Research And Implementation Of Recognition And Classification Algorithm For Sentiment Texts

In recent years,the social media information in the Internet has grown rapidly.It is very meaningful to make good use of these network text data and to mine and analyze valuable sentiment information.Therefore,the emotional sentiment analysis has become a research hotspot in the field of NLP.A large number of web texts include subjective sentences and objective sentences,in which subjective sentences are the research objects of sentiment analysis.Therefore,this paper firstly implements the extraction of subjective text and conducts sentiment analysis on subjective texts,and experiments whether it can assist the calculation of sentiment analysis.The main work of this paper includes:(1)In the process of identifying sentiment texts,this paper takes the part-of-speech features of words as an important clue for subjective and objective sentences classification.According to the results of part-of-speech tagging,the word-of-speech categories of nine subjective features are summarized.This paper designs a subjective and objective sentences classification model based on part-of-speech features and CNN.The part of speech is combined with the word vector and then input to the CNN.The dimension of the part-of-speech vector and the size of the convolution kernel are determined by experiments.The experimental results show that the precision of the model with part-of-speech features is increased by 2%compared with the original model,and the precision is improved by 4%compared with the traditional subjective and objective classification models using N-POS and support vector machines.(2)In the sentiment analysis of subjective texts,the turning and summing sentences in the text determine the emotional polarity of the whole sentence.Therefore,this paper proposes an auxiliary sentiment analysis method based on trunk analysis.This paper summarizes the conjunctions dictionary,which contains 33 turning conjunctions and concluding conjunctions,and identifies the trunk of the whole sentence by matching with the dictionary.The experimental results show that the precision of the original model is increased by 1%after adding the trunk analysis.(3)In each clause of the text,emotional words show different emotional strength because they are modified by different adverb of degree.Therefore,this paper proposes an sentiment analysis method of weighted word vector.According to the different degree adverbs,the word vector is given the corresponding weight by combining the degree adverb dictionary.The experimental results show that the precision of the sentiment analysis method model based on the weighted word vector is improved by 1.6%compared with the original model(4)Finally,this paper sets up a comparative experiment of six groups of sentiment analysis models.The precision of the system is improved by 2.6%,based on based on the method of emotional weight word vector and trunk analysis.Compared with the traditional sentiment analysis method based on sentiment dictionary and support vector machine,the precision is improved by 8%.The experimental results show that the precision of the sentiment analysis model based on LSTM neural network is improved after adding the weighted word vector and trunk analysis of the sentences.This model also sets up a comparative experiment for sentiment analysis of dataset without classifying subjective and objective sentences.The results show that the text after the subjective and objective classification reduces the noise caused by the objective text and improves the precision of the sentiment analysis.