Cross-lingual Sentiment Analysis Based On Sentiment Transfer

With the exponential growth of a large number of subjective data on the Internet,the analysis of these rich data has attracted the attention of researchers.At the same time,the speed of globalization of Internet is accelerating,and the characteristics of multi-lingual information are becoming more and more obvious.Therefore,it is a hot spot to explore some of the languages that have started earlier and studied with rich resource to help the languages which are lack of resources.In this paper,two methods of cross-language sentiment analysis based on sentiment correlation are proposed to solve the problem of cross-language research task.The main innovations of this paper are as follows:(1)we propose to transfer the sentiment knowledge across languages by finding the sentiment connection which in this paper refers to the lexicon-level sentiment alignment across languages.To capture the degree of the alignment,we consider the sentiment projection as a two-step semi-supervised cross-lingual sentiment voting.(2)In this paper,we propose to capture the document-level sentiment connection across languages(called cross-lingual sentiment relation)for cross-language sentiment analysis by joining two convolutional neural networks(CNNs)as a novel bi-view CNN.Inspired by relation embedding learning,we first project the extracted parallel sentiments into a bilingual sentiment relation space,then capture the relation by subtracting them with an error-tolerance.Finally,we predict the bilingual sentiment relation type as their sentiment polarity.(3)In this paper,We evaluate the proposed models on the dataset of an open cross-lingual sentiment analysis task in NLP&CC 2013.Experiments on several public datasets demonstrate the effectiveness and efficiency of the proposed approaches.