Research On Sentiment Analysis For Web Texts Based On Semantic Relatedness

In the Web2.0 era,people can take the initiative to express and disseminate their opinions,attitudes,feelings and emotions on the network,generating a massive web text with subjective color.It is not unrealistic to analyze such large-scale network texts by artificial means.Therefore,this situation promotes the emergence and development of the technology for Sentiment Analysis.From now on,the technology for Sentiment Analysis has been successfully applied to public opinion analysis,product marketing,the stock price forecasting and so on.It has important theoretical value and practical value.As the web text has the characteristics of wide range of subjects,nonstandard format,short length,logical confusion and so on,has brought many difficulties and challenges to sentiment analysis.In view of this,we take network comment texts as research object,discussing the concept of semantic relatedness deeply,introducing topic model and word vector model into the sentiment analysis for web texts in order to effectively improve accuracy of sentiment classification.The specific contents are as follows:(1)The differences and relations between the concept of "Similarity" and the concept of "Relatedness" are introduced and summarized,and pointing out that the concept of "Relatedness" covers the concept of "Similarity".Then we clarify the definition of the word correlation,summarize the methods of calculating the wor correlation and described the main idea of the topic model and the word vector model as well as their excellent performance in digging the semantic association behind the words terms.(2)Sentiment classification for web texts based on the topic model,mainly carrying out the following studies:(1)study the context between the development and the current word with the rules of the emotion unit constructed in order to extract from the text of the web text sentiment classification helpful emotional information;(2)extracting key features form the emotional information based on topic model in order to build the vector space model,and the apply the machine learning classifier on the sentiment classification for web texts.The experimental results show the effectiveness of the method,and its performance on dimension reduction is better than the others.(3)The research on sentiment classification for Micoblog texts,mainly containing two parts.The first part is the mothed of automatic annotation for thesentiment orientation of Microblog smileys.The mothed primarily screen seed words with the artificial means and statistical methods,and then we design algorithm based on the seed words in order to recognize the sentiment orientation of the Microblog smileys.The second part is sentiment classification for microblog text,including:(1)extract the characteristics with the emotion dictionary,and then definte three types of features,including seed features,similar features and surplus features;(2)propose a algorithm based word2 vec to merge the features into feature sets in order to construct the text vector,and then we apply the machine learning classifier on sentiment classification for Microblog texts.The experimental results show the effectiveness of the mothed of automatic annotation for the sentiment orientation of Microblog smileys and the the mothed of sentiment analysis for Microblog texts base on word2 vec,and second method is good at dimension reduction.(4)We design and implement the experimental system based on our methods to explore the semtiment analysis.This system includes four modules,which are the module for data preprocessing,the module for constructing the sentiment lexicon,the module for sentiment classification and the module for exprimental reports.