Multilabel Sentiment Identification Using Modifier Structure

• Main Authors: Zhong Peng, 鍾鵬
• Other Authors: K. Robert Lai
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/hzzc26

碩士 === 元智大學 === 資訊工程學系 === 106 === In modern life, many people are suffering from mental illness problems. With the development of Internet services, there have been some professional online medical service platform. Choose to talk through the network through the psychological problems, in order to obtain the solution provided by professionals, has become a convenient and scruples the user an effective way of privacy. Internet medical service platform has accumulated a large number of social psychology text, if you can effectively tap this valuable resource, will give academic research to a new breakthrough direction, through the use of the results of these studies, the Internet service platform can also provide users with A better experience. Mental disease text is a medical text, usually a description of a condition will be summarized as multiple emotional tags, are multi-label text. It is a research direction of emotional analysis task in natural language processing. This paper discusses the problem of multi-label text emotion recognition in the field of mental illness, and then discusses the influence of modified structure on multi-label classification. In the study of emotional analysis, the study of emotional vocabulary has always been the focus, which is the important feature of the classification of text emotions. This paper analyzes the influence of the text containing the modified structure on the multi - label emotion recognition by analyzing the feature of the text by using the dictionary to construct the modified structure dictionary. This article considers three kinds of modifiers, negative words, degree adverbs and functional words, which are composed of different types of modifiers and emotional films. Negative modification structure can be used to reverse the semantic polarity of emotion words in the structure. The degree of modification structure can enhance or weaken the positive or negative degree of emotional words. The function structure can weaken the emotional intensity. This paper first discusses the feasibility of using the depth learning model to classify text in the natural language field, and proposes a classification model of several depth neural networks based on sentence level. Secondly, the classification model was used to carry out the multi - label classification experiment of mental illness text. Then, the structure of the modified structure dictionary, with the dictionary to extract the text containing the modified structure. Finally, the classification text of the feature text is classified and predicted, and the influence of the modified structure on the accuracy of the text emotion recognition is evaluated.