Yu Zhiwen

An adaptive in-vehicle multimedia recommender for group users

Rapid advances of mobile communication technologies enable todays vehicle to receive overabundant multimedia content from different broadcasters. In addition, the in-vehicle multimedia content is often viewed by a group of passengers traveling in buses, trains, and airplanes. Adaptive vehicular multimedia systems are required to provide personalized multimedia content for group users by taking care of the majoritys preferences. This paper proposes an adaptive in-vehicle multimedia recommendation system serving for group users. It presents the recommender system architecture. The main technologies: user identification and profile aggregation, user profile merging algorithm, and recommendation strategy are described. This paper also presents the evaluation methodology and experimental results.

Design, implementation, and evaluation of an agent-based adaptive program personalization system

The abundance of DTV (digital television) programs precipitates a need for new tools to help people find interesting programs and personalize TV viewing experience. We describe the design, implementation, and evaluation of an agent based adaptive program personalization system, which is designed to assist users by adapting to their personal preferences. We provide the architecture of the system, details its components, and presents the inter agent communication mechanism for the multiagent system. In our system, the feature representation and similarity measurement is based on VSM (vector space model), while the preference knowledge learning algorithm is using relevance feedback. We describe the prototype implementation, which has been implemented on PC/Linux. Thirdly, it describes our experiments and performance evaluation on the prototype. The experiment results are encouraging, which shows that the system proposed is useful to TV consumers. Finally, we summarizes the conclusions, and ends with directions for future work.

A hybrid filter for program personalization

The current exponential growth of DTV programs precipitates a need for new filtering techniques to provide program personalization for people. Several filtering methods, notably Boolean filtering and similarity-based filtering, have been proposed in information filtering. This paper first briefly describes the architecture of program personalization, XML based metadata representation, and discusses Boolean and similarity-based filtering, as well as their advantages and disadvantages. Then a new hybrid filter is proposed. The hybrid filter consists of VSM based filtering, which is a similarity-based filtering, and XML query filtering, which is a Boolean filtering. This hybrid filter is effective and efficient for program personalization in future DTV with local storage.