Some Experiments with Blind Feedback and Re-ranking for Chinese Information Retrieval

In our formal runs, we have experimented with hybrid-term indexing and bigram indexing because hybrid-term indexing is a more distinct type of indexing strategy for better pooling and because bigram indexing usually gives robust (near) good results. We have also used our pseudo-relevance feedback (PRF) methods. In the informal runs, we have experimented with our previous re-ranking strategy, called title re-ranking. This strategy rewards documents which title terms match with the terms in the title query. Title re-ranking is able to improve the effectiveness performance for both short and long queries when bigram indexing is used. For formal runs, our best relax MAP achieved was 36% and 51% using PRF, for title queries and long queries respectively. For informal runs, our best relax MAP achieved was 43% for title queries and 50% for long queries using both PRF and merging retrieval lists.