Research On Fine-grained Short Text Emotional Analysis Based On Mixed Word Vector Method

Emotional analysis of short text is one of the hot research directions in natural language processing.Using deep learning to capture emotional information hidden in text is a popular short text emotional analysis method at present,but it relies on data sets and has low accuracy in dealing with fine-grained emotional analysis tasks.Emotional vocabulary and rule-based methods are often used in long text analysis.This method is introduced into short text analysis task,which is called hybrid word vector method for short text analysis.It can extract some text features by using emotional thesaurus,improve the description ability of word vector to text,and improve the accuracy of fine-grained short text emotional analysis.The emotional element vectors are constructed by the emotional categories of words in the vocabulary,the syntactic features of the text and the expression symbols contained in the text.The mixed word vectors are synthesized together with the word vectors and input into the embedding layer of the in-depth learning network.At the same time,BiLSTM network with Attention mechanism is used to construct deep learning network.The effectiveness and efficiency of the hybrid word vector method are evaluated using the popular short text dataset,Chinese Weibo.The granularity of affective analysis is divided into four categories: aversion,anger,sadness and happiness.Experiments show that the hybrid word vector method can significantly improve the accuracy of fine-grained short text affective analysis.It achieves 69.63% accuracy and improves 7.12% accuracy in data set.The cost is only to increase the code volume by hundreds of lines,and to increase a small amount of text data processing time and module training time.