Analysis of Sentiment on Movie Reviews Using Word Embedding Self-Attentive LSTM
S. Sivakumar, R. Rajalakshmi
Abstract
In the contemporary world, people share their thoughts rapidly in social media. Mining and extracting knowledge from this information for performing sentiment analysis is a complex task. Even though automated machine learning algorithms and techniques are available, and extraction of semantic and relevant key terms from a sparse representation of the review is difficult. Word embedding improves the text classification by solving the problem of sparse matrix and semantics of the word. In this paper, a novel architecture is proposed by combining long short-term memory (LSTM) with word embedding to extract the semantic relationship between the neighboring words and also a weighted self-attention is applied to extract the key terms from the reviews. Based on the experimental analysis on the IMDB dataset, the authors have shown that the proposed architecture word-embedding self-attention LSTM architecture achieved an F1 score of 88.67%, while LSTM and word embedding LSTM-based models resulted in an F1 score of 84.42% and 85.69%, respectively.