Sentic Panalogy: Swapping Affective Common Sense Reasoning Strategies and Foci

Sentic Panalogy: Swapping Affective Common Sense Reasoning Strategies and Foci Erik Cambria, Daniel Olsher, Kenneth Kwok {cambria, olsher, kenkwok}@nus.edu.sg Cognitive Science Programme, Temasek Laboratories National University of Singapore, 5A Engineering Drive 1, Singapore 117411 http://sentic.net Abstract To show the effectiveness of the proposed approach, termed sentic panalogy, we employ it for the natural lan- guage processing (NLP) task of sentiment analysis, for which a faceted and nuanced analysis is mostly needed. The structure of the paper is as follows: the first section provides some background information on sentiment anal- ysis; the second section introduces the concept of affective common sense reasoning and explains why and how this can aid sentiment analysis; the third and fourth sections describe the implementation of the switch among different strategies and among the foci around which such strategies are devel- oped, respectively; the fifth section provides an evaluation of the proposed approach; the last section, finally, comprises concluding remarks and future directions. An important difference between traditional AI systems and human intelligence is our ability to harness common sense knowledge gleaned from a lifetime of learning and experi- ences to inform our decision-making and behavior. This al- lows humans to adapt easily to novel situations where AI fails catastrophically for lack of situation-specific rules and gener- alization capabilities. In order for machines to exploit com- mon sense knowledge in reasoning as humans do, moreover, we need to endow them with human-like reasoning strategies. In problem-solving situations, in particular, several analogous representations of the same problem should be maintained in parallel while trying to solve it so that, when problem-solving begins to fail while using one representation, the system can switch to one of the others. Sentic panalogy is a technique that aims to emulate such process by exploiting graph-mining and dimensionality-reduction techniques to dynamically inter- change both different reasoning strategies and the foci around which such strategies are developed. Keywords: AI; NLP; Cognitive systems; Sentic computing. Sentiment Analysis Introduction Sentiment analysis is a branch of the broad field of text data mining and refers generally to the process of extracting in- teresting and non-trivial patterns or knowledge from unstruc- tured text documents. It can be viewed as an extension of data mining or knowledge discovery from (structured) databases (Fayyad, Piatetsky, & Smyth, 1996; Simoudis, 1996). As the most natural form of storing information is text, sentiment analysis is believed to have a commercial potential higher than that of data mining. Sentiment analysis, however, is also a much more complex task as it involves dealing with text data that are inherently unstructured and fuzzy. It is a multi- disciplinary research area that involves the adoption of tech- niques in fields such as text analysis, information retrieval and extraction, auto-categorization, machine learning, clustering, and visualization. Most of the existing approaches to opinion mining and sen- timent analysis rely on the extraction of a vector representing the most salient and important text features, which is later used for classification purposes. Some of the most commonly used features are term frequency (Wu, Luk, Wong, & Kwok, 2008) and presence (Pang, Lee, & Vaithyanathan, 2002). The latter, in particular, is a binary-valued feature vectors in which the entries merely indicate whether a term occurs or not, and formed a more effective basis for polarity classification. This is indicative of an interesting difference between typ- ical topic-based text categorization. While a topic is more likely to be emphasized by frequent occurrences of certain keywords, overall sentiment may not usually be highlighted through repeated use of the same terms. Emotions are different Ways to Think (Minsky, 2006) that our mind triggers to deal with different situations we face in our lives. Our decision-making and problem-solving skills, in fact, are strictly dependent on both our common sense knowl- edge about the world and the appraisal associated with this (Scherer, Shorr, & Johnstone, 2001). The capability to ac- cordingly compress and exploit such information, which we term affective common sense reasoning (Cambria, Olsher, & Kwok, 2012), is a fundamental component in human experi- ence, cognition, perception, learning, and communication. For this reason, we cannot prescind from emotions in the development of intelligent user interfaces: if we want com- puters to be really intelligent, not just have the veneer of in- telligence, we need to give them the ability to recognize, un- derstand, and express emotions. Furthermore, in order not to get stuck and to be able to tackle different problems from dif- ferent perspectives, an intelligent machine should not have a unique way to deal with a task, but rather be endowed with different reasoning strategies and with the capability to ac- cordingly switch among these. This work further develops a recently proposed approach (Cambria, Mazzocco, Hussain, & Durrani, 2011) for the em- ulation of the human capability to switch between different perspectives and find novel ways to look at things. Such ap- proach is inspired by Minsky’s notion of ‘panalogy’ (parallel analogy), which states that several analogous representations of the same problem should be maintained in parallel while trying to solve it (Minsky, 2006).