Preference Reasoning and Aggregation

Опубликовано 8 июля 2016, 1:17

Preference reasoning and aggregation are the main ingredients of a very active area of research which brings together multi-agent systems, knowledge representation, combinatorial optimisation, and voting theory. Preferences are often used in collective decision making: each agent expresses its preferences over a set of possible decisions, and such preferences are then aggregated to return the collective decision. The main challenges in this area are related to: handling a large set of candidate decisions with a combinatorial structure, as well as a possibly very large set of agents expressing their preferences; choosing among several formalisms to model preferences compactly; the presence of indifference, incomparability, and uncertainty in the preferences; and of course computational concerns. After describing the main lines of activities and the most significant results in this area, I will discuss the possible uses of preference modelling, reasoning, and aggregation in two application domains: sentiment analysis and stable matching problems. In sentiment analysis, preferences have to be extracted by a massive amount of textual data, such as tweets, then they have to be modelled faithfully, and then aggregated in an efficient way, in order to generate an accurate collective opinion. This poses several challenges in terms of handling big data, extracting relevant information from it, and learning the missing one. In stable matching problems, there are two sets of agents, each agent expresses preferences over the agents in the other set, and the aim is to find a stable matching according to the preferences. Both systematic and local search algorithms have been used to solve such problems. But to handle problems of large size, such algorithms need to be much faster, so distribution/parallelization techniques can be of great help in this respect.