SCM: a new user behaviour model for web search

S. I. Nikolenko^ab, A. A. Fishkov<sup>c

a</sup> St. Petersburg Department of V. A. Steklov Institute of Mathematics, Russian Academy of Sciences
^b St. Petersburg Academic University — Nanotechnology Research and Education Centre of the Russian Academy of Sciences (the Academic University)
^c Saint-Petersburg State Technical University

Abstract: User behaviour models represent one of the most important directions of studies aimed at enhancing web search; such models are usually based on probabilistic graphical models and are learned from logs of user actions (click logs). In this work, we introduce a new user behaviour model called SCM (session click model). We show that the new model has simpler inference algorithms but in practice gives better results than existing models.

Keywords: user behaviour models, probabilistic graphical models, expectation propagation.

UDC: 004.8

Received: 06.06.2012