Home // DATA ANALYTICS 2017, The Sixth International Conference on Data Analytics // View article

Aspect Term Extraction from Customer Reviews using Conditional Random Fields

Authors:
Hardik Dalal
Qigang Gao

Keywords: Aspect-based Sentiment Analysis; Aspect-term Extraction; Data Analytics; Conditional Random Fields

Abstract:
E-commerce customers generate a vast amount of information about services and products using comments and blogs. Customer reviews serve as one source of this information and they are a critical aspect of e-Business. Reviews are a vital source of feedback and they also help businesses to determine market trends, demographics, and develop knowledge about their competition. Collecting reviews from customers is only half of the challenge. The other half includes mining these reviews to gain insights. Sentiment Analysis techniques help to extract sentiments and determine the perceived product quality or level of customer satisfaction. Our work is focused on detecting product features from customer reviews which, is a part of Aspect Level Sentiment Analysis research. We address the task by expressing it as a sequence-labeling problem in which features are required to be labeled from sentences. The process is similar to that of Named Entity Extraction (NER). However, we are now targeting a different type of entity, i.e., product features. In comparison to NER, Aspect Term Extraction (ATE) poses unique challenges and we address them using Conditional Random Field (CRF), a conditional probability based model. Using dependency parsing, we have engineered a set of optimum features that allow for promising results.

Pages: 73 to 79

Copyright: Copyright (c) IARIA, 2017

Publication date: November 12, 2017

Published in: conference

ISSN: 2308-4464

ISBN: 978-1-61208-603-3

Location: Barcelona, Spain

Dates: from November 12, 2017 to November 16, 2017