Multi-aspect Sentiment Analysis with Topic Models Modeling

Abstract

Abstrac t-We have tried to examine the viability of topic model based methodologies to two mu lti-aspect sentiment analysis tasks: mult i-aspect sentence labeling and mult iaspect rating prediction. Fo r o ne o f th e t as ks o f sentence labeling, we propose a weakly -superv ised app roach that utilizes only minimal prior knowledge-in the form of seed words-to uphold an immed iate correspondence between topics and aspects. This correspondence will be utilized to name sentences with execution that approaches a fully supervised standard. For multi-aspect rating prediction, we find that general evaluations can be utilized as a part of conjunction with our sentence labeling to accomplish sensible execution contrasted with a fully supervised baseline. At the point when highest level perspective evaluations are accessible, we find that topic model based characteristics can be utilized to enhance unsophisticated supervised pattern execution, in concurrence with past mult i-aspect rating prediction work...