Information retrieval (IR) systems are inherently temporal. Documents change, indexes acquire new documents, and systems answer or "field" queries differently over time. The vision of this project is to capitalize on this temporality to improve the models used for predicting document relevance. The approach is based on a novel probabilistic framework to allow temporal factors to improve IR effectiveness. The framework situates temporality as a key factor in predicting the document relevance. Initial work focuses on established text retrieval settings, estimating document relevance to keyword queries. However, emerging domains such as social media and volunteer-maintained knowledge bases have an inherent temporality that demands new models. Thus, during the project, research pursues problems of filtering and topic evolution. Methods developed in this project will be experimentally evaluated using standard datasets. The project's expected outcome includes improved models and algorithms for retrieving, filtering, and organizing textual data that arrives incrementally over time.
The project will benefit society in two ways. IR systems play a key role in people?s daily information use. This project will advance the public's ability to negotiate an increasingly complex information landscape, because the expected outcomes will improve search engine technology. Research results will be disseminated primarily via academic conferences and journals. Work will also be stored in an archival institutional repository, affording the public long-term access to results. Progress and general information about the project will be published on the project Web site (http://timer.lis.illinois.edu). The project will provide research experience for students and will advance scientific education. In addition, course materials will be developed to support on-line teaching of information retrieval to non-technical students.
