1,068
个编辑
更改
跳到导航
跳到搜索
第1行：
第1行： − + − 文本情感分析（也称为意见挖掘）是指用自然语言处理、文本挖掘以及计算机语言学等方法来识别、提取、量化和研究原素材中的情感状态和主观信息。情感分析被广泛应用于来源于用户的素材，如评论和调查回复，在线和社交媒体；也被应用于来源于卫生保健的素材，其应用范围涵盖了从市场营销到客户服务到临床医学的各个方面。+ − − − The objective and challenges of sentiment analysis can be shown through some simple examples. − + − * Coronet has the best lines of all day cruisers. − * Bertram has a deep V hull and runs easily through seas. − * Pastel-colored 1980s day cruisers from Florida are ugly. − * I dislike old [[cabin cruiser]]s. 第21行：
第14行： − + − − * I do not dislike cabin cruisers. ([[Negation]] handling) − * Disliking watercraft is not really my thing. (Negation, inverted [[word order]]) − * Sometimes I really hate [[Rigid-hulled inflatable boat|RIBs]]. ([[Adverbial]] modifies the sentiment) − * I'd really truly love going out in this weather! (Possibly [[sarcastic]]) − * Chris Craft is better looking than Limestone. (Two [[brand name]]s, identifying the target of attitude is difficult). − * Chris Craft is better looking than Limestone, but Limestone projects seaworthiness and reliability. (Two attitudes, two brand names). − * The movie is surprising with plenty of unsettling plot twists. (Negative term used in a positive sense in certain domains). − * You should see their decadent dessert menu. (Attitudinal term has shifted polarity recently in certain domains) − * I love my mobile but would not recommend it to any of my colleagues. (Qualified positive sentiment, difficult to categorise) 第43行：
第26行： − + − A basic task in sentiment analysis is classifying the ''polarity'' of a given text at the document, sentence, or feature/aspect level—whether the expressed opinion in a document, a sentence or an entity feature/aspect is positive, negative, or neutral. Advanced, "beyond polarity" sentiment classification looks, for instance, at emotional states such as enjoyment, anger, disgust, sadness, fear, and surprise. <ref name=":2"> Vong Anh Ho, Duong Huynh-Cong Nguyen, Danh Hoang Nguyen, Linh Thi-Van Pham, Duc-Vu Nguyen, Kiet Van Nguyen, Ngan Luu-Thuy Nguyen. "Emotion Recognition+ + − 情感分析的最底层的任务是识别给定的情感评论文本中的极性倾向是正面的、负面的还是中性的。按照处理文本的粒度不同，情感分析可以分为篇章级、句子级和词语级三个研究层次。高级的“超极性”情感分类研究关注有如情绪状态等，如享受、愤怒、厌恶、悲伤、恐惧和惊讶。<ref name=":2" /> + − Precursors to sentimental analysis include the General Inquirer,<ref name=":3">Stone, Philip J., Dexter C. Dunphy, and Marshall S. Smith. "The general inquirer: A computer approach to content analysis." MIT Press, Cambridge, MA (1966).</ref> which provided hints toward quantifying patterns in text and, separately, psychological research that examined a person's [[psychological state]] based on analysis of their verbal behavior.<ref name=":4">Gottschalk, Louis August, and Goldine C. Gleser. The measurement of psychological states through the content analysis of verbal behavior. Univ of California Press, 1969.</ref> − 情感分析的先驱包括 ''the General Inquirer''，''<ref name=":3" />'' 这为文本和心理学研究中的量化模式提供了线索，即根据对一个人的语言行为的分析来研究其心理状态。<ref name=":4" />+ − − − Subsequently, the method described in a patent by Volcani and Fogel,<ref name=":5">{{cite patent 第72行：
第52行： − + − 随后，Volcani和Fogel<ref name=":5" /> 在一项专利中描述的方法专门研究了情感并根据不同的情感尺度识别了文本中的单个单词和短语。一个基于他们的研究建立的称为EffectCheck的系统则提供了同义词，可以用来增加或减少在每个尺度的诱发情绪的水平。 − + − Many other subsequent efforts were less sophisticated, using a mere polar view of sentiment, from positive to negative, such as work by Turney,<ref name="Turney02" /> and Pang<ref name="PangAl02"> 第88行：
第66行： − + 第98行：
第76行： − + 第108行：
第86行： − + − − 之后许多的研究都没有那么复杂，仅仅使用了正负面的情感极性视角，比如Turney<ref name="Turney02" />和Pang<ref name="PangAl02" />分别使用了不同的方法来识别产品评论和电影评论的极性。这项工作是在篇章级的粒度层次进行的。人们还可以在多层次上对篇章的极性进行分类，Pang<ref name="PangLee05" />和Snyder<ref name="SnyderBarzilay07" /> 等人曾尝试这样做：Pang和Lee<ref name="PangLee05" />拓展了仅仅将电影评论分为正面或负面的基本任务，并以三星或四星的尺度预测电影的评级；而Snyder<ref name="SnyderBarzilay07" /> 对餐馆评论进行了深入分析，预测特定餐馆的各个方面的评级，例如食物和氛围(以五星的尺度)。 − − First steps to bringing together various approaches—learning, lexical, knowledge-based, etc.—were taken in the 2004 [[AAAI]] Spring Symposium where linguists, computer scientists, and other interested researchers first aligned interests and proposed shared tasks and benchmark data sets for the systematic computational research on affect, appeal, subjectivity, and sentiment in text.<ref name=":6">Qu, Yan, James Shanahan, and Janyce Wiebe. "Exploring attitude and affect in text: Theories and applications." In AAAI Spring Symposium) Technical report SS-04-07. AAAI Press, Menlo Park, CA. 2004.</ref> − + − Even though in most statistical classification methods, the neutral class is ignored under the assumption that neutral texts lie near the boundary of the binary classifier, several researchers suggest that, as in every polarity problem, three categories must be identified.Moreover, it can be proven that specific classifiers such as the [[Maximum entropy probability distribution|Max Entropy]]<ref name="Vryniotis13">+ 第125行：
第99行： − + 第135行：
第109行： − + − − 尽管在大多数统计分类方法中，根据中性文本位于二元分类器边界附近的假设，中性类常常忽略了，但一些研究者建议在每个极性问题中必须确定三个类别。此外，研究也证明引入中立类可以提高某些分类器的整体准确率，如最大熵（Max Entropy）<ref name="Vryniotis13" /> 和支持向量机（SVMs）<ref name="KoppelSchler06" /> 等特定分类器。原则上由两种方法可以进行中性分类。一是，算法首先识别出中性分类后将其过滤，再根据正面和负面的情感二分类对其他内容进行评估。二是，一步构建包含中性、正面和负面三种类别的分类。<ref>{{Cite journal|last1=Ribeiro|first1=Filipe Nunes|last2=Araujo|first2=Matheus|date=2010|title=A Benchmark Comparison of State-of-the-Practice Sentiment Analysis Methods|url=https://www.researchgate.net/publication/286302059|journal=Transactions on Embedded Computing Systems |volume=9 |issue=4}}</ref> 第二种方法通常会涉及到估计所有类别的概率分布（比如[[Nltk|NLTK]]实现的naive Bayes分类器）。是否以及如何使用中性分类取决于数据的性质：如果数据被清晰地分类为中性、正面和负面的语言，那么过滤掉中性语言并关注正面和负面情感的极性是有意义的。相比之下，如果数据大部分是中性的，对正面和负面影响的偏差很小，这种策略就会使其更难明确区分两极。 − A different method for determining sentiment is the use of a scaling system whereby words commonly associated with having a negative, neutral, or positive sentiment with them are given an associated number on a −10 to +10 scale (most negative up to most positive) or simply from 0 to a positive upper limit such as +4. This makes it possible to adjust the sentiment of a given term relative to its environment (usually on the level of the sentence). When a piece of unstructured text is analyzed using [[natural language processing]], each concept in the specified environment is given a score based on the way sentiment words relate to the concept and its associated score.<ref name=":7">{{Cite journal|last1=Taboada|first1=Maite|last2=Brooke|first2=Julian|date=2011|title=Lexicon-based methods for sentiment analysis|url=http://dl.acm.org/citation.cfm?id=2000518|journal=Computational Linguistics |volume=37 |issue=2 |pages=272–274|doi=10.1162/coli_a_00049|citeseerx=10.1.1.188.5517|s2cid=3181362}}</ref><ref name=":8">{{Cite journal|last1=Augustyniak|first1=Łukasz|last2=Szymański|first2=Piotr|last3=Kajdanowicz|first3=Tomasz|last4=Tuligłowicz|first4=Włodzimierz|date=2015-12-25|title=Comprehensive Study on Lexicon-based Ensemble Classification Sentiment Analysis|journal=Entropy|language=en|volume=18|issue=1|pages=4|doi=10.3390/e18010004|bibcode=2015Entrp..18....4A|doi-access=free}}</ref><ref name=":9">{{Cite journal|last1=Mehmood|first1=Yasir|last2=Balakrishnan|first2=Vimala|date=2020-01-01|title=An enhanced lexicon-based approach for sentiment analysis: a case study on illegal immigration|url=https://doi.org/10.1108/OIR-10-2018-0295|journal=Online Information Review|volume=44|issue=5|pages=1097–1117|doi=10.1108/OIR-10-2018-0295|issn=1468-4527}}</ref>This allows movement to a more sophisticated understanding of sentiment, because it is now possible to adjust the sentiment value of a concept relative to modifications that may surround it. Words, for example, that intensify, relax or negate the sentiment expressed by the concept can affect its score. Alternatively, texts can be given a positive and negative sentiment strength score if the goal is to determine the sentiment in a text rather than the overall polarity and strength of the text.<ref name="SentiStrength2010">+ 第164行：
第136行： − 另一种不同的识别情感的方法是使用一个量表系统，在这个系统中负面、中性和正面相关的词语被赋予了-10到+10的取值，代表着从最负面到最正面，或者是简单地从0到正面的上限，如+4。这使得我们能够根据环境（通常是在句子语境的层次上）调整特定语言的情感极性程度。当使用自然语言处理对一段非结构化文本进行分析时，基于情感词与概念的关联方式及其相关分数，对指定环境中的每个概念进行评分。<ref name=":7" /><ref name=":8" /><ref name=":9" /> 。这使得人们可以对情感有更深入的理解，因为现在依据相周围可能发生的变化调整一个概念的情感程度，例如，强化、缓和或否定概念所表达的情感的词语会影响它的得分。或者，如果目的是确定文本中的情感而不是文本的整体极性和强度，则可以给文本一个正面和负面的情感强度得分。<ref name="SentiStrength2010" />+ + − There are various other types of sentiment analysis like- Aspect Based sentiment analysis, Grading sentiment analysis (positive,negative,neutral), Multilingual sentiment analysis and detection of emotions.+ − − 还有各种其他类型的情感分析，如功能/属性为基础的情感分析、分级情感分析(正面、负面、中性) 、多语言情感分析和情感识别。 − − === Subjectivity/objectivity identification 主观性/客观性识别 === − This task is commonly defined as classifying a given text (usually a sentence) into one of two classes: objective or subjective.<ref name="PangLee08Subjectivity">{{cite book 第183行：
第151行： − + 第199行：
第167行： − + 第207行：
第175行： − + 第216行：
第184行： − + − 这一任务被普遍地定义为将给定的文本识别为主观和客观两个类别。<ref name="PangLee08Subjectivity" /> 这个问题有时甚至比极性分类更加难以解决。<ref name="MihalceaAl07" /> 词或短语的主观性取决于特定的上下文语境，客观的篇章有时候又包含了主观的句子（比如，一篇新闻中引用了其他人的观点）。此外，正如Su<ref name="SuMarkert08" /> 所提到的，结果在很大程度上依赖于注释文本时使用的主观性的定义。然而，Pang<ref name="PangLee04" /> 的研究表明，在对篇章文本进行极性分类之前去掉文本中的客观句子有助于提高模型的表现。+ − − The term objective refers to the incident carry factual information.<ref name="Wiebe 2005 486–497">{{Cite journal|last1=Wiebe|first1=Janyce|last2=Riloff|first2=Ellen|date=2005|editor-last=Gelbukh|editor-first=Alexander|title=Creating Subjective and Objective Sentence Classifiers from Unannotated Texts|url=https://link.springer.com/chapter/10.1007%2F978-3-540-30586-6_53|journal=Computational Linguistics and Intelligent Text Processing|series=Lecture Notes in Computer Science|volume=3406|language=en|location=Berlin, Heidelberg|publisher=Springer|pages=486–497|doi=10.1007/978-3-540-30586-6_53|isbn=978-3-540-30586-6}}</ref> − − 客观指的是具有事实信息的事件。<ref name="Wiebe 2005 486–497" /> − − * Example of an objective sentence: 'To be elected president of the United States, a candidate must be at least thirty-five years of age.' − The term subjective describes the incident contains non-factual information in various forms, such as personal opinions, judgment, and predictions. Also known as 'private states' mentioned by Quirk et al.<ref name=":10">{{Cite book|last1=Quirk|first1=Randolph|title=A Comprehensive Grammar of the English Language (General Grammar)|last2=Greenbaum|first2=Sidney|last3=Geoffrey|first3=Leech|last4=Jan|first4=Svartvik|publisher=[[Longman]]|year=1985|isbn=1933108312|pages=175–239}}</ref> In the example down below, it reflects a private states 'We Americans'. Moreover, the target entity commented by the opinions can take several forms from tangible product to intangible topic matters stated in Liu(2010).<ref name="Liu2010" /> Furthermore, three types of attitudes were observed by Liu(2010), 1) positive opinions, 2) neutral opinions, and 3)negative opinions.<ref name="Liu2010" />+ − 主观这个术语描述的事件包含各种形式的非事实信息，如个人意见、判断和预测。也被Quirk等人称为“私人状况（private states）”。<ref name=":10" /> 在下面的例子中，它反映了“我们美国人”这样一个私人状态。此外，被评论的目标实体可以是从有形到无形的话题事项等多种形式（Liu,2010）。<ref name="Liu2010" /> 此外，刘（2010）还观察到三种类型的态度: 1)正面的观点，2)中性的观点，3)负面的观点。<ref name="Liu2010" /> − * Example of a subjective sentence: 'We Americans need to elect a president who is mature and who is able to make wise decisions.' − This analysis is a classification problem.<ref name=":1" /> − Each class's collections of words or phrase indicators are defined for to locate desirable patterns on unannotated text. For subjective expression, a different word list has been created. Lists of subjective indicators in words or phrases have been developed by multiple researchers in the linguist and natural language processing field states in Riloff et al.(2003).<ref name=":11">{{Cite journal|last1=Riloff|first1=Ellen|last2=Wiebe|first2=Janyce|date=2003-07-11|title=Learning extraction patterns for subjective expressions|journal=Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing|series=EMNLP '03|volume=10|location=USA|publisher=Association for Computational Linguistics|pages=105–112|doi=10.3115/1119355.1119369|s2cid=6541910|doi-access=free}}</ref> A dictionary of extraction rules has to be created for measuring given expressions. Over the years, in subjective detection, the features extraction progression from curating features by hands in 1999 to automated features learning in 2005.<ref name=":12">{{Cite journal|last1=Chaturvedi|first1=Iti|last2=Cambria|first2=Erik|last3=Welsch|first3=Roy E.|last4=Herrera|first4=Francisco|date=November 2018|title=Distinguishing between facts and opinions for sentiment analysis: Survey and challenges|url=https://sentic.net/subjectivity-detection.pdf|journal=Information Fusion|volume=44|pages=65–77|doi=10.1016/j.inffus.2017.12.006|via=Elsevier Science Direct|doi-access=free}}</ref> At the moment, automated learning methods can further separate into supervised and [[Unsupervised learning|unsupervised machine learning]]. Patterns extraction with machine learning process annotated and unannotated text have been explored extensively by academic researchers.+ − − 每个类别的单词或短语指标集合都是为了在未注释的文本上找到理想的模式而定义的。对于主观表达，已经建立了一个不同的单词列表。Riloff等人(2003)指出，语言学家和自然语言处理领域的多位研究人员已经开发出了单词或短语的主观指标列表。<ref name=":11" /> 必须为测量给定的表达方式创建一个提取规则的字典是非常必要的。多年来，在主观性识别方面，从1999年的手工特征提取发展到了2005年的自动特征学习。<ref name=":12" />目前，自动学习方法可以进一步分为监督学习和无监督学习。利用机器学习对文本进行注释和去注释的模式提取方法已经成为学术界研究的热点。 − − However, researchers recognized several challenges in developing fixed sets of rules for expressions respectably. Much of the challenges in rule development stems from the nature of textual information. Six challenges have been recognized by several researchers: 1) metaphorical expressions, 2) discrepancies in writings, 3) context-sensitive, 4) represented words with fewer usages, 5) time-sensitive, and 6) ever-growing volume.
无编辑摘要
'''Sentiment analysis''' (also known as '''opinion mining''' or '''emotion AI''') is the use of [[natural language processing]], [[Text analytics|text analysis]], [[computational linguistics]], and [[biometrics]] to systematically identify, extract, quantify, and study affective states and subjective information. Sentiment analysis is widely applied to [[voice of the customer]] materials such as reviews and survey responses, online and social media, and healthcare materials for applications that range from [[marketing]] to [[Customer relationship management|customer service]] to clinical medicine.
'''文本情感分析'''（也称为'''意见挖掘'''）是指用自然语言处理、文本挖掘以及计算机语言学等方法来识别、提取、量化和研究原素材中的情感状态和主观信息。情感分析被广泛应用于来源于用户的素材，如评论和调查回复，在线和社交媒体；也被应用于来源于卫生保健的素材，其应用范围涵盖了从市场营销到客户服务到临床医学的各个方面。
== 案例 ==
== Examples 案例 ==
情感分析的目的和挑战可以通过一些简单的例子来说明。
情感分析的目的和挑战可以通过一些简单的例子来说明。
=== Simple cases 简单例子 ===
=== 简单例子 ===
* Coronet 拥有全天式游艇中最好的航线。
* Coronet 拥有全天式游艇中最好的航线。
* Bertram有一个深v型的船身，可以轻松通过海洋。
* Bertram有一个深v型的船身，可以轻松通过海洋。
* 我不喜欢旧的游艇。
* 我不喜欢旧的游艇。
=== More challenging examples更具挑战性的例子 ===
=== 更具挑战性的例子 ===
* 我不是不喜欢游艇（I do not dislike cabin cruisers）。(否定处理)
* 我不是不喜欢游艇（I do not dislike cabin cruisers）。(否定处理)
* 我喜欢自己的手机，但不会向任何同事推荐（I love my mobile but would not recommend it to any of my colleagues）。(有保留的积极情绪，很难归类）
* 我喜欢自己的手机，但不会向任何同事推荐（I love my mobile but would not recommend it to any of my colleagues）。(有保留的积极情绪，很难归类）
== Types类型 ==
== 类型 ==
情感分析的最底层的任务是识别给定的情感评论文本中的极性倾向是正面的、负面的还是中性的。按照处理文本的粒度不同，情感分析可以分为篇章级、句子级和词语级三个研究层次。高级的“超极性”情感分类研究关注有如情绪状态等，如享受、愤怒、厌恶、悲伤、恐惧和惊讶。<ref name=":2"> Vong Anh Ho, Duong Huynh-Cong Nguyen, Danh Hoang Nguyen, Linh Thi-Van Pham, Duc-Vu Nguyen, Kiet Van Nguyen, Ngan Luu-Thuy Nguyen. "Emotion Recognition
for Vietnamese Social Media Text". In Proceedings of the 2019 International Conference of the Pacific Association for Computational Linguistics (PACLING 2019), Hanoi, Vietnam (2019).</ref>
for Vietnamese Social Media Text". In Proceedings of the 2019 International Conference of the Pacific Association for Computational Linguistics (PACLING 2019), Hanoi, Vietnam (2019).</ref>
情感分析的先驱包括 ''the General Inquirer''，''<ref name=":3">Stone, Philip J., Dexter C. Dunphy, and Marshall S. Smith. "The general inquirer: A computer approach to content analysis." MIT Press, Cambridge, MA (1966).</ref>'' 这为文本和心理学研究中的量化模式提供了线索，即根据对一个人的语言行为的分析来研究其心理状态。<ref name=":4">Gottschalk, Louis August, and Goldine C. Gleser. The measurement of psychological states through the content analysis of verbal behavior. Univ of California Press, 1969.</ref>
随后，Volcani和Fogel<ref name=":5">{{cite patent
| country = USA
| country = USA
| number = 7,136,877
| number = 7,136,877
| class =
| class =
| url = http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=%2Fnetahtml%2FPTO%2Fsearch-adv.htm&r=2&p=1&f=G&l=50&d=PTXT&S1=(fogel.INNM.+AND+volcani.INNM.)&OS=in/fogel+and+in/volcani&RS=(IN/fogel+AND+IN/volcani)
| url = http://patft.uspto.gov/netacgi/nph-Parser?Sect1=PTO2&Sect2=HITOFF&u=%2Fnetahtml%2FPTO%2Fsearch-adv.htm&r=2&p=1&f=G&l=50&d=PTXT&S1=(fogel.INNM.+AND+volcani.INNM.)&OS=in/fogel+and+in/volcani&RS=(IN/fogel+AND+IN/volcani)
}}</ref> looked specifically at sentiment and identified individual words and phrases in text with respect to different emotional scales. A current system based on their work, called EffectCheck, presents synonyms that can be used to increase or decrease the level of evoked emotion in each scale.
}}</ref> 在一项专利中描述的方法专门研究了情感并根据不同的情感尺度识别了文本中的单个单词和短语。一个基于他们的研究建立的称为EffectCheck的系统则提供了同义词，可以用来增加或减少在每个尺度的诱发情绪的水平。
之后许多的研究都没有那么复杂，仅仅使用了正负面的情感极性视角，比如Turney<ref name="Turney02" />和Pang<ref name="PangAl02">
{{cite conference
{{cite conference
| first1 = Bo | last1 = Pang
| first1 = Bo | last1 = Pang
| url = http://www.cs.cornell.edu/home/llee/papers/sentiment.home.html
| url = http://www.cs.cornell.edu/home/llee/papers/sentiment.home.html
}}
}}
</ref> who applied different methods for detecting the polarity of [[product review]]s and movie reviews respectively. This work is at the document level. One can also classify a document's polarity on a multi-way scale, which was attempted by Pang<ref name="PangLee05">
</ref> 分别使用了不同的方法来识别产品评论和电影评论的极性。这项工作是在篇章级的粒度层次进行的。人们还可以在多层次上对篇章的极性进行分类，Pang<ref name="PangLee05">
{{cite conference
{{cite conference
| first1 = Bo | last1 = Pang
| first1 = Bo | last1 = Pang
| url = http://www.cs.cornell.edu/home/llee/papers/pang-lee-stars.home.html
| url = http://www.cs.cornell.edu/home/llee/papers/pang-lee-stars.home.html
}}
}}
</ref> and Snyder<ref name="SnyderBarzilay07">
</ref> 和Snyder<ref name="SnyderBarzilay07">
{{cite conference
{{cite conference
| first1 = Benjamin | last1 = Snyder
| first1 = Benjamin | last1 = Snyder
| url = http://people.csail.mit.edu/regina/my_papers/ggranker.ps
| url = http://people.csail.mit.edu/regina/my_papers/ggranker.ps
}}
}}
</ref> among others: Pang and Lee<ref name="PangLee05" /> expanded the basic task of classifying a movie review as either positive or negative to predict star ratings on either a 3- or a 4-star scale, while Snyder<ref name="SnyderBarzilay07" /> performed an in-depth analysis of restaurant reviews, predicting ratings for various aspects of the given restaurant, such as the food and atmosphere (on a five-star scale).
</ref>等人曾尝试这样做：Pang和Lee<ref name="PangLee05" />拓展了仅仅将电影评论分为正面或负面的基本任务，并以三星或四星的尺度预测电影的评级；而Snyder<ref name="SnyderBarzilay07" /> 对餐馆评论进行了深入分析，预测特定餐馆的各个方面的评级，例如食物和氛围(以五星的尺度)。
在2004年AAAI春季研讨会上，语言学家、计算机科学家和其他感兴趣的研究人员首次将各种方法——学习、词汇、基于知识等——结合起来，提出了共享任务和基准数据集，以便对文本中的情感、吸引力、主观性和情感进行系统的计算研究。<ref name=":6" />
在2004年AAAI春季研讨会上，语言学家、计算机科学家和其他感兴趣的研究人员首次将各种方法——学习、词汇、基于知识等——结合起来，提出了共享任务和基准数据集，以便对文本中的情感、吸引力、主观性和情感进行系统的计算研究。<ref name=":6">Qu, Yan, James Shanahan, and Janyce Wiebe. "Exploring attitude and affect in text: Theories and applications." In AAAI Spring Symposium) Technical report SS-04-07. AAAI Press, Menlo Park, CA. 2004.</ref>
尽管在大多数统计分类方法中，根据中性文本位于二元分类器边界附近的假设，中性类常常忽略了，但一些研究者建议在每个极性问题中必须确定三个类别。此外，研究也证明引入中立类可以提高某些分类器的整体准确率，如最大熵（Max Entropy）<ref name="Vryniotis13">
{{cite conference
{{cite conference
| first = Vasilis | last = Vryniotis
| first = Vasilis | last = Vryniotis
| url = http://blog.datumbox.com/the-importance-of-neutral-class-in-sentiment-analysis/
| url = http://blog.datumbox.com/the-importance-of-neutral-class-in-sentiment-analysis/
}}
}}
</ref> and [[Support vector machine|SVMs]]<ref name="KoppelSchler06">
</ref> 和支持向量机（SVMs）<ref name="KoppelSchler06">
{{cite conference
{{cite conference
| first1 = Moshe | last1 = Koppel
| first1 = Moshe | last1 = Koppel
| citeseerx = 10.1.1.84.9735
| citeseerx = 10.1.1.84.9735
}}
}}
</ref> can benefit from the introduction of a neutral class and improve the overall accuracy of the classification. There are in principle two ways for operating with a neutral class. Either, the algorithm proceeds by first identifying the neutral language, filtering it out and then assessing the rest in terms of positive and negative sentiments, or it builds a three-way classification in one step.This second approach often involves estimating a probability distribution over all categories (e.g. [[Naive Bayes classifier|naive Bayes]] classifiers as implemented by the [[Nltk|NLTK]]).Whether and how to use a neutral class depends on the nature of the data: if the data is clearly clustered into neutral, negative and positive language, it makes sense to filter the neutral language out and focus on the polarity between positive and negative sentiments. If, in contrast, the data are mostly neutral with small deviations towards positive and negative affect, this strategy would make it harder to clearly distinguish between the two poles.
</ref> 等特定分类器。原则上由两种方法可以进行中性分类。一是，算法首先识别出中性分类后将其过滤，再根据正面和负面的情感二分类对其他内容进行评估。二是，一步构建包含中性、正面和负面三种类别的分类。<ref>{{Cite journal|last1=Ribeiro|first1=Filipe Nunes|last2=Araujo|first2=Matheus|date=2010|title=A Benchmark Comparison of State-of-the-Practice Sentiment Analysis Methods|url=https://www.researchgate.net/publication/286302059|journal=Transactions on Embedded Computing Systems |volume=9 |issue=4}}</ref> 第二种方法通常会涉及到估计所有类别的概率分布（比如[[Nltk|NLTK]]实现的naive Bayes分类器）。是否以及如何使用中性分类取决于数据的性质：如果数据被清晰地分类为中性、正面和负面的语言，那么过滤掉中性语言并关注正面和负面情感的极性是有意义的。相比之下，如果数据大部分是中性的，对正面和负面影响的偏差很小，这种策略就会使其更难明确区分两极。
另一种不同的识别情感的方法是使用一个量表系统，在这个系统中负面、中性和正面相关的词语被赋予了-10到+10的取值，代表着从最负面到最正面，或者是简单地从0到正面的上限，如+4。这使得我们能够根据环境（通常是在句子语境的层次上）调整特定语言的情感极性程度。当使用自然语言处理对一段非结构化文本进行分析时，基于情感词与概念的关联方式及其相关分数，对指定环境中的每个概念进行评分。<ref name=":7">{{Cite journal|last1=Taboada|first1=Maite|last2=Brooke|first2=Julian|date=2011|title=Lexicon-based methods for sentiment analysis|url=http://dl.acm.org/citation.cfm?id=2000518|journal=Computational Linguistics |volume=37 |issue=2 |pages=272–274|doi=10.1162/coli_a_00049|citeseerx=10.1.1.188.5517|s2cid=3181362}}</ref><ref name=":8">{{Cite journal|last1=Augustyniak|first1=Łukasz|last2=Szymański|first2=Piotr|last3=Kajdanowicz|first3=Tomasz|last4=Tuligłowicz|first4=Włodzimierz|date=2015-12-25|title=Comprehensive Study on Lexicon-based Ensemble Classification Sentiment Analysis|journal=Entropy|language=en|volume=18|issue=1|pages=4|doi=10.3390/e18010004|bibcode=2015Entrp..18....4A|doi-access=free}}</ref><ref name=":9">{{Cite journal|last1=Mehmood|first1=Yasir|last2=Balakrishnan|first2=Vimala|date=2020-01-01|title=An enhanced lexicon-based approach for sentiment analysis: a case study on illegal immigration|url=https://doi.org/10.1108/OIR-10-2018-0295|journal=Online Information Review|volume=44|issue=5|pages=1097–1117|doi=10.1108/OIR-10-2018-0295|issn=1468-4527}}</ref>。这使得人们可以对情感有更深入的理解，因为现在依据相周围可能发生的变化调整一个概念的情感程度，例如，强化、缓和或否定概念所表达的情感的词语会影响它的得分。或者，如果目的是确定文本中的情感而不是文本的整体极性和强度，则可以给文本一个正面和负面的情感强度得分。<ref name="SentiStrength2010">
{{cite journal
{{cite journal
| first1 = Mike
| first1 = Mike
</ref>
</ref>
还有各种其他类型的情感分析，如功能/属性为基础的情感分析、分级情感分析(正面、负面、中性) 、多语言情感分析和情感识别。
===主观性/客观性识别 ===
这一任务被普遍地定义为将给定的文本识别为主观和客观两个类别。<ref name="PangLee08Subjectivity">{{cite book
| first1 = Bo
| first1 = Bo
| last1 = Pang
| last1 = Pang
| chapter-url = http://www.cs.cornell.edu/home/llee/opinion-mining-sentiment-analysis-survey.html
| chapter-url = http://www.cs.cornell.edu/home/llee/opinion-mining-sentiment-analysis-survey.html
}}
}}
</ref> This problem can sometimes be more difficult than polarity classification.<ref name="MihalceaAl07">{{cite conference
</ref>这个问题有时甚至比极性分类更加难以解决。<ref name="MihalceaAl07">{{cite conference
|first1 = Rada
|first1 = Rada
|last1 = Mihalcea
|last1 = Mihalcea
|archive-date = 2010-07-08
|archive-date = 2010-07-08
}}
}}
</ref> The subjectivity of words and phrases may depend on their context and an objective document may contain subjective sentences (e.g., a news article quoting people's opinions). Moreover, as mentioned by Su,<ref name="SuMarkert08">{{cite conference
</ref>词或短语的主观性取决于特定的上下文语境，客观的篇章有时候又包含了主观的句子（比如，一篇新闻中引用了其他人的观点）。此外，正如Su<ref name="SuMarkert08">{{cite conference
| first1 = Fangzhong | last1 = Su
| first1 = Fangzhong | last1 = Su
| first2 = Katja | last2 = Markert
| first2 = Katja | last2 = Markert
| url = http://www.comp.leeds.ac.uk/markert/Papers/Coling2008.pdf
| url = http://www.comp.leeds.ac.uk/markert/Papers/Coling2008.pdf
}}
}}
</ref> results are largely dependent on the definition of subjectivity used when annotating texts. However, Pang<ref name="PangLee04">{{cite conference
</ref>所提到的，结果在很大程度上依赖于注释文本时使用的主观性的定义。然而，Pang<ref name="PangLee04">{{cite conference
| first1 = Bo | last1 = Pang
| first1 = Bo | last1 = Pang
| first2 = Lillian | last2 = Lee
| first2 = Lillian | last2 = Lee
| url = http://www.cs.cornell.edu/home/llee/papers/cutsent.home.html
| url = http://www.cs.cornell.edu/home/llee/papers/cutsent.home.html
}}
}}
</ref> showed that removing objective sentences from a document before classifying its polarity helped improve performance.
</ref>的研究表明，在对篇章文本进行极性分类之前去掉文本中的客观句子有助于提高模型的表现。
客观指的是具有事实信息的事件。<ref name="Wiebe 2005 486–497">{{Cite journal|last1=Wiebe|first1=Janyce|last2=Riloff|first2=Ellen|date=2005|editor-last=Gelbukh|editor-first=Alexander|title=Creating Subjective and Objective Sentence Classifiers from Unannotated Texts|url=https://link.springer.com/chapter/10.1007%2F978-3-540-30586-6_53|journal=Computational Linguistics and Intelligent Text Processing|series=Lecture Notes in Computer Science|volume=3406|language=en|location=Berlin, Heidelberg|publisher=Springer|pages=486–497|doi=10.1007/978-3-540-30586-6_53|isbn=978-3-540-30586-6}}</ref>
* 客观句的例子：“要当选美国总统，候选人必须年满35岁。”
* 客观句的例子：“要当选美国总统，候选人必须年满35岁。”
主观这个术语描述的事件包含各种形式的非事实信息，如个人意见、判断和预测。也被Quirk等人称为“私人状况（private states）”。<ref name=":10">{{Cite book|last1=Quirk|first1=Randolph|title=A Comprehensive Grammar of the English Language (General Grammar)|last2=Greenbaum|first2=Sidney|last3=Geoffrey|first3=Leech|last4=Jan|first4=Svartvik|publisher=[[Longman]]|year=1985|isbn=1933108312|pages=175–239}}</ref>在下面的例子中，它反映了“我们美国人”这样一个私人状态。此外，被评论的目标实体可以是从有形到无形的话题事项等多种形式（Liu,2010）。<ref name="Liu2010" /> 此外，刘（2010）还观察到三种类型的态度: 1)正面的观点，2)中性的观点，3)负面的观点。<ref name="Liu2010" />
* 主观句的例子：“我们美国人需要选出一位成熟且能够做出明智决定的总统。”
* 主观句的例子：“我们美国人需要选出一位成熟且能够做出明智决定的总统。”
这种分析是一个分类的问题。<ref name=":1" />
这种分析是一个分类的问题。<ref name=":1" />
每个类别的单词或短语指标集合都是为了在未注释的文本上找到理想的模式而定义的。对于主观表达，已经建立了一个不同的单词列表。Riloff等人(2003)指出，语言学家和自然语言处理领域的多位研究人员已经开发出了单词或短语的主观指标列表。<ref name=":11">{{Cite journal|last1=Riloff|first1=Ellen|last2=Wiebe|first2=Janyce|date=2003-07-11|title=Learning extraction patterns for subjective expressions|journal=Proceedings of the 2003 Conference on Empirical Methods in Natural Language Processing|series=EMNLP '03|volume=10|location=USA|publisher=Association for Computational Linguistics|pages=105–112|doi=10.3115/1119355.1119369|s2cid=6541910|doi-access=free}}</ref>必须为测量给定的表达方式创建一个提取规则的字典是非常必要的。多年来，在主观性识别方面，从1999年的手工特征提取发展到了2005年的自动特征学习。<ref name=":12">{{Cite journal|last1=Chaturvedi|first1=Iti|last2=Cambria|first2=Erik|last3=Welsch|first3=Roy E.|last4=Herrera|first4=Francisco|date=November 2018|title=Distinguishing between facts and opinions for sentiment analysis: Survey and challenges|url=https://sentic.net/subjectivity-detection.pdf|journal=Information Fusion|volume=44|pages=65–77|doi=10.1016/j.inffus.2017.12.006|via=Elsevier Science Direct|doi-access=free}}</ref>目前，自动学习方法可以进一步分为监督学习和无监督学习。利用机器学习对文本进行注释和去注释的模式提取方法已经成为学术界研究的热点。
然而，研究人员认识到在为表达方式分类制定一套固定的规则集方面存在一些挑战。规则开发中的大部分挑战源于文本信息的性质。一些研究人员已经认识到了六个挑战： 1)隐喻性的表达，2)写作中的差异，3)上下文敏感性，4)时间敏感性，5)代表性词用法较少以及6)不断增长的数量。
然而，研究人员认识到在为表达方式分类制定一套固定的规则集方面存在一些挑战。规则开发中的大部分挑战源于文本信息的性质。一些研究人员已经认识到了六个挑战： 1)隐喻性的表达，2)写作中的差异，3)上下文敏感性，4)时间敏感性，5)代表性词用法较少以及6)不断增长的数量。