社交网络评论中的反语识别研究

发布时间：2018-04-17 04:33

本文选题：反语识别 + 文本分类　；参考：《云南财经大学》2017年硕士论文

【摘要】：随着信息技术的蓬勃发展,网络中的文档成为主要的信息载体,是人们生活中主要信息来源。随着互联网Web2.0时代的到来,人们从被动接受门户网站的信息,转为主动在社交媒体上发布、共享、传播信息。由于用户参与信息产生过程中,网络信息的内容形式也变得多种多样。这些关于观点性的内容对社会舆情分析、电子商务等方面都有着重要的意义和实用价值。针对这种观点性文本的情感分类逐渐演变为一个热点问题。在网络文本中经常出现反语,他们利用反语来表达主观的以及深层次的观点,反语的使用将大大增加情感分析的难度。为了提高情感分类的准确度,需要探讨反语识别,因此本文研究的是反语识别。本文主要从两个方面研究反语的识别:一是基于规则的反语识别方法;二是基于机器学习反语识别方法。对于基于规则的反语识别方法,本文中提出了两种反语识别的规则——歇后语规则、违反常识规则,这也是本文的创新之处。对违反常识规则部分引入了矛盾关系检测,通过矛盾关系中的反义词规则和否定词规则,来判断文本是否是违反了常识规则。满足上述两种规则中的其中一种的文本则判定为反语,对于不满足规则的文本,使用机器学习方法训练分类器,识别出反语。训练分类器过程中,构建特征体系——英文词、特定的语气词、文本中特定的词汇、网络词汇、谐音字、连续的标点符号,将分词后并删除停用词后的所有不重复词也纳入特征体系。计算只使用机器学习方法进行反语识别的性能,同时也计算规则结合机器学习方法进行反语识别的性能,并比较这两种方法的性能。本文发现本文中提出的规则结合机器学习方法对反语识别相对只使用机器学习方法是更有效的。
[Abstract]:With the rapid development of information technology, network documents become the main carrier of information, is the main source of information in people's life. With the advent of the Internet Web2.0 era, people accept the information portal from passive to active in social media release, sharing and dissemination of information. Because the user participation in the information network process. The content of the information form has become diverse. These views about the content analysis of social public opinion, e-commerce and other aspects are of great significance and practical value. In this view the text sentiment classification has gradually evolved into a hot issue. Irony often appear in the network text, they use the language to express subjective and deep view of irony use will greatly increase the difficulty of sentiment analysis. In order to improve the accuracy of sentiment classification, to explore irony recognition, because This paper is the study of irony recognition. This article mainly from the two aspects of irony recognition: one is the irony recognition method based on rules; two is the irony recognition method based on machine learning. The irony recognition method based on rules, this paper puts forward two kinds of irony recognition rules of Xiehouyu rules, in violation of common sense rules. This is also the innovation of this paper. Any violation of common sense rules introduced by the contradiction between detection, the contradiction between the opposite rule and negation rules, to determine whether the text is in violation of common sense rules. Meet the above two kinds of rules in which a text is judged for irony, does not meet the rules of the text. The use of machine learning methods to train the classifier, recognize the irony. The process of training classifier, feature construction system -- English specific words, modal words, vocabulary specific text in network Vocabulary, homophonic words, punctuation, continuous, word and delete the stop words after all words are included in the feature system. Calculation only using machine learning methods for performance of irony recognition, but also the calculation rules combined with machine learning method for recognition of irony, and compare the performance of these two methods. We find that the proposed rule learning method with the machine of irony recognition relative only use the machine learning method is more effective.

【学位授予单位】：云南财经大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.1

【参考文献】