S. Carberry et al. (Eds.): UMAP 2013, LNCS 7899, pp. 356–358, 2013. © Springer-Verlag Berlin Heidelberg 2013 Multilingual vs. Monolingual User Models for Personalized Multilingual Information Retrieval M. Rami Ghorab, Séamus Lawless, Alexander O’Connor, Dong Zhou, and Vincent Wade Centre for Next Generation Localisation, Knowledge & Data Engineering Group, School of Computer Science & Statistics, Trinity College Dublin, Ireland {ghorabm,seamus.lawless,alex.oconnor,vincent.wade}@css.tcd.ie, dongzhou1979@hotmail.com Abstract. This paper demonstrates that a user of multilingual search has differ- ent interests depending on the language used, and that the user model should re- flect this. To demonstrate this phenomenon, the paper proposes and evaluates a set of result re-ranking algorithms based on various user model representations. Keywords: User Modeling, Personalization, Multilingual Web Search. 1 Introduction Today’s Web is becoming increasingly multilingual, and users are increasingly faced with the challenge of finding documents which resolve their information needs in collections of different languages. Multilingual Information Retrieval (MIR) has been well studied [1], but many unexplored questions remain about modeling users inter- ests for Personalized MIR (PMIR). User modeling for personalized search has been studied, but only on a monolingual level [2]. Taking multilinguality into consideration should change the way user information is modeled to improve PMIR. This research argues that users’ search behavior is influenced by language; this de- pends on the combination of their language capabilities, and the availability of content in various languages. For example, a user may use a certain language when searching for certain types of content (e.g. native language to search for news), yet use another language when searching for other types (e.g. English to search for technical content). Furthermore, a user may click on results of certain languages depending on the type of information sought. This behavior suggests that a user may have multiple personas (facets) in Web search. In support of this argument, this paper proposes and compares multilingual vs. monolingual approaches to adapt search results in PMIR. This in- volves evaluating a set of result re-ranking algorithms combined with various repre- sentations of the user’s interests in both multilingual and monolingual form. The evaluation showed that maintaining user models that account for multilinguality, can achieve significant improvements in retrieval effectiveness.