TREC 2004 Web Track Experiments at CAS-ICT

This report presents CAS-ICT’s experiments on the Mixed query task of the TREC2004 Web track. Our work focused on combining different Web page evidences together to improve the overall retrieval performance. Four kinds of evidences, including body content(C), anchor texts (AT), basic structural information (S0) and extended structural information (S1) were considered for retrieval. Six combination functions were investigated in our experiments. The experimental results show that most functions can improve the retrieval performance. Some heuristic re-ranking techniques were also introduced and tested in the task. No query classification was made during the experiments.