Sequential Patterns-Based Rules for Aspect-Based Sentiment Analysis

摘要

Expansion, availability and acceptance of World Wide Web (WWW) have dramatically changed the way people think and express their opinions and sentiments. Due to the veracity of the internet, people are inclined to purchase online and used to express their experiences, in the form ofreviews, on the buyer’s or merchant’s websites, social media, web forums and blogs. These experiences of people contain valuable information for the manufacturers to know customer’s satisfaction against their products and for other potential buyers to acquire sufficient knowledgebefore selecting a purchase. Aspect-based sentiment analysis deals with the extraction of such valuable information or knowledge from customer’s reviews. One of the key tasks of the aspect-based sentiment analysis is to learn the association or relationship among user’s opinionsand their targets. In the recent years, dependency tree-based approaches have proved their significance to accomplish this task. The grammatical and language constraints make these approaches vulnerable. In this paper, we have used sequential pattern mining approach to identify associationamong opinions and their targets. First we mined sequential patterns from customer reviews and secondly we have defined certain rules based on the mined sequential patterns. The experimental results elaborate the effectiveness of sequential patterns in the form of sequential rules.