NOVEL MULTI-LEVEL ASPECT BASED SENTIMENT ANALYSIS FOR IMPROVED ROOT-CAUSE ANALYSIS

Abstract

Aspect extraction and sentiment identification are the two important tasks to provide effective root cause analysis. This work presents a Multi-Level Aspect based Sentiment Analysis (MLASA) model that integrates the aspect extraction and sentiment identification modules to provide effective root cause analysis. The aspect extraction module performs token filtration, followed by rule based aspect identification. The heterogeneous multi-level sentiment identification phase performs aspect based sentiment identification. First level performs magnitude along and polarity identification of text, while the second level performs polarity identification using multiple machine learning models. The results are aggregated and ranked based on aspect significance and sentiment magnitude. Experiments and comparisons show effective performance of the MLASA model.