Research

Peer Networks, Society and the Word of Mouth in Web Search

Professor Menczer

In spite of progress in crawling performance, indexing, retrieval and ranking, the ``one engine fits all'' model of search cannot scale well with the growth, dynamics, and heterogeneity of the Web and its users. One central weakness in the centralized model of search is the lack of context. Users are viewed as independent information seekers, and information resources as passive repositories of knowledge.

Professor Menczer addresses these limitations through the study of new adaptive techniques to synergistically combine two emerging technologies: social bookmarking and peer networks. Social/collaborative engines aggregate and share user recommendations, opinions, and annotations. Mechanisms include tagging/folksonomies, ratings, voting, shared categories, and hierarchical similarity (see our own project GiveALink.org). They leverage word of mouth mechanisms to accelerate the propagation of worthy memes.

Research on peer networks has produced robust architectures that are ideal for brokering among individual needs and catering to communities. In particular we have shown that peer-based search systems driven by simple distributed adaptive query routing algorithms can spontaneously organize into small-world networks with efficient communication and with emerging clusters capturing semantic locality.