Research On Sentiment Classification And Evaluation Object Recognition Of English Review Text

In recent years,with the rapid development of e-commerce and social media,more and more people are active in the network and accompanied by the generation of massive text data,which contains a lot of emotional information,in order to mine and obtain valuable emotions in the text,the research on sentiment analysis of product review text has received more and more attention.Text sentiment analysis is to mine people's attitudes and emotions expressed in subjective web texts.This article considers the importance of different sentences when judging the sentiment tendency of texts,and performs sentiment classification of English review text based on key sentence extraction.At the same time,this article identifies the fine-grained evaluation objects in the review texts,which makes the sentiment analysis work more detailed.The specific research work is as follows:In the study of sentiment classification based on key sentence extraction,for the existing methods,most of them extract sentimental key sentences based on feature attribute weights,and ignore the problems of Title and context information,this paper proposes the sentiment classification based on Title and weighted TextRank the study.Firstly,the emotional key sentences are extracted through two ways,on the one hand,an extended semantic rule template and an emotional dictionary are constructed,based on this,calculate the Title's emotional score to determine Title's contribution,and determine the number of Title as emotional key sentence;on the other hand,according to the idea of weighted TextRank algorithm,a sentient sentence weighted directed graph is constructed in the document body to extract key sentences,and four factors affecting the weight of directed edges are introduced in detail.Then,the sentiment classification results obtained from the key sentences,detail sentences and full-text sentences are used as features,and xgboost algorithm is used to further determine the sentiment of the text.Finally,the experimental results show that the key sentence extraction algorithm proposed in this paper can greatly improve the effect of emotion classification,and verify the advantages of emotional key sentences and their complementarity with detailed sentences.In the research of identification of evaluation objects,in order to solve the problems of artificially setting features in statistical learning methods and the deep model relying on a large amount of labeled data and lack of prior knowledge,this paper proposes the evaluation object identification research that combines knowledge guidance and parallel models for integrated learning.Firstly,the splicing word vector is constructed as the input of the model,which consists of three parts: the feature word vector,the Glove word vector,and the Elmo weighted average word vector.The feature word vector is embedded with a product name domain dictionary.Then built four parallel deep models for ensemble learning,each individual deep model is composed of the following parts: a two-way long-term and short-term memory model LSTM?an attention model Attention?a conditional random field CRF or Softmax function.Then use the parallel model recognition results and prior knowledge such as part-of-speech and syntactic analysis tree relations,and propose a rule template incorporating knowledge guidance to improve the voting strategy.Finally,the experimental results show the effectiveness of the parallel individual depth model constructed in this paper and the rule template incorporating knowledge guidance to improve the voting strategy.