| With the rapid development of network and the rise of social media,the amount of text data generated by users is increasing.The texts generated by Internet users involve a wide range and are becoming shorter and shorter,with less information,faster update speed.Therefore,how to mine the hidden semantic information and judge emotional tendencies from these texts is a important topic in the field of natural language procession and text retrieval at present.We can use the topic model to mine the potential topic information in texts,which can effectively solve the above problems.Text feature extraction is an important task in text sentiment analysis,which has a great impact on the effect of sentiment analysis.Besides,the traditional text feature extraction algorithm is mainly aimed at long texts such as news and novels.If it is directly applied to short text scenarios,it will not achieve the desired effect.So an efficient short text sentiment classification method is urgently needed.Based on the topic sentiment model,this thhesis takes Chinese microblog as the research object,and conducts research on microblog text sentiment analysis and new word extraction.Firstly,in view of the existing emotional topic model ignoring the emotional issues of documents and words,the emotional modeling of documents and words is carried out at the same time.We also added the user’s personality characteristics and emoji characteristics into the model.Based on this,the Joint Sentiment and Topic Model based on User personality and Time(UPT-JST)is proposed.Extensive experiments show that the UPT-JST performs better than the traditional models.Secondly,in view of UPT-JST’s lack of attention to lexical semantic relations,the deep learning method is applied to the topic model,which is called as PUPTJST(Promotional UPT-JST),and the word embedding is adopted to calculate the correlation between words.We defined the probability distribution function and added the sentiment dictionary and emoji dictionary as the prior knowledge into UPT-JST.Experiments show that the PUPT-JST can effectively deal with the sentiment analysis problem of microblog text.Finally,a Chinese new word discovery algorithm is proposed,which is based on the combination of improved word embedding and rule screening.The effect of the new Chinese word detection in the microblog field is verified by experiments. |
