PIP CONICET

Personalized Recommendation to Groups of Users based on Multi-agent Systems

PIP Project 2015-2017

Funded by CONICET

Recommendation systems have established themselves as powerful tools to help reduce the information overload faced by users of software applications in the processes of information search, as they help filter the retrieved items, using different techniques to identify those items that best satisfy the preferences or needs of users. Traditionally, these systems have focused on meeting the needs of a particular user, analyzing your preference profile as well as your browsing history or previous searches. In the In recent years, it has begun to be considered that the users of an application are not always users individual, but can also be groups of users. This new approach makes the task of generation of recommendations that satisfy the group as a whole depends not only on opinions individual but also social factors, such as the role of each member, their personality versus conflict situations, and group dynamics, among others.

In this context, this project proposes to investigate a group recommendation approach based on a multi-agent architecture, where each of the group members is represented by a agent. Each individual agent will be able to learn and interpret both the preferences and interests of his user as their social profile, which will determine potential social influences and / or personalities before conflict situations. The recommended approach adopted will be to create a group model, that is, from the individual profiles of the group members, a model will be obtained that reflects the preferences of the group as a whole. For the construction of this model, the individual agents begin a negotiation process where each agent has individual objectives that, in many cases will be conflictive, and must negotiate with the rest of the agents which elements make up the group profile (eg pairs of attributes and values ​​that define characteristics of movies or tours). Thus the agents have a Common “global” objective: to reach an agreement for the construction of the model that defines the group. The rules established for negotiation will be given not only by individual preferences, but also by also by the social profile. Individual agents will analyze different social factors in order to recognize possible changes of opinion and detect different types of social influence. On the other hand, it is also use negotiation techniques between agents in the final decision-making stage, that is, at the moment to select one or more of the different alternatives or suggestions generated by the recommendation engine from the group model obtained.

PI: Silvia Schiaffino

Research Team:

- Ariel Monteserín

- Andrés Díaz Pace

- Christian Villavicencio