The Sentiment Orientation Analysis Of Texts Based On Network Comments

With the rapid development of computer technology and the Internet, network comme nts have become an important part of the Internet and they often contain a lot of of valuable information. How to extract valuable information from these massive comm ents is increasingly becoming a concerned topic.Then we design and implement a network comment sentiment orientation analysis system, which mainly consists of five parts: data preprocessing module, the evaluation unit recognition module, text sentiment orientation formula module, feature selection module based on semantics and machine learning module. At the same time, we focus on how to use sentiment features to construct valid sentiment orientation analysis algorithms and have made the following findings:(1) In the sentiment orientation analysis based on semantics, the sentiment dictionary is very important. And to some extent, the semantic orientation of texts can be affected by negative words and extent qualifier words. Therefore, we design some sentiment orientation calculation algorithms, such as the combination of sentiment dictionary and negative words, the combination of sentiment dictionary, negative words and extent qualifier words.(2) In the sentiment orientation analysis based on machine learning, the feature selection is particularly important. Evaluation units, which are the key-value pairs of evaluation objects and evaluation words, have strong sentiment properties. Then a feature selection algorithm combining evaluation units and word frequency is designed and can be used in Naive Bayes classifier, polynomial Naive Bayes classifier and support vector machine classifier.Finally, according to the above findings, our proposed algorithms, such as sentiment orientation analysis formula based on semantics and the feature selection algorithm are tested in the network comment sentiment orientation analysis system. The experimental results show these proposed algorithms are feasible and effective.