Research On Cross-lingual Sentiment Classification Technology Based On Product Reviews

With the rapid development of science and technology,the Internet is so full of users' comments on goods or services.The most valuable information is the sentiment information that people have product reviews.This information has a very good research on sentiment classification.However,due to the uneven distribution of quality and quantity of sentiment resources in different languages,it is impossible to better classify the sentimentally resourced languages.In order to address this problem,researchers usually use language with rich sentiment resources to solve the cross-lingual sentiment classification problem of resourcepoor language.The traditional cross-lingual sentiment classification mainly uses a machine translation system to convert a language translation to another language,and then to perform sentiment classification under the language,but the quality of machine translation seriously affects the classification accuracy.The current cross-lingual sentiment classification problem is how to better reduce the gap between diverse languages.To solve this problem,this paper analyzes in detail the problems faced by cross-lingual sentiment classification and finds the mapping between two distinct languages.Relations,cross-lingual sentiment classification through this mapping,the main work are as follows:1)Laplace mapping is used to improve the cross-lingual sentiment classification algorithm(CLSCL)based on structure-based learning,and an improved algorithm(MCLSCL)is proposed.According to the probable relationship between the source language and target language,the pivot is selected.Feature word pairs,and finally learn a mapping function with the selected pivot feature word pairs and use this function to perform crosslingual sentiment classification.2)A cross-lingual sentiment classification method based on Autoencoder is proposed,which learns a shared representation(BLSR)through Autoencoder in two different languages(source language and target language).For learning and processing two different languages through Autoencoder,the shared representation space is reached.After learning more about the space,the training data(source language)is mapped into space and the classifier is trained.The test data(target language)is further mapped to space and carry out testing.3)The two proposed algorithms are selected from the data of DVD and book reviews in the NLP&CC 2013 Cross-Language Emotion Classification Dataset.Through the experimental results,the cross-language emotion classification algorithm and the original algorithm are proposed in this paper.Comparing experiments on the same data set has higher accuracy;the second method has higher accuracy of sentiment classification on the dataset than BSWE.