Summary: Artificial intelligence (AI) has reached virtually every aspect of our lives. With increased computing power, networking resources, storage facilities and specialized, efficient algorithms, the age-old objective to have machines think and act autonomously is becoming more and more a reality. Yet there still remains an important barrier to be crossed: it turns out extremely difficult to capture the “meaning” of inherently imprecise concepts in a way that allows computers to reason with them in a logical, human-like way. For example, in the context of a travel web platform, it would be too simplistic to reduce the popularity of a hotel to its average user rating; rather, one would expect a machine to perform sentiment analysis to user comments, i.e., extract positive and negative opinions from them (requiring it to deal with ambiguity, irony, reviewer credibility, etc.), match this information with individual user preferences (which may be expressed as a list of restrictions and wishes, each with their own weight), and justify its recommendations.
Fuzzy sets model and process vague concepts and partial truth, while rough sets handle incomplete information in an approximate, granular way. In this project, we aim to expand their use for machine learning tasks, with a specific view towards sentiment analysis as a challenging and timely topic. As a particular case study, we plan to evaluate the research results in a virtual travel agent that assists a user with planning an itinerary.