This research proposal will focus on retrieval models for Asian languages (monolingual IR) as well as effective merging strategies for both data fusion and collection fusion problems (merging various search results achieved by different IR models). In this research proposal we only consider textual information and exclude any spoken text, images or video sequences.
There are three objectives in carrying out this research: (1) Obtain better knowledge of the relative retrieval effectiveness of various probabilistic models (Okapi), Divergence from Randomness or language models) when using different indexing schemes for Asian languages (monolingual IR). (2) Design and implement effective data fusion (when searching the same document collection using different search systems) and collection merging procedures (searching into distinct document collections). 3) Propose a simple and effective query translation strategy capable of effectively crossing language barriers (bilingual IR). In this case, the request will be written in one given language and document collections in another.
This research is based in part on our participation on two NTCIR evaluation campaigns (http://research.nii.ac.jp/ntcir/) where we obtain very good IR performance.