Experiment on sentiment embedded comparison interface

详细信息    查看全文

• 作者：Li Chen ; ^{lichen@comp.hkbu.edu.hk" class="auth_mail} ; Feng Wang ^{fwang@comp.hkbu.edu.hk" class="auth_mail} ; Luole Qi ^{llqi@comp.hkbu.edu.hk" class="auth_mail} ; Fengfeng Liang ^{ffliang@comp.hkbu.edu.hk" class="auth_mail}
• 关键词：Consumer reviews ; E-commerce ; Product comparison ; Sentiment analysis ; Feature-based review summarization ; User study
• 刊名：Knowledge-Based Systems
• 出版年：July 2014
• 年：2014
• 卷：64
• 期：Complete
• 页码：44-58
• 全文大小：2731 K

文摘

Because the large amount of product reviews has been appearing in the current e-commerce sites, it becomes increasingly important to summarize these reviews, so as to support online buyers’ information-seeking and decision-making process. However, little work has investigated how to present the sentiment information (as extracted from reviews) on the user interface, especially in the interface of supporting users to compare products. In this manuscript, we design three alternative sentiment-embedded comparison interfaces based on popular techniques, respectively called opinion table, opinion bar chart, opinion cloud. We then report results from two user studies on the developed interfaces. The first user study verified (1) the important role of comparison matrix in users’ decision process, (2) the benefit of incorporating reviews into the comparison interface, and (3) the positive effect of showing features’ sentiment info on aiding users’ product comparison. Motivated by the first study’s results, we performed the second user study to in depth compare the three alternative designs empirically. It turns out that the opinion bar chart, that mainly visualizes numerical feature sentiment scores via bars and qualitative adjective words via tool tip window, achieved significantly higher user assessments in terms of perceived information sufficiency, perceived ease of use and perceived cognitive effort. Users also behaved more active in opinion bar chart by manipulating the extracted features while less frequently viewing the raw textual reviews. The opinion cloud, that primarily visualizes the feature-associated opinion words in form of adjusted tags, was shown with better performance than opinion table, but slightly lower favor than opinion bar chart. In addition, this study revealed the effectiveness of showing opinion features (i.e., features with sentiment) in allowing users to examine the similarity and contrast across multiple products, and hence enabling them to make an informed and confident decision at the end.