Sentiment Analysis Across Languages Based on Domain-Specific Emotional Dictionary

Abstract

With the rapid pace of globalization, various comments and communication platforms in different languages spring up online. Correspondingly, cross-language sentiment analysis becomes particularly important. In this paper, we describe our experience of participation in the sixth Chinese Opinion Analysis Evaluation task 2. For the deficient training corpus of specified target language, a classified method, which gets sentiment orientation across language relied on emotional dictionary, is proposed in this paper, with aid of translation tools and training corpus in Chinese. The special way goes as follows: First, Fisher criterion is used to identify benchmark words. Considering the correlation between words, words and sentences, words and documents, as well as impacts of denied words on different sentiment words, clustering gets completed by the improved information bottleneck algorithm. Subsequently, a Chinese emotional dictionary based on sentence structure will be built. Then, calculate the sentiment weights of those translated texts by referring to negative word dictionary and degree adverb dictionary, further, the trend of texts can be identified. Evaluation results have proven that the proposed method is feasible.