs into consideration and then take a future step to optimize sentiment classification based on multiple characteristics. Micro-blog post is propagated through their forwarding, commentaries, sharing. The forwarding, commentaries and sharing usually means the user agrees with the original user, and they have the same sentiment on the topic. At the same time, micro-blogs published by the same person within a short timeframe 1 should have a consistent sentiment about the same topic. Based on these four kinds of relation features, we can construct a graph using the input micro-blog collection of a given topic. As illustrated in Figure  3 , each node in the graph indicates a micro-blog post. The four kinds of edges indicate forwarding (solid line), commentaries (long dash line), sharing relations (dash line) and being published by the same person (round dotted line) 1 Taking one week (7 days) as the unit in our experiment respectively. The isolated nodes have not any relations with other micro-blogs.",
                "cite_spans": [
                    {
                        "start": 1094,
                        "end": 1095,
                        "text": "1",
                        "ref_id": null
                    }
                ],
                "ref_spans": [
                    {
                        "start": 852,
                        "end": 861,
                        "text": "Figure  3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Sentiment Optimization Based on Relation Features",
                "sec_num": "7"
            },
            {
                "text": "We consider that the sentiment of a micro-blog post only depends on its content and immediate neighbors, and then compute the sentiment of the current node with the following formula: is all immediate neighbors of node . The sentiment label of the micro-blog post only depending on its content is represented by",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "\u2211 = = = \u03a6 )) ( ( , ))) ( ( ( ))) ( ( | ) ( ( )) ( | ) ( ( d N d c d N p d N c d p d c d p \u03bb \u03bb \u03bb \u03bb \u03c4 \u03bb (3) Where ) (d",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "d ) (d N d )) ( | ) ( ( , d c d p d c \u03c4 \u03bb \u03c0 = = .",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "According to relaxation labeling algorithm described in (Angelova and Weikum, 2006) , we can simplify the formula (3) into (4), and use formula (4) to iteratively estimate the sentiment for all micro-blogs in the graph.",
                "cite_spans": [
                    {
                        "start": 56,
                        "end": 83,
                        "text": "(Angelova and Weikum, 2006)",
                        "ref_id": null
                    }
                ],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "1 , ) ) ) ( ) ( ( ( )) ( ( ) 1 ( ) ( ' ' , ) ( , ' > = \u2227 = \u22c5 = \u03a6 \u2211 \u220f \u2212 \u2208 r c d c d p d N r d N d d c r d c \u03bb \u03bb \u03bb \u03c0 (4)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "Where r is iterative superscript. With the shorthand notation we can rewrite this into:",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": ") ) ( ) ( ( ' ' , ' c d c d p c c = \u2227 = = \u03bb \u03bb \u03c6 \u2211 \u220f \u2212 \u2208 \u22c5 = \u03a6 )) ( ( ) 1 ( ) ( , , ) ( , ) ( ' ' d N r d N d c c d c r d c \u03bb \u03c6 \u03c0 (5)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "The original sentiment label of each node in Fig.3 is computed by the algorithm described in section 6. After the iteration ends, for any microblog in Fig.3 , the sentiment label that has the maximum is considered the final label. ",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 45,
                        "end": 50,
                        "text": "Fig.3",
                        "ref_id": null
                    },
                    {
                        "start": 151,
                        "end": 156,
                        "text": "Fig.3",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "Fig.3 Relationship graph of micro-blogs about a given topic",
                "sec_num": null
            },
            {
                "text": "According to the evaluation in section 8.2, we respectively adopt the algorithms in sections 5, 6 and 7 to determine the emotional orientation for the topics {iphone5, \u8881\u9686\u5e73, \u5317\u4eac\u7231\u60c5\u6545\u4e8b}. As shown in Table 6 , we conclude that extending topic is a vital factor, especially it has an obvious effect on sentiment analysis based on phrase path. The mainly reason is that the distance between topic word and sentiment word influences emotional judgment. Extending topic largely shortens the distance and improves the topic coverage, and promotes sentiment analysis performance.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 194,
                        "end": 201,
                        "text": "Table 6",
                        "ref_id": "TABREF6"
                    }
                ],
                "eq_spans": [],
                "section": "Results",
                "sec_num": "8.3"
            },
            {
                "text": "Because there is no annotated micro-blog post corpus publicly available for evaluation of microblog topic sentiment classification, we design topic-focused web crawler and respectively crawl 1,000 micro-blogs on three hot topics {iphone5, \u8881 \u9686 \u5e73 , \u5317 \u4eac \u7231 \u60c5 \u6545 \u4e8b }. After removing duplicate micro-blogs, we obtain 983 \"iphone5\", 993 \"\u8881\u9686\u5e73\", and 998 \"\u5317\u4eac\u7231\u60c5\u6545\u4e8b\". Two persons from our lab manually classify each micro-blog post as positive, negative or neutral towards the topics. Among the micro-blogs, 83 of them are neutral-subjective disagreement. In order to manually determine the sentiment for the next step, we assume that these micro-blogs are neutral. 103 of them are positive-negative disagreement, we adopt three people ballot to determine emotional orientation, and finally obtain 1,526 positive, 685 negative and 763 neutral microblogs. Take each 500 of them as the training corpus and the other 1,474 as the test set.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resources and Pre-processing of data",
                "sec_num": "8.1"
            },
            {
                "text": "In order to conveniently analyse the results and comparisons, the precision in Tables 7 and 9 is the proportionality of correct micro-blogs identified, containing positive, negative and neutral microblogs.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Resources and Pre-processing of data",
                "sec_num": "8.1"
            },
            {
                "text": "Using sentiment words to determine emotional orientation plays an important role for sentiment analysis. However, as for micro-blog post, the effects of internet slang and emoticons are more significant, the result is described in Table 7 . This is because internet slang is simple and flexible to write and expresses rich meanings, emoticons are easy to use and have obvious emotional orientation, they have been used pervasively in social network and welcomed by many Chinese internet users.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 231,
                        "end": 238,
                        "text": "Table 7",
                        "ref_id": "TABREF5"
                    }
                ],
                "eq_spans": [],
                "section": "Resources and Pre-processing of data",
                "sec_num": "8.1"
            },
            {
                "text": "Sentiment analysis on micro-blog topic is evaluated by Precision, Recall, F-measure and Coverage. The coverage is used to measure integrity and adequacy for the test, and help us understand the test coverage scope. ",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "\u00d7 = \u2212 (8)",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "However, in order to improve the performance of the evaluation, we take relation features between micro-blogs as an important factor. As seen in Figure 3 , there are several micro-blogs which are not connected with any other microblogs. For these micro-blogs, our sentiment optimization will have no effect. The following table 8 shows the percentages of the micro-blogs in the collection which have at least one related micro-blog post according to various relation types. Where #system_correct is the correct result from system, #system_proposed is the whole number of micro-blogs from system, #person_correct is the number of micro-blogs that has been annotated correctly by people, #weibo_topic is the number of micro-blogs containing topic words, #weibo_total is the whole number of micro-blogs in the collection. According to Table 8 , for 80.3% of the tweets concerning the topics, we can use optimization algorithm to determine their emotional orientation. That means our sentiment optimization based on relation features is potentially useful for most of the micro-blogs. The results reported in Table 9 show that every kind of relation influences on the precision of emotional orientation. (Long Jiang et al., 2011) 83.9",
                "cite_spans": [
                    {
                        "start": 1200,
                        "end": 1225,
                        "text": "(Long Jiang et al., 2011)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 145,
                        "end": 153,
                        "text": "Figure 3",
                        "ref_id": null
                    },
                    {
                        "start": 832,
                        "end": 839,
                        "text": "Table 8",
                        "ref_id": null
                    },
                    {
                        "start": 1105,
                        "end": 1112,
                        "text": "Table 9",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "As shown in Table 9 , compared to 85.6% in (Long Jiang et al., 2011) , the precision of our optimization algorithm is increased by more than one percentage points. This mainly attributed to three aspects: firstly, taking internet slang and emoticons as Micro-blog's characteristics is greatly suitable for Micro-blog's sentiment analysis. Secondly, the relations between microblogs also play an import role to determine emotional orientation, especially the sharing relation. Thirdly, to some extent the different data set can also influence the precision. We also apply the same data set into (Long Jiang et al., 2011) , and the precision is 83.9% which is still lower than our algorithm.",
                "cite_spans": [
                    {
                        "start": 43,
                        "end": 68,
                        "text": "(Long Jiang et al., 2011)",
                        "ref_id": "BIBREF10"
                    },
                    {
                        "start": 594,
                        "end": 619,
                        "text": "(Long Jiang et al., 2011)",
                        "ref_id": "BIBREF10"
                    }
                ],
                "ref_spans": [
                    {
                        "start": 12,
                        "end": 19,
                        "text": "Table 9",
                        "ref_id": "TABREF7"
                    }
                ],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "What's more, as a result of micro-blog content published with time characteristic, we can use time characteristic to analyse and forecast micro-blog topic's emotion tendency. We do experiment for the topic \"\u5317\u4eac\u7231\u60c5\u6545\u4e8b\" published within one week, as shown in Figure 4 , the horizontal axis represents time characteristic, and the vertical axis represents the numbers of micro-blogs having emotional orientation. With the emergence of new media, sentiment analysis for new media is becoming more and more important. In this paper, we make full use of multiple characteristics on micro-blog post, and design two kinds of algorithms to achieve microblog topic's sentiment analysis. In future the work will focus on the following two aspects: 1) Micro-blog topic's expansion algorithm.",
                "cite_spans": [],
                "ref_spans": [
                    {
                        "start": 254,
                        "end": 262,
                        "text": "Figure 4",
                        "ref_id": null
                    }
                ],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "2) The effect produced by the attention relation between users as well as the numbers of fans for micro-blog topic sentiment analysis.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "The relations between micro-blog users have equal importance with the multiple characteristics came from micro-blog post itself, they also play a positive effect for micro-blog's sentiment analysis.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "The Evaluation of Micro-blog Topic Sentiment Analysis",
                "sec_num": "8.2"
            },
            {
                "text": "http://www.tweetfeel.com/ 2 http://twendz.waggeneredstrom.com/ 3 http://teittersentiment.appspot.com/",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://nlp.stanford.edu/downloads/lex-parser.shtml 5 http://www.csie.ntu.edu.tw/~cjlin/libsvm 6 http://s.weibo.com/ 7 http://t.qq.com/",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            },
            {
                "text": "http://www.clr.org.cn/",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "",
                "sec_num": null
            }
        ],
        "back_matter": [
            {
                "text": "This paper is financially supported by National Natural Science Foundation of China (No. 61132009) and National Key Technology R&D Program (No. 2012BAH14F06). We would like to thank the anonymous reviewers for many valuable comments and helpful suggestions.",
                "cite_spans": [],
                "ref_spans": [],
                "eq_spans": [],
                "section": "Acknowledgments",
                "sec_num": null
            }
        ],
        "bib_entries": {
            "BIBREF0": {
                "ref_id": "b0",
                "title": "Sentiment Analysis of Twitter Data",
                "authors": [
                    {
                        "first": "Apoorv",
                        "middle": [],
                        "last": "Agarwal",
                        "suffix": ""
                    },
                    {
                        "first": "Xie",
                        "middle": [],
                        "last": "Boyi",
                        "suffix": ""
                    },
                    {
                        "first": "Ilia",
                        "middle": [],
                        "last": "Vovsha",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Proceedings of the Workshop on Language in Social Media (LSM",
                "volume": "",
                "issue": "",
                "pages": "30--38",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Apoorv Agarwal, Xie Boyi, Ilia Vovsha, et al. Sentiment Analysis of Twitter Data [C]. Proceedings of the Workshop on Language in Social Media (LSM 2011). Portland, Oregon, 2011: 30-38",
                "links": null
            },
            "BIBREF1": {
                "ref_id": "b1",
                "title": "Mining Newsgroups Using Networks Arising From Social Behavior",
                "authors": [
                    {
                        "first": "Rakesh",
                        "middle": [],
                        "last": "Agrawal",
                        "suffix": ""
                    },
                    {
                        "first": "Ramakrishnan",
                        "middle": [],
                        "last": "Sridhar Rajagopalan",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Srikant",
                        "suffix": ""
                    }
                ],
                "year": 2003,
                "venue": "Proceedings of the 12th international conference on World Wide Web",
                "volume": "",
                "issue": "",
                "pages": "529--535",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Rakesh Agrawal, Sridhar Rajagopalan, Ramakrishnan Srikant, et al. Mining Newsgroups Using Networks Arising From Social Behavior [C]. Proceedings of the 12th international conference on World Wide Web. Budapest, Hungary: WWW'03, 2003: 529-535",
                "links": null
            },
            "BIBREF2": {
                "ref_id": "b2",
                "title": "Graph-based Text Classification: Learn from Your Neighbors",
                "authors": [
                    {
                        "first": "Ralitsa",
                        "middle": [],
                        "last": "Angelova",
                        "suffix": ""
                    },
                    {
                        "first": "Gerhard",
                        "middle": [],
                        "last": "Weikum",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Ralitsa Angelova, Gerhard Weikum. Graph-based Text Classification: Learn from Your Neighbors [C].",
                "links": null
            },
            "BIBREF3": {
                "ref_id": "b3",
                "title": "Proceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval",
                "authors": [],
                "year": 2006,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "485--492",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Proceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval. Seattle, Washington, USA: SIGIR'06, 2006: 485-492",
                "links": null
            },
            "BIBREF4": {
                "ref_id": "b4",
                "title": "Robust Sentiment Detection on Twitter from Biased and Noisy Data",
                "authors": [
                    {
                        "first": "Luciano",
                        "middle": [],
                        "last": "Barbosa",
                        "suffix": ""
                    },
                    {
                        "first": "Feng",
                        "middle": [],
                        "last": "Junlan",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proceedings of COLING 2010",
                "volume": "",
                "issue": "",
                "pages": "36--44",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Luciano Barbosa and Feng Junlan. Robust Sentiment Detection on Twitter from Biased and Noisy Data. Proceedings of COLING 2010. Beijing, China, 2010: 36-44",
                "links": null
            },
            "BIBREF5": {
                "ref_id": "b5",
                "title": "Classifying Sentiment in Microblogs: Is Brevity an Advantage",
                "authors": [
                    {
                        "first": "Adam",
                        "middle": [],
                        "last": "Bermingham",
                        "suffix": ""
                    },
                    {
                        "first": "Alan",
                        "middle": [],
                        "last": "Smeaton",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Proceedings of the 19th ACM international conference on Information and knowledge management",
                "volume": "",
                "issue": "",
                "pages": "1833--1836",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Adam Bermingham, Alan Smeaton. Classifying Sentiment in Microblogs: Is Brevity an Advantage? [C]. Proceedings of the 19th ACM international conference on Information and knowledge management. Toronto, Ontario, Canada: CIKM'10, 2010: 1833-1836",
                "links": null
            },
            "BIBREF7": {
                "ref_id": "b7",
                "title": "Proceedings of the International Conference on Chinese Computing",
                "authors": [],
                "year": 2005,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "66--73",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Dependency Analyzer for Chinese [C]. Proceedings of the International Conference on Chinese Computing 2005. Singapore: COLIPS Publication, 2005: 66-73",
                "links": null
            },
            "BIBREF8": {
                "ref_id": "b8",
                "title": "A New Method to Compute Semantic Orientation",
                "authors": [
                    {
                        "first": "Du",
                        "middle": [],
                        "last": "Weifu",
                        "suffix": ""
                    },
                    {
                        "first": "Tan",
                        "middle": [],
                        "last": "Songbo",
                        "suffix": ""
                    },
                    {
                        "first": "Yun",
                        "middle": [],
                        "last": "Xiaochun",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "J]. Journal of Computer Research and Development",
                "volume": "46",
                "issue": "10",
                "pages": "1713--1720",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Du Weifu, Tan Songbo, Yun Xiaochun, et al. A New Method to Compute Semantic Orientation [J]. Journal of Computer Research and Development, 2009, 46(10): 1713-1720",
                "links": null
            },
            "BIBREF9": {
                "ref_id": "b9",
                "title": "Twitter Sentiment Classification using Distant Supervision",
                "authors": [
                    {
                        "first": "Alec",
                        "middle": [],
                        "last": "Go",
                        "suffix": ""
                    },
                    {
                        "first": "Richa",
                        "middle": [],
                        "last": "Bhayani",
                        "suffix": ""
                    },
                    {
                        "first": "Huang",
                        "middle": [],
                        "last": "Lei",
                        "suffix": ""
                    }
                ],
                "year": 2009,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Alec Go, Richa Bhayani, Huang Lei. Twitter Sentiment Classification using Distant Supervision. CS224N Project Report, Stanford, 2009",
                "links": null
            },
            "BIBREF10": {
                "ref_id": "b10",
                "title": "Target-dependent Twitter Sentiment Classification",
                "authors": [
                    {
                        "first": "Jiang",
                        "middle": [],
                        "last": "Long",
                        "suffix": ""
                    },
                    {
                        "first": "Yu",
                        "middle": [],
                        "last": "Mo",
                        "suffix": ""
                    },
                    {
                        "first": "Zhou",
                        "middle": [],
                        "last": "Ming",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics",
                "volume": "",
                "issue": "",
                "pages": "151--160",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Jiang Long, Yu Mo, Zhou Ming, et al. Target-dependent Twitter Sentiment Classification [C]. Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics. Portland, Oregon, 2011: 151-160",
                "links": null
            },
            "BIBREF11": {
                "ref_id": "b11",
                "title": "Thumbs up? Sentiment Classification using Machine Learning Techniques",
                "authors": [
                    {
                        "first": "B",
                        "middle": [],
                        "last": "Pang",
                        "suffix": ""
                    },
                    {
                        "first": "L",
                        "middle": [],
                        "last": "Lee",
                        "suffix": ""
                    },
                    {
                        "first": "S",
                        "middle": [],
                        "last": "Vaithyanathan",
                        "suffix": ""
                    }
                ],
                "year": 2002,
                "venue": "Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP)",
                "volume": "",
                "issue": "",
                "pages": "79--86",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Pang B, Lee L, Vaithyanathan S. Thumbs up? Sentiment Classification using Machine Learning Techniques [C]. Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP). Philadelphia, PA: Association for Computational Linguistics, 2002: 79-86",
                "links": null
            },
            "BIBREF12": {
                "ref_id": "b12",
                "title": "Measuring praise and criticism: inference of semantic orientation from association",
                "authors": [
                    {
                        "first": "P D",
                        "middle": [],
                        "last": "Turney",
                        "suffix": ""
                    },
                    {
                        "first": "M",
                        "middle": [
                            "L"
                        ],
                        "last": "Littman",
                        "suffix": ""
                    }
                ],
                "year": null,
                "venue": "",
                "volume": "",
                "issue": "",
                "pages": "",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Turney P D, Littman M L. Measuring praise and criticism: inference of semantic orientation from association [J].",
                "links": null
            },
            "BIBREF14": {
                "ref_id": "b14",
                "title": "A Method of Text Sentiment Classification Based on Weighted Rough Membership",
                "authors": [
                    {
                        "first": "Wang",
                        "middle": [],
                        "last": "Suge",
                        "suffix": ""
                    },
                    {
                        "first": "Li",
                        "middle": [],
                        "last": "Deyu",
                        "suffix": ""
                    },
                    {
                        "first": "Wei",
                        "middle": [],
                        "last": "Yingjie",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "J]. Journal of Computer Research and Development",
                "volume": "48",
                "issue": "5",
                "pages": "855--861",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Wang Suge, Li Deyu, Wei Yingjie. A Method of Text Sentiment Classification Based on Weighted Rough Membership [J]. Journal of Computer Research and Development, 2011, 48(5): 855-861",
                "links": null
            },
            "BIBREF15": {
                "ref_id": "b15",
                "title": "An opinion mining system for Chinese automobile reviews",
                "authors": [
                    {
                        "first": "Yao",
                        "middle": [],
                        "last": "Tianfang",
                        "suffix": ""
                    },
                    {
                        "first": "Nie",
                        "middle": [],
                        "last": "Qingyang",
                        "suffix": ""
                    },
                    {
                        "first": "Li",
                        "middle": [],
                        "last": "Jianchao",
                        "suffix": ""
                    }
                ],
                "year": 2006,
                "venue": "Proceedings of the Frontiers of Chinese Information Processing",
                "volume": "",
                "issue": "",
                "pages": "260--281",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Yao Tianfang, Nie Qingyang, Li Jianchao, et al. An opinion mining system for Chinese automobile reviews. In: Cao Youqi, Sun Maosong, eds. Proceedings of the Frontiers of Chinese Information Processing. Beijing: Tsinghua University Press, 2006: 260-281 (in Chinese with English abstract)",
                "links": null
            },
            "BIBREF16": {
                "ref_id": "b16",
                "title": "Automatic Parsing of Chinese Functional Chunks",
                "authors": [
                    {
                        "first": "Zhou",
                        "middle": [],
                        "last": "Qiang",
                        "suffix": ""
                    },
                    {
                        "first": "Zhao",
                        "middle": [],
                        "last": "Yingze",
                        "suffix": ""
                    }
                ],
                "year": 2007,
                "venue": "",
                "volume": "21",
                "issue": "",
                "pages": "18--24",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zhou Qiang, Zhao Yingze. Automatic Parsing of Chinese Functional Chunks [J]. Journal of Chinese Information Processing. 2007, 21(5): 18-24",
                "links": null
            },
            "BIBREF17": {
                "ref_id": "b17",
                "title": "Appraisal Expression Recognition Based on Syntactic Path",
                "authors": [
                    {
                        "first": "Qin",
                        "middle": [],
                        "last": "Zhao Yanyan",
                        "suffix": ""
                    },
                    {
                        "first": "Che",
                        "middle": [],
                        "last": "Bing",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Wanxiang",
                        "suffix": ""
                    }
                ],
                "year": 2011,
                "venue": "Journal of Software",
                "volume": "22",
                "issue": "5",
                "pages": "887--898",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zhao Yanyan, Qin Bing, Che Wanxiang, et al. Appraisal Expression Recognition Based on Syntactic Path [J]. Journal of Software. 2011, 22(5): 887-898",
                "links": null
            },
            "BIBREF18": {
                "ref_id": "b18",
                "title": "Sentiment Analysis",
                "authors": [
                    {
                        "first": "Qin",
                        "middle": [],
                        "last": "Zhao Yanyan",
                        "suffix": ""
                    },
                    {
                        "first": "Liu",
                        "middle": [],
                        "last": "Bing",
                        "suffix": ""
                    },
                    {
                        "first": "",
                        "middle": [],
                        "last": "Ting",
                        "suffix": ""
                    }
                ],
                "year": 2010,
                "venue": "Journal of Software",
                "volume": "21",
                "issue": "8",
                "pages": "1834--1848",
                "other_ids": {},
                "num": null,
                "urls": [],
                "raw_text": "Zhao Yanyan, Qin Bing, Liu Ting. Sentiment Analysis [J]. Journal of Software. 2010, 21(8): 1834-1848",
                "links": null
            }
        },
        "ref_entries": {
            "FIGREF4": {
                "text": "Fig.4 Emotion tendency of the topic \"\u5317\u4eac\u7231\u60c5\u6545\u4e8b\"",
                "type_str": "figure",
                "num": null,
                "uris": null
            },
            "TABREF0": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Seed words with positive emotion",
                "content": "<table><tr><td>\u8f89\u714c</td><td>\u7f8e\u5999</td><td>\u6f02\u4eae</td><td>\u4ff1\u4f73</td><td>\u52a8\u542c</td></tr><tr><td>\u4f53\u9762</td><td>\u6df3\u7f8e</td><td>\u826f\u597d</td><td>\u51fa\u8272</td><td>\u5b8c\u7f8e</td></tr><tr><td>\u7f8e\u4e3d</td><td>\u7cbe\u82f1</td><td>\u4f18\u79c0</td><td>\u9ad8\u624b</td><td>\u5148\u8fdb</td></tr><tr><td>\u5feb\u4e50</td><td>\u6700\u4f73</td><td>\u4f18\u8d28</td><td>\u5e78\u798f</td><td>\u79ef\u6781</td></tr></table>"
            },
            "TABREF1": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Seed words with negative emotion",
                "content": "<table><tr><td/><td/><td/><td colspan=\"3\">\u7f6a\u6076</td><td colspan=\"3\">\u8bc5\u5492</td><td/><td>\u8d23\u5907</td><td/><td/><td>\u4e11\u964b</td><td colspan=\"2\">\u4e11\u6076</td><td/><td/></tr><tr><td/><td/><td/><td colspan=\"3\">\u85d0\u5c0f</td><td colspan=\"3\">\u7d2f\u8d58</td><td/><td>\u9519\u8bef</td><td/><td/><td>\u5931\u8d25</td><td colspan=\"2\">\u9ebb\u70e6</td><td/><td/></tr><tr><td/><td/><td/><td colspan=\"3\">\u4e0d\u826f</td><td colspan=\"3\">\u6076\u610f</td><td/><td>\u8272\u60c5</td><td/><td/><td>\u66b4\u529b</td><td colspan=\"2\">\u8ba8\u538c</td><td/><td/></tr><tr><td/><td/><td/><td colspan=\"3\">\u9b54\u9b3c</td><td colspan=\"3\">\u91ce\u86ee</td><td/><td>\u8150\u8d25</td><td/><td/><td>\u6d41\u6c13</td><td colspan=\"2\">\u6b8b\u9177</td><td/><td/></tr><tr><td colspan=\"18\">So emotional orientation of the test word is</td></tr><tr><td colspan=\"9\">computed as follows:</td><td/><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td>SO</td><td>(</td><td>word</td><td>)</td><td>=</td><td colspan=\"2\">\u2208 Pset \u2211 pword PMI</td><td>(</td><td>word</td><td>,</td><td>pword</td><td>)</td><td>\u2212</td><td>\u2211 \u2208 Nset nword PMI</td><td>(</td><td>word</td><td>,</td><td>nword</td><td>)</td></tr></table>"
            },
            "TABREF2": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Part of the dictionary of internet slang",
                "content": "<table><tr><td>Internet slang</td><td>Meaning</td><td>Polarity</td></tr><tr><td>\u8fbe\u4eba</td><td>\u9ad8\u4eba</td><td>Positive</td></tr><tr><td>\u72c2\u9876</td><td>\u5f3a\u70c8\u652f\u6301</td><td>Positive</td></tr><tr><td>\u841d\u8389</td><td>16 \u5c81\u4ee5\u4e0b\u7684\u53ef\u7231\u5c0f\u5973\u5b69</td><td>Positive</td></tr><tr><td>\u7070\u5e38\u6851\u5fc3</td><td>\u975e\u5e38\u4f24\u5fc3</td><td>Negative</td></tr><tr><td>\u83dc\u9e1f</td><td>\u7f51\u4e0a\u4f4e\u624b</td><td>Negative</td></tr></table>"
            },
            "TABREF3": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Part of the dictionary of emoticons",
                "content": "<table><tr><td>Emoticons</td><td>Meaning</td><td>Polarity</td></tr><tr><td/><td>good</td><td>Positive</td></tr><tr><td/><td>\u7ed9\u529b</td><td>Positive</td></tr><tr><td/><td>\u9119\u89c6</td><td>Negative</td></tr><tr><td/><td>\u6012\u9a82</td><td>Negative</td></tr></table>"
            },
            "TABREF4": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Part of the dictionary of user-generated emoticons",
                "content": "<table><tr><td>Emoticons</td><td>Meaning</td><td>Polarity</td></tr><tr><td>:o)</td><td>\u5927\u7b11</td><td>Positive</td></tr><tr><td>:)</td><td>\u5fae\u7b11</td><td>Positive</td></tr><tr><td>:(</td><td>\u4f24\u5fc3</td><td>Negative</td></tr><tr><td>T_T</td><td>\u54ed\u6ce3</td><td>Negative</td></tr></table>"
            },
            "TABREF5": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "",
                "content": "<table><tr><td/><td/><td/><td/><td/><td/><td/><td/><td>The influence of internet slang and emoticons</td></tr><tr><td>Pr</td><td colspan=\"2\">ecision</td><td>=</td><td>#</td><td>system system _ #</td><td colspan=\"2\">proposed correct _</td><td>(6)</td><td>for emotional orientation Characteristics Precision(%)</td></tr><tr><td colspan=\"2\">Re F</td><td colspan=\"5\">correct correct ecision _ _ ecision Pr person system measure call # = # + Pr 2</td><td>call call Re Re \u00d7</td><td>(7)</td><td>Content characteristics -Sentiment words -Internet slang -Emoticons</td><td>83.9 81.7 75.3 71.3</td></tr></table>"
            },
            "TABREF6": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "The comparisons of algorithms",
                "content": "<table><tr><td>Extending topic</td><td>Sentiment analysis</td><td colspan=\"9\">Precision(%) Positive Negative Neutral Positive Negative Neutral Positive Negative Neutral Recall(%) F-measure(%)</td><td>Coverage (%)</td></tr><tr><td/><td>Based on phrase path</td><td>80.5</td><td>53.2</td><td>23.9</td><td>43.9</td><td>31.3</td><td>73.4</td><td>56.8</td><td>39.4</td><td>36.1</td><td/></tr><tr><td>Before extending</td><td>Multiple characteristics-based</td><td>87.4</td><td>62.5</td><td>54.7</td><td>80.9</td><td>70.3</td><td>65.8</td><td>84.0</td><td>66.2</td><td>59.7</td><td>68.7</td></tr><tr><td/><td>Based on relation features</td><td>90.8</td><td>59.1</td><td>73.0</td><td>86.7</td><td>70.3</td><td>76.0</td><td>88.7</td><td>64.2</td><td>74.5</td><td/></tr><tr><td/><td>Based on phrase path</td><td>81.8</td><td>55.0</td><td>43.9</td><td>68.5</td><td>58.9</td><td>69.6</td><td>74.6</td><td>56.9</td><td>53.8</td><td/></tr><tr><td>After extending</td><td>Multiple characteristics-based</td><td>92.9</td><td>69.0</td><td>69.2</td><td>83.3</td><td>78.4</td><td>90.5</td><td>87.8</td><td>73.4</td><td>78.4</td><td>82.6</td></tr><tr><td/><td>Based on relation features</td><td>93.6</td><td>60.9</td><td>82.9</td><td>91.1</td><td>75.7</td><td>77.2</td><td>92.3</td><td>67.5</td><td>79.9</td><td/></tr><tr><td colspan=\"5\">Table 8 The proportion of micro-blogs having at least</td><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td colspan=\"3\">one related micro-blog in the collection</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>Relations features</td><td/><td>Proportion(%)</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>Forwarding</td><td/><td>39.8</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>Commentaries</td><td/><td>32.7</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>Sharing</td><td/><td>22.3</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td colspan=\"2\">Published by the same person</td><td/><td>25.9</td><td/><td/><td/><td/><td/><td/><td/><td/></tr><tr><td/><td>All</td><td/><td>80.3</td><td/><td/><td/><td/><td/><td/><td/><td/></tr></table>"
            },
            "TABREF7": {
                "type_str": "table",
                "html": null,
                "num": null,
                "text": "Comparison of contributions made by every",
                "content": "<table><tr><td>kind of relations</td><td/></tr><tr><td>Relation features</td><td>Precision(%)</td></tr><tr><td>Forwarding</td><td>67.3</td></tr><tr><td>Commentaries</td><td>61.5</td></tr><tr><td>Sharing</td><td>78.8</td></tr><tr><td>Published by the same person</td><td>70.9</td></tr><tr><td>All</td><td>86.7</td></tr></table>"
            }
        }
    }
}