A Fine-grained Sentiment Analysis Study Of Online Comment Texts

With the explosive growth of the network commentary text,the commentary carries a large number of user sentiment information,analysis of the overall tendency of the comments cannot meet the needs of current users,the urgent need for more granularity of the level of sentiment analysis,and the user expressed free Resulting in the accuracy of the word segmentation is too low affect the sentiment elements of extraction and loss of implicit sentiment information and other issues need to be resolved.In this paper,we first analyze the two kinds of text preprocessing tasks of spam filter and Chinese word segmentation.Secondly,we extract the sentiment elements based on the CRFs model,and then classify and realize the extraction of the implicit sentiment objects and perform the aggregation treatment on the sentiment objects.Then we propose a method to calculate the sentiment intensity of the opposing viewpoints of the post-aggregation feature.The work of this paper mainly includes the following four aspects:(1)Aiming at the problem of text preprocessing,identifying the spam comment based on the constructed feature classification and constructing the user dictionary to improve the Chinese word segmentation.This paper firstly classifies the text based on the characteristics of the construction,including the subjective and objective text classification,filters out the comments data of the garbage view information,preserves the true and valuable comment text information,carries on the sentiment analysis task,and carries on the intention group division,facilitates the subsequent semantic emotion aggregation processing The Chinese word segmentation is based on the NLPIR word segmentation system.The user lexicon can be constructed based on the unlisted words such as the new words,the network vocabulary and the domain terms.It can correct the error of the word segmentation,improve the accuracy of the emotion object extraction,and supplement the sentiment lexicon,reduce the loss of user sentiment information and make full use of the sentiment information contained in the comments for fine-grained sentiment analysis.(2)Based on the CRFs model to extract sentiment elements,the task of emotion,emotion and modifier is transformed into a structured sequence label task.The conditional random field model is used to identify the sentiment elements.Firstly,the feature templates and annotation sets are selected.Then,the CRFs is used to identify the sentiment elements,and the product feature points are constructed from the explicit emotion characteristics-emotion words and tag sets.Naive Bayesian classifier was used to identify the implicit emotion objects in sentences.Finally,through the meaning of the code to achieve sentiment object aggregation,to improve the sparseness of the problem.(3)An algorithm for analyzing the sentiment intensity of opposing view based on contextual emotional disambiguation is proposed.This paper first defined the sentiment ambiguities words based on the dynamic polarity of the sentiment words,then uses the association rules to extract the sentiment ambiguous words collocation set,PMI pruning and filtering to construct the sentiment ambiguity lexicon of the triad form(sentiment object,sentiment word,sentiment tendency),and then we introduce the network lexicon,negative lexicon and degree adverb lexicon constructed by sentiment analysis,then put forward the algorithm of calculating the sentiment intensity of opposing viewpoints.Finally,based on sentiment intensity to generate the opposite view sentiment summary to complete the fine-grained sentiment analysis,the experiment shows the construction of this lexicon and sentiment intensity calculation method of effectiveness.(4)The prototype system of fine-grained sentiment analysis in comment is designed and implemented.Based on the theoretical basis of this paper,we can realize the fine-grained sentiment analysis system.The function modules of the system can complete the whole process of comment collection,spam filtering,Chinese word segmentation,sentiment element extraction and fine-grained sentiment analysis,and finally provide the user with intuitive view Fine-grained analysis of intensity information.