A framework for the delivery and evaluation of personalised multilingual information retrieval

Abstract

The amount of content provided in different languages on the Web is growing every day. The best answer to a user's query may not necessarily be available in his/her own language, but may reside in the diverse, multilingual corpora of the Web. Furthermore, as Internet penetration increases around the world, the number of multilingual users who seek and interact with information on the Web is also increasing. Personalised Information Retrieval (PIR) aims to help users in satisfying their information needs in a more accurate and less time-consuming manner. The user's search can be personalised by keeping track of his/her personal information and interests, and using this information for query adaptation and result-list adaptation. However, current search personalisation approaches do not pay adequate attention to the effect of multilinguality (of both the users and the content). This has a significant impact on the way PIR services should be delivered and evaluated. The study reported in this thesis argues that users’ searches are influenced by language. For example, a multilingual user, whose native language is not English, may prefer to use his/her native language when seeking certain types of content on the Web (e.g. news), yet choose to use English when seeking other types of content (e.g. technical content). Furthermore, in multilingual search, the user may choose to click on documents originating from certain languages depending on the type of information sought. The study reported in this thesis shows that taking multilinguality into consideration significantly affects PIR. The study therefore introduces the notion of Personalised Multilingual Information Retrieval (PMIR) and proposes a novel framework for the delivery and evaluation of PMIR services. This entailed designing, implementing, and evaluating a set of algorithms for multilingual user modelling, multilingual query adaptation, and multilingual result-list adaptation. Furthermore, this entailed designing and implementing a framework that enables evaluating the compartmentalisation and the combination of PMIR elements. The evaluation shows the success of the multilingual approach to search personalisation and highlights the benefits of the PMIR framework. The methodology undertaken for this study involved: theoretical investigation, an industry case study, user studies, and empirical evaluation. The PMIR framework and the personalisation approaches proposed in this study contribute to the areas of Personalisation and Information Retrieval as they advance research concerning how to model Web users, how to retrieve information that adequately satisfies their information needs, and how to make this information accessible to them.