Sentiment Analysis Of Weibo Comments Based On Emotional Word Vector Optimization

With the rapid development of social platforms such as micro-blog,the analysis technology based on micro-blog corpus has attracted much attention from academia.Among them,sentiment analysis technology has become a hot research field.In sentiment analysis tasks,word vectors based on corpus context information can not distinguish words with the same context but different semantics,which will lead to "words with opposite semantics and high similarity of word vectors",thus reducing the accuracy of emotional analysis.In order to solve the above problems,this paper focuses on the optimization of the emotional word vector,and proposes an optimization model of the emotional word vector combined with the emotion dictionary.In addition,emotion dictionary is an important research tool in the field of natural language processing,and it is of great significance to the study of its construction.Because of the need to use emotion dictionary,this paper studies its construction method,and proposes an emotion dictionary construction method which combines SO-PMI algorithm,HowNet vocabulary similarity and word vector similarity,and applies it to the proposed emotional word vector optimization model.The main work of this paper is as follows:(1)Considering the need for emotion dictionary labeled with emotional intensity level in this paper,we propose an emotion dictionary construction method based on the existing research.This method combines SO-PMI algorithm,HowNet vocabulary similarity and word vector similarity,makes up for the shortcomings of the above methods,and can semi-supervise the construction of emotion dictionary with emotional intensity division.Experiments show that our emotion dictionary construction method is superior to the traditional SO-PMI algorithm,and the emotional intensity level provided by it can provide a reference for artificial emotional intensity classification.(2)In order to solve the problem of "opposite semantics but high similarity of word vectors" in sentiment analysis tasks,this paper first discusses the causes of the above problems and the influence of word vectors being adjusted,and then gets inspiration from the existing methods,and proposes an optimization method for the adjustment of emotional word vectors.This method combines emotion dictionary and judges the similarity of word vectors,and chooses an emotional word vector as the starting point to optimize and adjust.By exchanging the vector representation of emotional words within a certain similarity range,the emotional words can be exchanged to a more suitable location for their existence.In this way,the vectors of emotional words with opposite meanings are far away from each other,and the vectors of emotional words with the same meanings are close to each other according to their intensity.Then,the scope of optimization is enlarged slowly by breadth-first search,and finally the optimization of the emotional word vector is completed.(3)We combine the emotional word vector optimization model and emotion dictionary construction method proposed in this paper to conduct sentiment analysis of microblog comments.We use the optimized word vectors for the training of sentiment analysis model,and try to use the CNN model and the Bi-LSTM model for the sentiment analysis respectively,and make the experimental analysis and comparison.The experimental results show that compared with the sentiment analysis model based on the original word vector before optimization,the sentiment analysis model based on the optimized word vector in this paper improves both the overall accuracy and the F1 values of positive and negative emotional polarity,thus verifying the validity of our optimized word vector.