Bidirectional Independently Long Short-Term Memory and Conditional Random Field Integrated Model for Aspect Extraction in Sentiment Analysis

Abstract

Aspect extraction or feature extraction is a crucial and challenging task of opinion mining that aims to identify opinion targets from opinion text. Especially, how to explore these aspects or features from unstructured comments is a matter of concern. In this paper, we propose a novel supervised learning approach using deep learning technique for the above-mentioned aspect extraction task. Our model combines a bidirectional independently long short-term memory (Bi-IndyLSTM) with a Conditional Random Field (CRF). This integrated model is trained on labelled data to extract feature sets in opinion text. We employ a Bi-IndyLSTM with word embeddings achieved by training GloVe on the SemEval 2014 data set. There are 6086 training reviews and 1600 testing reviews on two domains, Laptop and Restaurant of the SemEval 2014 data set. Experimental results showed that our proposed Bi-IndyLSTM-CRF aspect extraction model in sentiment analysis obtained considerably better accuracy than the state-of-the-art methods.