• Home
• Research
• Publications
• Projects
• Contact

I am currently working (partly) on my PhD. My research is focused on the so-called problem of attribute selection in conversational search systems. In basic terms, given a system capable of having a dialogue with the user as the means to progressively constrain a search space, what are the right questions to ask at every step in order to produce a fast and enjoyable search? This simple formulation hides the complexity of the problem, which touches on a number of areas:

• Dialogue Management
  This is one of the key problems of conversational interfaces: how to build a dialogue manager that provides a convincing interlocutor to the user. In particular, I have been investigating the special features necessary for dialogue management in the context of in-home environments.
• Information Theory
  An information-theoretical analysis of the search space is necessary to determine which attributes provide more even splits and potentially faster searches. The Maximum Entropy Method is used for this problem.
• User Modelling
  Keeping a profile of the user is necessary to tailor the search process to the particularities of each user. I am investigating the use of IBL (Instance-Based Learning) to store user profiles, and also the application of user stereotypes to speed up the learning process and to avoid the blank slate problem.
• Machine Learning
  I have a general interest in machine learning, and in particular, in algorithms for enabling the system to adapt and to learn about the user. IBL (Instance-Based Learning) is one example, but reinforcement learning is also used.
• Simulated Users
  Having a simulated user which captured the essential features of a real person would be a huge benefit to research in conversational interfaces. Alas, the construction of simulated users is a difficult problem, filled with potential dangers. I am investigating how some of these problems can be overcome.
• Multi-Agent Systems
  I am especially interested in the problem of peer-based rating in multi-agent communities. As far as my research is concerned, this problem manifests itself in the rewarding/punishment of the multiple profile heuristics. However, this problem also manifests itself in other domains, such as in the realm of online communities.
• Browsing Assistants
  Browsing assistants provide the user with an intelligent guide to navigate through large volumes of information. Combined with the conversational search process, these could provide the means the tackle content overload situations.
• Multimodal Interaction
  Dialogue does not necessarily mean speech. The integration of multiple modalities will allow the user to conduct the search process using, for example, holiday souvenirs as shortcuts to search space constraints.