Multidimensional Topic Model For Oriented Sentiment Analysis Based On Long Short-term Memory

With the rapid development of Internet applications, people began to use large-scale of new media technology to achieve efficient information exchange. Microblog as the representative of social media to attract tens of thousands of users, people anywhere can express their true thoughts and ideas through it. By digging microblog text position, beliefs, opinions, emotions, likes or dislikes and other subjective information, emotional orientation analysis were made which has great practical value for consumers, businesses and government departments to judge.The mean research is about how to determine the entire essay emotional tendency according to microblog short text context which focuses on the classification algorithm and propose a new model to solve the problem. We verify the effectiveness of the algorithm through simulation system. The main research contents and results as follows:First, we build a model of feature recognition based on punctuations. Microblogs emotional tendencies could be judged effectively according to special sentence which patch on different criteria in the sentence layer.Second, to enhance the correlation between the contexts, the sentence increased dimensions to structure the three-dimensional long short-term memory model to reduce the impact of short-term memory on the accuracy of the model. It could increase the accuracy of feature extraction and reduce the influences on initial weight distribution to the accuracy.Third, on the basis of three-dimensional model, we constructed high dimensional long short-term memory model. Based on the word-related increased, it increase the association of phrase-to-phrase and sentence-to-sentences. It reduced the influence from special syntactical structure to judge the accuracy of sentence emotional tendency.Finally, we present a multi-dimensional topic sentiment analysis model, based on the constituent elements of microblog(word, meaning group, sentence, and topic) with hierarchical division and analysis. According to different levels of elements, we add microblog topics tab to increase the topic restrictions, select the appropriate sub-models, and give full play to the advantages of each sub-model to increase the accuracy and portability of the model.According to the above study design the experimental. The results show that the proposed model has good overall performance, and effectively improves the accuracy of Chinese microblog sentiment tendencies, meanwhile reduces the amount of training data and the complexity of matching calculation.