Sentence-Level Sentiment Analysis via Sequence Modeling

This paper presents a method of improving the performance of sentence-level sentiment analysis. Sentiment analysis is generally understood as a task that requires a deep understanding of the sentence structure (e.g., word order and non-local dependency). To attack this problem without the sentence parsing, we propose a novel approach that decomposes a sentence into a series of sub-sequences. Sentence-level polarity is then determined by classifying within sub-sequences and by fusing the obtained sub-sequences polarities. Extensive evaluations are conducted on one benchmark dataset for sentence polarity detection. Experimental results show that the performance of our proposed method outperforms two baselines based on Support Vector Machines (SVMs) and Logistic Regression (LR), respectively.