A Hybrid Capsule Network with Attention and BiLSTM for Opinion Mining in Text

Abstract

Examining the product review or review on web service facilitate to increase the product quality or web service. By means, comments from online shopping sites such as Amazon, Flipkart, EBay etc will not only assist the users to purchase the product however besides be able to guide the producer/supplier to identify the advantages and disadvantages of the goods. Mining online shopping websites with their information will becomes a major important task. Web Mining plays a major important role to mine the details of online websites efficiently. Web mining is the process of data mining that learns without human intervention and mine information obtained from the documents in web and also services. Sentiment categorization and web mining has become a truly significant task recently, with profound business and research impact. Machine learning algorithms and soon after deep learning methods have been the market leaders in sentiment analysis. The advent of capsule networks has been a landmark event in deep learning. It has been truly proficient in image processing. In case of text classification, standalone capsule networks are not optimally suitable. Here, a hybrid BiLSTM-Capsule framework is introduced for sentiment analysis of web texts of reviews from various datasets. In the model beginning, there is a bidirectional LSTM layer after which is an attention layer and a final capsule layer. This review analysis will helps to improve the products from amazon, increase the movie quality. The analysis of outcome depending on MR, IMDB, SST and Amazon datasets indicated the introduced framework performs better than some benchmark deep learning models. Significantly, the BiLSTM-Capsule can put its words in sentimental trend showing the capsules’ attributes without utilizing the linguistic knowledge.