Site visit in Helsinki on 5 May 2003

The coordination unit made a site visit at the Advanced Research Unit of HIIT, NNRC, and CKIR in Ruoholahti, Helsinki on 5 May 2003. Academy Research Fellows Samuel Kaski and Petri Myllymäki and Docent Ilpo Kojo together with their research groups presented their research in their project Proactive Information Retrieval by Adaptive Models of Users' Attention and Interests (PRIMA).

The general goal of the project is to develop models underlying true personal assistants, models that learn from the actions of people to model their intentions and actions, and use the models for disambiguating the users' vague commands and anticipating their actions. A specific goal of the project concerns information retrieval: the project tries to infer what is relevant or interesting based on eye movements, document content, interest history and collaborative filtering.

Feedback in information retrieval is usually both necessary and very helpful. Giving explicit feedback is, however, costly and laborious. Implicit feedback would enhance many applications. Implicit data is also available in different forms, such as eye movements of the user, the user's history in the search session and the user's general interest profile. Collaborative filtering can be used to find relevant information, as well. The project's idea is to combine such rich implicit feedback information with modelling.

The Vision Team at CIKR has performed a pilot study on short text documents, where the users are asked to find certain topics in a list of relevant and irrelevant headlines. A head-mounted gaze tracking system (SMI iView) is used to track eye movements. During the task, eye fixations on certain parts of the text are observed and measured. Preliminary results show that there seems to be a significant difference between fixation times if the relevant headlines are found in the first half of the list compared to if they are found in the second half or not at all. The number of fixations per word is also significantly higher when there is no answer as compared to when there is at least one answer (no matter where in the list). The group will continue with the experiments. One goal is also to determine personal search strategies from the search paths (i.e. that paths the eyes are moved over the text).

The project team will build models to estimate how relevant documents are based on the measured eye movements. Models will also be developed for inferring not only which documents are relevant but also what parts of a document are relevant. Also models of interest profiles will be used in anticipating which documents are relevant. Collaborative filtering, especially probabilistic modelling of co-occurrence data plays an important role in these profiles. In the end, all models should be combined to form a single statistical model, which is then analyzed from the cognitive science point of view.

Results
The results of the project will be demonstrated in an application (applications) that will be able to perform more accurate and proactive information retrieval, that will be able to suggest new interesting documents while the user is reading or browsing information, that filters documents (and emails) and that is able to highlight interesting parts in a document.

More information
For more information, please contact Academy Research Fellow Samuel Kaski at the Helsinki University of Technology or see the home page of the group at http://www.cis.hut.fi/projects/mi/prima/. See also the project's web page at this site.

The research consortium consists of the following units:

• Helsinki University of Technology, Laboratory of Computer and Information Science, Neural Networks Research Centre (NNRC): Academy Research Fellow Samuel Kaski, researchers Jarkko Salojärvi, Eerika Savia, affiliated researchers Kai Puolamäki, Janne Sinkkonen
• Helsinki School of Economics, Center for Knowledge and Innovative Research, Vision Team: Docent Ilpo Kojo, researcher Jaana Simola
• Helsinki Institute for Information Technology, Complex Systems Computation Group: Academy Research Fellow Petri Myllymäki, researchers Miikka Miettinen, Ville Tuulos

Greger Lindén
Programme Coordinator
Greger.Linden@cs.helsinki.fi