Movie Aspects Identification Model (MAIM) for Aspect Based Sentiment Analysis

Based Sentiment Analysis techniques have been widely applied in several application domains. During the last two decades, these techniques have been mostly developed for the domain of product and service reviews. However, very few Aspect Based Sentiment Techniques have been proposed for the domain of movie reviews. In contrast to most studies that focus on movie specific aspects such as Script, Director, and Actor, this work focus on NER (Named Entity Recognition) in order to find out entity-specific aspects. Consequently, MAIM (Movie Aspects Identification Model) is proposed that can extract not only movie-specific aspects but can also identify Named Entities (NEs) such as Person Name and Movie Title. The three main contributions in this paper are (i) identification of infrequent aspects, (ii) identification of NEs, and (iii) identification of N-gram opinion words as an entity. MAIM is implemented using BiLSTM-CRF (Bidirectional Long Short-Term Memory - Conditional Random Field) hybrid technique and tested on movie reviews dataset. The results showed a precision score of 89.9%, recall of 88.9%, and f1-score of 89.4%. The results of the hybrid model are compared with the baseline models i.e., CRF (Conditional Random Field) and LSTM-CRF (Long Short-Term Memory - Conditional Random Field) and shown hybrid model outperforms both models in term of precision, recall and f1-score.