A Sequential Pattern Rule-Based Approach For Explicit And Implicit Aspect Extraction From Online Product Reviews
Loading...
Date
2017-08
Authors
Ahmad Rana, Toqir
Journal Title
Journal ISSN
Volume Title
Publisher
Universiti Sains Malaysia
Abstract
With the tremendous growth of the World Wide Web (WWW) and acceptance of online shopping, WWW has become a prime source of users‘ generated reviews. These reviews can play an important role for companies and governments to know the public opinions which are articulated within the reviews. Aspect-based sentiment analysis deals with the extraction of such user generated fine-grained opinions and their targets (aspects) from online reviews. The main task focuses on the identification of the aspects which are the targets of users‘ opinions or sentiments from online reviews. Most of the current approaches rely on syntactic patterns for aspect extraction task. However, these approaches heavily depend upon grammatical and language constraints. On the contrary, users do not give much importance to these constraints which makes these approaches vulnerable. In this work, sequential pattern mining approach has been adopted for exploring associations between opinions and aspects, and defining sequential patterns-based rules for aspect extraction. Using the sequential patterns-based rules, a proposed two-fold rules-based model (TF-RBM) extracts the explicit aspects. The first fold extracts aspects associated with domain independent opinions and the second fold extracts aspects associated with domain dependent opinions. The frequency- and similarity-based approaches are then applied to improve the aspect extraction accuracy of the proposed model. For implicit aspects, a multi-level knowledge-based model has been proposed which performs the task in two steps. In the first step, the proposed model extracts opinions and implicit aspect clues based upon the sequential patterns-based rules along with manually crafted rules for implicit aspects. The second step uses a multi-level knowledge-based approach to identify implicit aspect with the help of extracted implicit aspect clues and opinion words. For the experimental evaluation, the proposed model is compared with the state-of-the-art and recent approaches for explicit aspect extraction. The proposed approach shows better results in terms of recall and F1-score. For implicit aspects, the proposed model is the first attempt to deal with diverse nature of aspect and hence no comparable work is available.
Description
Keywords
WWW has become a prime source , of users‘ generated reviews