BabelSenticNet: A Commonsense Reasoning Framework for Multilingual Sentiment Analysis

Abstract

SenticNet is a concept-level knowledge base used to develop commonsense reasoning algorithms for sentiment analysis tasks. One of the challenges that this resource must overcome is its lack of availability for languages aside from English. Prototype algorithms have been recently proposed to create non-English language concept-level knowledge databases, but they rely on a number of heterogeneous resources that complicate comparison, reproducibility and maintenance. This paper proposes an easy and replicable method to automatically generate SenticNet for a variety of languages, obtaining as a result BabelSenticNet. We use statistical machine translation tools to create a high coverage SenticNet version for the target language. We then introduce an algorithm to increase the robustness of the translated resources, relying on a mapping technique, based on WordNet and its multilingual versions. SenticNet versions for 40 languages have been made available. Human-based evaluation on languages belonging to different families, alphabets and cultures proves the robustness of the method and its potential for utility in future research on multilingual concept-level sentiment analysis.