An interactive multi-agent reasoning model for sentiment analysis: a case for computational semiotics

Semiotics is a domain that studies signs. For Peircean semiotics, a sign is not a dyadic entity, composed of a word and its meaning. Instead, meaning-making is a process of signification borne out of a strictly triadic relationship; in which a representative of a sign (word(s)) stands for its object (or meaning,) but never in a vacuum, and always for an interpretant. For Peirce, it is this third, this interpretant, through which the sign displays its meaning. What is even more important is that this rational process of signification is never being carried out by a single Mind, it requires a community of reasoners. In semiotic terms this article translates the sentiment analysis problem as follows: A sentence/comment is a representamen which has a sentiment value (its object) for an evolving community of reasoning agents (interpretant.) This article presents an interactive multi-agent system in which the agents implicitly model other agents, with a semiotic based approach towards sentiment analysis. It then tests the system on an original student evaluation of teachers dataset, compares the results with deep learning and other baseline techniques, and aims to propose semiotics as a reparative alternative to the dominant dichotomies—rule-based and data-based camps within artificial intelligence.