Research And Implementation Of Commentary Sentiment Analysis System Based On Word2Vec New Word Rdcognition

With the rapid development of the Internet and social networks,emotional tendency analysis has become an important research content of information content management.In the past,the research and implementation of sentiment analysis mainly used sample supervised learning and the use of sentimental dictionaries.However,there are both shortcomings:1.The use of sample supervised learning methods is very sensitive to the quality of the sample,requires a large amount of quality training samples,and requires different model data for the sentiment analysis of texts in different topics and fields,which has great limitations in practical use.2.Using the emotional dictionary for analysis,due to the lagging of the dictionary due to the emergence of new words in the network and the quality of the dictionary itself,the emotional dictionary can not be matched to valid emotional words,causing analysis failure.In addition to the above issues,the recognition of new emotional words and the expansion of the dictionary mainly rely on Word2 vec for word vectorization and semantic similarity calculation.However,due to the fact that the words with opposite emotions tend to have similar contexts,Word2 vec can not distinguish between positive and negative words,resulting in misclassification of emotional words and seriously affecting the quality of the dictionary.This paper aims at improving the insufficiency of sentiment analysis using emotion dictionary,researching new word recognition technology and sentence-level emotion analysis technology,and implements a system that combines the automatic expansion of sentiment dictionary and sentiment analysis.The major work completed in this paper includes:1.The algorithm of word2 vec model was trained by combining the knowledge of antonymous and antonymous words,and the engine for the recognition of new emotions and the expansion of the dictionary were studied and implemented,and the problem of distinguishing words with opposite emotions in word vector representation was mainly solved.2.This paper proposes a new rule for the classification of emotional words,and designs and implements a sentence-level sentiment analysis system for Chinese sentences.3.Realize the collection of basic data.Mainly Tencent news and commentary data(last half year).And synonym-antonym data from the word forest network,about 97 w.4.Based on the SSH framework,an interactive website for task management and result analysis of web crawlers,emotional dictionaries,and sentiment analysis systems was designed and implemented.The system's emotional new word recognition module is verified on the dataset of Tencent News.Compared with the original Word2 vec model in the recognition of emotional new words,the error rate of the fusion word vector model has dropped by an average of 80% or more,effectively improving emotions.Dictionary automatically expands the quality.The highest accuracy rate of sentiment analysis for news commentary is up to 92.3%.