Study On Topical Fine-grained Sentiment Analysis Using Syntax And Semantics

There is huge social and commercial valuein the massive Internet information. The traditional coarse-grained sentiment analysis technology cannot meet the growing demand of Internet users. How to use a fine-grained sentiment analysis technologyto analyze the network information automatically, to provide more useful information for decision makers to understand the user’s consumption habits, to help users understand the more information about a product, to provide the basis for real data, is becoming a new trend.Considering researches on fine-grained sentiment analysis did not pay much attention on syntaxesand semanticx, this paperfocuses on fine-grained sentiment analysison the basis of sentiment key sentence extraction, opinion targets identification and topic-related sentiment analysis. The main research work and contributions are listed as follows:1. For sentiment key sentence extraction, an algorithm of sentiment lexicon expansion based on PMI is proposed firstly, and then a keyword lexicon construction algorithm based on topic model and graph model is put forward, and then an extraction algorithm of dependency template is also proposed. Finally, these three sentiment, keyword and dependency features are combine for sentiment key sentence extraction using lexical semantics and syntactic dependencies.2. For opinion targets identification, a new domain lexicon construction method, which combines POS template, dependency parsing, semantic role labeling and phrase structure analysis, is proposed. And after expansion by Word Embedding, the domain lexicon is embedded in sequence labellingmodel of CRF to identify opinion targets.3. For the topic-related sentiment orientation analysis, this paper first proposes a feature extraction method which combines local and global information. Then, a sentiment feature extraction approach based on Word Embedding, dependency analysis and K-means clustering is proposed. Lastly, these two features with other basic features are combined for SVM to analyze the sentiment orientation of the each topic, and finally accomplish the fine-grained sentiment analysis.