I am interested in all aspects of information retrieval (theory, experimentation, evaluation and applications) in the textual and multimedia domain. My research focuses around the following three themes: (i) Adaptive and personalized search systems; (ii) Multimodal interaction for information retrieval; (iii) Multimedia mining and search.

Due to the developments in technology and computing, and the way people need to use information for their day to day life, the retrieval problem has evolved into delivering information to the user based on his/her context or situation. Interaction is moving away from people to computer to people to environment. Information search is moving away from searching a well collected repository (such as a scholarly collection) to searching heterogeneous collections (like World Wide Web, personal desktop data, organizational data sets). Information need arises in various situations and the relevance of that information depends on the context. My research focuses on solving practical problems of importance and is largely shaped by the interaction with companies such as Kodak, Sharp, Yahoo! Research, and other European companies dealing with media industry.

A number of research issues need to be addressed before solving the retrieval problem. They are: context modelling; adaptive retrieval models; media mining and retrieval; evaluation methodologies. I investigate these issues under the following three themes.

Adaptive and personalized search systems

Three important stages in an information retrieval process are: query formulation; the matching process and the result presentation. These three aspects are inter-related and we addressed these problems from the perspective of the searchers. We look at this issue from personalising to userís current information need. This is done by mining user interaction data in order to infer user needs. We have investigated novel forms of result presentation strategies, based on various forms of surrogates. In the context of web search retrieval, we have looked into the issue of presenting information from fine to coarse granularities. The enhanced interaction based on these factors was mined to infer user interests. We have looked into the problems of relevance feedback techniques and devised a scheme based on implicit feedback. The approach exploited user interaction data to devise a technique to infer user interests.

Evaluation: One of the hindering blocks in studies like above is the issues with evaluation methodologies. User cantered evaluations are paramount but at the same time expensive and time consuming. In order to benchmark various adaptive search models one has to revoke tedious trial and error approach. We have addressed this problem by devising a simulated evaluation methodology. This is based on the classical evaluation methodology but extending to incorporate adaptive and personalized retrieval. In addition to benchmarking various models, it also allows us to conceptually compare various retrieval interfaces.

Context Modelling: These advances were complemented by devising methodologies to analyze user experimental data. We have created a methodology to analyze user experimental data using machine learning techniques. The objective is to identify contextual factors that affect the relevance of documents. We have studied the effect of task complexity, topics, interface features, search stage and search experience. This is a methodological study and the tools we are developing will be useful to analyze logged data from interactive experimental study.

Current Research Projects

Personalised information retrieval: Understanding long-term user needs and pro-actively fetching information is an important activity. The objective is on understanding and capturing the evolving needs of users in order to build personalised retrieval systems. Adaptive search models are an issue that need to be addressed in order to solve this problem. The following sub-projects contribute to my activity.
  • News video modelling and retrieval
  • User friendly systems for personal multimedia data management
  • Intelligent information modelling and tracking systems
Context Modelling:
  • Collaborative search systems
  • Interactive search systems (Text, TREC VID etc.)
  • Adaptive Test Collections ( for context-sensitive retrieval)
  • Management Information Systems for the web

Multimodal interaction for information retrieval

One of the fundamental problems in information retrieval is understanding relevance. Given that information seeking is becoming situation sensitive and multi-modal in nature, we investigate multi-modal information seeking process. We address this problem from two perspectives: One is to find relationship of relevance with modality. Second is to develop innovative multimodal search systems.

User feedback is considered to be a critical element in the information seeking process, especially in relation to relevance assessment. Current feedback techniques determine content relevance with respect to the cognitive and situational levels of interaction that occurs between the user and the retrieval system. However, apart from real-life problems and information objects, users interact with intentions, motivations and feelings, which can be seen as critical aspects of cognition and decision-making. We explore the role of emotions in the information seeking process. Results of our initial study show that the latter not only interweave with different physiological, psychological and cognitive processes, but also form distinctive patterns, according to specific task, and according to specific user.

Multi-modality is not explored in information retrieval. Retrieval can be based on any form of modality: speech, gesture, eye-tracking. In this work we investigate the role of multi-modality in information seeking. We build exemplar multi-modal search systems and study their effectiveness in information seeking.

Current Research Projects

  • Affective search systems using emotion analysis and eye tracking. If we know the relationship between affective features and relevance, we will be able to develop better personalised retrieval models. This research is based on the Multimodal interaction (MIAUCE project) and is by investigating search models that use emotional features in information seeking.
  • Multi-modal search systems for everyday use: Interactive search systems based on everyday objects of use.
  • Affective summarization

Multimedia mining and search.

Building innovative interfaces: We continue our work on multimedia retrieval by building novel interactive interfaces which helps the user to understand the relevance of retrieved items and also facilitate us to mine user interaction data in order to build adaptive search systems. Such high-level of interaction will allow us to characterise and model user context.

Retrieval and mining models: Retrieval model is the basic block that facilitates effective retrieval. However, in multimedia retrieval role of retrieval models are not studied. We will continue to explore the development of retrieval models for multimedia data.

Annotation issues: Semantic annotation is key to multimedia management. We have studied the role of graph models in semantic annotation. We will continue this investigation in order to develop techniques for semi-automatic and automatics annotation techniques.

Current Research Projects

  • News video modelling and retrieval
  • User friendly systems for personal multimedia data management
  • Novel interfaces for multimedia search
  • Multimedia mining models
  • Multimedia search models
  • Event mining and detection
  • Multimedia retrieval models







Last modified: Thu Aug 30 17:39:57 GMT 2007