Research And Implementation Of Chinese Explanatory Opinion Relationship Recognition Method Based On Machine Learning

In recent years, with the rapid development of various forms of user networks, blog, product reviews and news editorials and other forms of user-generated content(User-Generated Content, UGC) have came into being. People visit a large number of comments will have to spend a lot of energy, and therefore, how to quickly and effectively help users find the information attracted widespread attention and research, Chinese explanatory opinion relations recognition is one of the problem.Chinese explanatory opinion elements recognition refers to the massive use of NLP techniques to analyze data, identify attributes evaluations and explanatory phrases, and complete the identification of explanatory opinion elements' relations on the basis of entity recognition. Based on the characteristics of Chinese language product reviews, discusses the relationship between attributes evaluations and explanatory comments sentence and use word embedding and dependency relationship feature to identify the relationship between explanatory opinion elements. Specifically, the paper will study from the following three aspects:(1) Explanatory opinion elements recognition based on machine learning: the explanatory opinion elements can be divided into attribute, evaluation and explanatory phrase. The identification of attribute and evaluation will be completed in the framework of CRFs with word, part of speech, keyword tag and relationship mark characteristics. The explanatory phrase will be identified by the method of sentence embedding in the framework of SVMs. Experimental results show that keyword tag and relationship mark can effectively improve the identification result of attribute and evaluation and show that sentence embedding have some effect in identifying explanatory opinion.(2) Explanatory opinion relations recognition based on dependency relationship: The dependency relationship method, mainly using distance information, dependency relationship information and position information as SVMs features. Experimental results show that dependency method has a good effect in the relationship identification of attribute-evaluation, has some effect in the relationships identification of attribute-attribute and opinion-explain.(3) Explanatory opinion relations recognition based on word embedding: The word embedding method, mainly using word embedding as the feature of a classifier to complete the experiment, use Similarity degree and rule to improve the result. Experimental results show that word embedding method has better effect than dependency method.