Research On Text Emotional Multi-label Classification Based On Deep Learning

The rapid development of the Internet has injected new vitality into e-commerce platforms and social platforms.The number of interactive websites has increased rapidly.People are more and more inclined to express their opinions and opinions through social media.As a result,a large amount of text data including people's sentiments and emotions has been accumulated on the Internet.How to effectively use natural language processing technology to analyze and mine such data has become an hot problem to monitor public opinion,analyze consumer behavior and serve e-commerce.For text sentiment analysis,the traditional supervised learning method based on single label has been difficult to meet the needs of various kinds of text processing.It is of great significance to study the classification method of multi-label for sentiment and emotion in text,and to use multi-label method reasonably to process various emotional text data.In recent years,in-depth learning has achieved good results in the field of natural language processing.In-depth learning distributes data representation and learns deep-seated abstract features of data by constructing complex networks,avoiding tedious feature construction work.Therefore,this paper mainly studies in-depth learning and multi-label classification technology,using the method of in-depth learning to carry out multi-label analysis of text emotions.The specific research contents and conclusions are as follows:(1)Text emotion multi-label classification problem analysis.Through the analysis of text emotion data,this paper determines the classification of text emotion in a fine-grained way.(2)Emotional multi-label classification based on tag features.When the text corresponds to multiple emotional tags,each tag contains different information of the text.The amount of information in the tag is relatively large.Rational use of tag features will be more conducive to the expression of the text,thus improving the performance of text multi-label classification.This paper presents a multi-label classification method for emotional text based on convolution neural network model fusing label features.This method uses convolution neural network to extract features of text and its corresponding emotional tags,and finally classifies multi-tags of text by fusing emotional text features and tag features as the overall representation of text.On NLPCC2014 Chinese micro-blog sentiment analysis data set,experiments results show that the average accuracy rate is 0.6227,which shows that the CNN model using tag features can improve the performance of micro-blog sentiment classification.(3)Hierarchical attention of LSTM text multi-label classification.In order to use the sentence structure and hierarchical information to express the text in a deep level,and at the same time to use the relevance information between labels,this paper presents a multi-label classification method of text emotion based on the hierarchical structure and attention model of the recurrent neural network.The method is divided into two processes,encoding and decoding.The encoding process represents the emotional text layer by layer from the word to the sentence and then to the text,and adds the attention of the word to the sentence and the attention of the sentence to the text,and represents the emotional text from bottom to top.The recurrent neural network is used to represent the text with multiple labels.The experimental results show that the method can effectively utilize the correlation between labels and improve the classification performance and the average accuracy rate reached 0.6086.