Linguistic Knowledge Based on Attention Neural Network for Targeted Sentiment Classification

Abstract

Deep learning approaches for targeted sentiment classification do not fully exploit linguistic knowledge. In this paper, we propose a Linguistic Knowledge based on Attention Neural Network (LKAN) to employ linguistic knowledge (e.g. sentiment lexicon, negation words, intensity words) to benefit targeted sentiment classification. Firstly, we extract linguistic knowledge words (e.g. sentiment lexicon, negation words, intensity words) in sentences by HowNet vocabulary. Then, we design an attention mechanism which drives the model to concentrate on such words and get a weighted combination of word embeddings as the final representation for the sentences. We evaluate our proposed approach on SemEval 2014 Task 4, whose performance as shown reaches the most advanced level.