The Design And Implementation Of Short Text Emotion Analysis System Based On Deep Learning

With the popularization of Internet technology,network life has been assimilated into people’s daily life.Many social platforms and e-commerce platforms are rising rapidly,and a growing number of people express their opinions and views on the Internet.These contents include the discussion of social hot issues,the evaluation of specific products and so on.Extracting valuable information from massive short text big data and classifying the short text published by users can not only better and faster recommend the required information for users,but also help the network supervision department supervise the network public opinion and stop the spread of harmful information on the network in time.In the emotion analysis task of short text,a major factor leading to the low accuracy of emotion analysis is that the characteristics of short text are relatively sparse,which is caused by the characteristics of short text its own lack of content.Therefore,short text is difficult to provide more emotional semantic information.To solve this problem,this thesis proposes a short text expansion algorithm based on emotional features and keyword association to alleviate the problem of sparsity.The algorithm improves the traditional TF-IDF and combines information gain to realize the quantitative calculation of emotional relevance,and proposes a word association network for text expansion.The algorithm first preprocesses the text,then calculates the quantitative value of emotional relevance of words in the text,filters out the word vectors with high relevance to form the keyword set,and then expands the keyword set by using the synonym library.Then the keyword set and word association network are used to associate to generate the candidate association word set.Finally,the emotional association degree is calculated to select the candidate word set with high correlation degree and add it to the original text data set to expand the text features.At the same time,in view of the shortcomings of many existing neural network models in the granularity control of semantic information extraction,and it is difficult to construct a deeper neural network structure.In this thesis,a multi-scale convolution and bidirectional independent recurrent neural network model based on attention is proposed.Taking advantage of the local feature extraction of convolution neural network,the model realizes the local feature extraction of different granularity through multiple scale convolution kernels,introduces the independent cyclic neural network which solves the gradient hierarchical decline problem existing in the traditional network,constructs a deeper neural network,and obtains richer global semantic feature information through two-way extraction,Finally,the emotion contribution is weighted by attention to improve the effect of text emotion classification.Through experiments,the proposed deep learning neural network model has improved the accuracy rate on the same micro-blog reviews dataset compared with other models.The accuracy rate,recall rate and F1 value of the model also have been improved.Finally,this thesis integrates the model into the system,develops a short text emotion analysis system based on deep learning,and realizes the application landing.