Unsupervised word-level affect analysis and propagation in a lexical knowledge graph

Abstract

Lexical sentiment analysis (LSA) is of central importance in extracting and analyzing user moods and views on the Web. Most existing LSA approaches have utilized supervised learning techniques applied on corpus-based statistics, requiring extensive training data, training time, and large statistical corpora which are not always available. Other studies have utilized unsupervised and lexicon-based approaches to match target words in a lexical knowledge base (KB) with seed words in a sentiment lexicon, usually suffering from the limited coverage or inconsistent connectivity of affective concepts. In this paper, we introduce LISA, an unsupervised word-level knowledge graph-based LSA framework. It uses different variants of shortest path graph navigation techniques to compute and propagate affective scores in a lexical-affective graph (LAG), created by connecting a typical lexical KB like WordNet, with a reliable affect KB like WordNet-Affect Hierarchy (where any other lexical or affective KB can be utilized). LISA was designed in two consecutive iterations, producing two main modules: i) LISA 1.0 for affect navigation, and ii) LISA 2.0 for affect propagation and lookup. LISA 1.0 suffered from the semantic connectivity problem shared by some existing lexicon-based methods, and required polynomial execution time. This led to the development of LISA 2.0, which i) processes affective relationships separately from lexical/semantic connections (solving the semantic connectivity problem of LISA 1.0), and ii) produces a sentiment lexicon which can be searched in logarithmic time (handling LISA 1.0’s efficiency problem). Experimental results on the ANEW dataset show that our approach, namely LISA 2.0, while completely unsupervised, is on a par with existing (semi)supervised solutions, highlighting its quality and potential.