An Investigation of User Behaviour Consistency for Context-Aware Information Retrieval Systems

An Investigation of User Behaviour Consistency for Context-Aware Information Retrieval Systems

Adam Grzywaczewski, Rahat Iqbal, Anne James, John Halloran
DOI: 10.4018/japuc.2009100105
(Individual Articles)
No Current Special Offers


Users interact with the Internet in dynamic environments that require the IR system to be context aware. Modern IR systems take advantage of user location, browsing history or previous interaction patterns, but a significant number of contextual factors that impact the user information retrieval process are not yet available. Parameters like the emotional state of the user and user domain expertise affect the user experience significantly but are not understood by IR systems. This article presents results of a user study that simplifies the way context in IR and its role in the systems’ efficiency is perceived. The study supports the hypothesis that the number of user interaction contexts and the problems that a particular user is trying to solve is related to lifestyle. Therefore, the IR system’s perception of the interaction context can be reduced to a finite set of frequent user interactions.
Article Preview


For the last few years we have been moving away from traditional desktop computing to a ubiquitous environment where information needs change dynamically. The internet is no longer a tool for academics only, but is being used for even the simplest every day activities. It is available not only through desktop computers but also semi-mobile laptops, mobile phones, game-consoles and many other Internet enable devices. The broader range of Internet applications creates a greater number of information needs which in modern search engines have to be translated into search queries. Expressing our thoughts precisely is very difficult especially if we consider the cultural, linguistic, emotional and situational contexts in which they are placed. Every human being is unique and that makes keyword-based representation, which search engines and Information Retrieval (IR) systems use, very limiting (Ferreira & Atkinson, 2005). Even for experienced users, formulating a good textual query in some situations is a challenging task (Ferreira & Atkinson, 2005). Therefore it is hypothesised that a computer system partially capable of understanding the user’s needs can significantly facilitate Internet search.

Current research aims at understanding a user’s intentions in multiple ways. Techniques such as relevance feedback, user profiling or personalisation are used to improve the context awareness of IR applications (Morita & Shinoda, 1994; Nichols, 1997; Spink & Losee, 1996; Oard & Kim, 1998; Kelly &Teevan, 2003). The most effective algorithms use implicit feedback to improve already existing search and ranking software and has been shown to increase the efficiency of search by up to 31% (Agichtein, Brill & Dumais, 2006; Agichtein, Brill, Dumais & Ragno, 2006) There is also much research that focusses on the validity of the approach in order to prove the theoretical potential of the method (Teevan, Dumais & Horvitz, 2007; Joachims, Granka, Pan, Hembrooke & Gay, 2005). Most modern approaches focus on estimating relevance based on universal patterns of human behaviour. Some approaches are based on parameters which do universally represent relevance, such as reading time (Kelly & Belkin, 2004; Claypool, Brown, Le, & Waseda, 2001) or activity on the page (Claypool, Brown, Le, & Waseda, 2001; Fox, Karnawat, Mydland, Dumais & White, 2005).

Existing IR systems provide some promising results but they have a number of limitations that are listed below.

  • Even though factors such as reading time can become very good estimators of relevance and the usage of click through information for pair wise estimation produces promising results (Joachims, 2002; Joachims, Granka, Pan, Hembrooke & Gay, 2005) recent research suggests high variability of human behaviour across various components that are the part of the search process.

  • Aspects such as domain expertise (Marshall & Byrd, 1998; White, Dumais & Teevan, 2009; Teevan, Dumais & Horvitz, 2007; Teevan, Alvarado, Ackerman & Karger, 2004), language and cultural background (Kralisch & Berendt, 2005; Kralisch & Berendt, 2004a), search experience (Aula & Käki, 2003; Zhang, Anghelescu & Yuan, 2005; Hsieh-Yee, 1993; Ylikoski, 2003; Hölscher & Strube, 2000; White, Ruthven & Jose, 2005), task the user performs (Turpin & Scholer, 2006; Teevan, Dumais & Horvitz, 2005; Kelly &Teevan, 2003; Teevan, Alvarado, Ackerman & Karger, 2004; Kelly & Belkin, 2004), personal character of the user (Heinstrom, 2003; Heinström, 2007), time of interaction (Lee & Park, 2006), user’s subjective perception of relevance (Spink & Losee, 1996; Krug, 2006), environmental settings, stress, time pressure, physical and emotional state and many others (Poblete & Baeza-Yates, 2008; Zhang & Koren, 2007; Ferrara, 2008; Morrison & Zander, 2008) become a source of a significant efficiency decrease in such applications.

  • As found by Dou et al. (Dou, Song, R., and Wen, 2007), personalisation can indeed significantly improve the quality of user search experience but only for some queries whereas for others not only does it fail to improve it, it can also negatively affect user performance.

Complete Article List

Search this Journal:
Open Access Articles: Forthcoming
Volume 11: 4 Issues (2019)
Volume 10: 4 Issues (2018)
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing