Research On Sentiment Classification Algorithm Based On Topic Fusion

With the explosive development of the Internet and intelligent terminals,huge social network groups and organizations have produced a large amount of network information.How to capture and analyze the emotional trend and development of the public in the unstructured network information is a very important research topic.Chinese text data has different forms of expression and has many characteristics such as grammatical specificity,semantic pluralism,implicit expression and so on,relative to English text data.In addition,most of the current Chinese text sentiment classification methods are relatively shallow learning methods and there are some defects such as relying on manual extraction of sample,limited the ability of text expression and so on.As the increasing of the amount of data with abundant representation methods,it is more difficult to get higher Chinese text sentiment classification accuracy.Therefore,combining the characteristics of Chinese text,to further improve the accuracy and performance of Chinese text sentiment classification is an urgent problem to be studied and solved in the field of public opinion classification.Combining the characteristics of Chinese text,the thesis carries out the research on Chinese text sentiment classification algorithm,and improves and integrates it on the existing algorithm.The main research contents are as follows:For the sentiment classification model of Chinese long text,the thesis proposes a feature fusion sentiment classification model based on convolutional neural network and the bidirectional gate recurrent unit,to solve the defect and limitation of text sentiment classification only using word vector as text feature in traditional deep learning algorithm.And using topic2 vec to obtain the topic and word vectors represented by Chinese text in the same vector space as the sentiment classification model input.Finally,verifying the validity of the model through the experimental comparison and analysis.For the sentiment classification algorithm of Chinese short text,for the lack of word vector information in the traditional sentiment classification algorithm,the topic information obtained by the P_BTM model is used to extend the word vector information to obtain more Chinese short text features,wherein the P_BTM model is improved by the BTM model,and the word vector of Chinese short text is obtained by the improved TF-IDF algorithm.Finally,verifying the effectiveness of the algorithm through the experimental comparison and analysis.