In this paper, we focus on a personalized information retrieval system based on multi-agent platform. Especially, they are capable of sharing information between them, for supporting collaborations between people. Personalization module has to be exploited to be aware of the corresponding user's browsing contexts (e.g., purposes, intention, and goals) at the specific moment. We want to recommend as relevant information to the estimated user context as possible, by analyzing the interaction results (e.g., clickstreams or query results). Thereby, we propose a novel approach to self-organizing agent groups based on contextual synchronization. Synchronization is an important requirement for online collaborations among them. This synchronization method exploits contextual information extracted from a set of personal agents in the same group, for real-time information sharing. Through semantically tracking of the users' information searching behaviors, we model the temporal dynamics of personal and group context. More importantly, in a certain moment, the contextual outliers can be detected, so that the groups can be automatically organized again with the same context. The co-browsing system embedding our proposed method was shown 52.7 % and 11.5 % improvements of communication performance, compared to single browsing system and asynchronous collaborative browsing system, respectively.
展开▼
