Sentiment analysis: dynamic and temporal clustering of product reviews

摘要

The increased availability of online reviews requires a relevant solution to draw chronological insights from review streams. This paper introduces temporal sentiment analysis by adopting the automatic contextual analysis and ensemble clustering (ACAEC) algorithm. ACAEC is a clustering algorithm which utilizes contextual analysis and a clustering ensemble learning. We propose chronological sentiment analysis using window sequential clustering (WSC) and segregated window clustering (SWC). WSC is a dynamic analysis, whereas SWC is solely based on the temporal characteristic of reviews. ACAEC is the base learning algorithm of WSC and SWC. ACAEC’s ensemble approach is enhanced using an additional weight scheme and an additional learner to improve WSC’s outcome. To understand the produced sentiment pattern, an unsupervised review selection is introduced which is based on review polarity. We also introduce consistency, a free-label measure to assess the algorithm’s performance. For this study, new sets of reviews are introduced, these being four airlines and an Australian property agent. In terms of accuracy and stability, the proposed methods are effective in processing a review series. Experiments show that the average accuracy rates of SWC and WSC reach 87.54% and 83.87%, respectively. In addition, it is robust against the so-called imbalanced windows problem. The suggested solutions are unsupervised i. e. domain-independent and suitable for the analysis of a large review series.