Multi-classification Sentiment Analysis Of Short Text Based On Deep Learning

With the rapid development of the Internet,social networks have become an important platform for people to express their ideas.Text sentiment analysis of social network data has become a hotspot in the field of information processing.Through the sentiment analysis of a text,we can capture the user's emotional inclination and obtain the user's requirements.Traditional sentiment analysis methods based on machine learning and sentimental dictionary need to select features.Since the selection of sentimental features is often subjective,it will lead to over-fitting classification model and insufficient generalization ability.Deep learning constructs a deep emotion model by effectively extracting the emotional features of text,which improves the accuracy of text sentiment analysis based on positive and negative factors.However,due to the lack of multi-category emotional corpus,the emotional inclination contained in the text cannot be accurately expressed.Therefore,this study is based on deep learning to study the multi-emotional classification of short micro-blog texts.The main work of this paper are as follows:For multi-category emotional corpus acquisition,an iterative corpus acquisition method based on emotional seed words was proposed.First,we extracted the emotional word set from the existing corpus,and expanded the emotional word set by using synonym forest.Then we calculated the TF-IDF weights of each affective word,and constructed the emotional seed word set by selecting the emotional words with higher weights.Finally,we used a variety of sub-word search strategies to collect emotional texts of micro-blog,and extended their automatic tags to the corresponding categories of micro-blog corpus,then completed corpus acquisition through multiple iterations.Experiments showed that this method could effectively solve the problem of insufficient and unbalanced samples of emotional corpus.In the construction of emotional classification model based on deep learning,we analyzed the weaknesses of existing CNN and LSTM networks in text sentiment analysis and proposed a multi-class affective analysis model based on Attention mechanism.The model used Attention mechanism to fuse the local features extracted from CNN network with the word order features extracted from LSTM model,and adopted the idea of integrated model at the classification level.Then the emotional features extracted from CNN network and LSTM network were joined together as the final emotional features extracted from the model.The comparative experiments,reveal that the accuracy of the model had been significantly improved.