Application Research Of Crosslingual Sentiment Classification Technology In Product Reviews

In pace with the rapid development of Internet technology and the continuous improvement of e-commerce,more and more users have begun to share their comments and views on products on the e-commerce platform,resulting in a large number of product comment texts.Sentiment classification of these product comment data has high commercial value and research significance.However,the sentiment resources of different languages are unevenly distributed in quality and quantity,and it is impossible to better classify languages with insufficient sentiment resources.Many small languages still have needs and research significance for sentiment classification.Therefore,the cross-lingual sentiment classification method uses the annotation data and sentiment resources of languages such as English to help other resource-poor languages to perform sentiment classification.In the study of crosslingual sentiment classification,the traditional method is to use parallel corpora or machine translation to associate the two languages.Sentiment expressions are very different between different languages,causing translation errors in the translation system itself.Machine translation does not solve the problem of language differences,and high-quality large-scale parallel corpora are not easy to obtain in many practical environments.To solve this problem,this paper takes the user product reviews as the analysis object,and implements cross-lingual sentiment classification by establishing a shared space independent of the two languages,so as to judge the sentiment tendency of product reviews.The main research contents are as follows:(1)A cross-language sentiment classification method based on shared space is proposed.We use TF-IDF and LDA algorithms to build bilingual dictionaries and perform crosslanguage semantic conversion.Meanwhile,choose polarizing and derogatory dictionaries to obtain polar information,learning a shared space independent of two languages.Different polarities words can be well distinguished on the shared space.In order to verify the effectiveness of the method,this paper selects three areas of user product reviews on Amazon website to conduct experiments.The experimental results on the NLP&&CC2013 crosslingual sentiment classification dataset show that this method can effectively improve the classification effect on the target language.(2)In cross-language sentiment classification,different parts of the text contribute differently to the sentiment of the text.This paper proposes a cross-lingual sentiment classification method(CLSA)that combines attention mechanisms and sentiment features.The sentiment context is extracted separately to obtain the sentiment features,and the attention mechanism is used to make the text pay more attention to the parts that are important to the sentiment.Finally,the acquired sentiment features are integrated into the GRU model,and the extracted sentiment semantic information is used to obtain sentiment attention and express the text.The final cross-lingual sentiment classification is carried out on the shared space.The experimental results show demonstrate that the technique proposed in the paper is performed on the same dataset as other methods,which proves the effectiveness of the method in solving cross-lingual sentiment classification tasks.