This year Fudan University takes part in the TREC conference for the second time. We have participated in four tracks of Filtering, Q&A, Web and Video. For filtering, we participate in the sub-task of adaptive and batch filtering. Vector representation and computation are heavily applied in filtering procedure. Four runs have been submitted, which includes one T10SU and one T10F run for adpative filtering, as well as another one T10SU and one T10F run for batch filtering. We have tried many natural language processing techniques in our QA system, including statistical sentence breaking, POS tagging, parsing, name entity tagging, chunking and semantic verification. Various sources of world knowledge are also incorporated, such as WordNet and geographic information. For web retrieval, relevant document set is first created by an extended Boolean retrieval engine, and then reordered according to link information. Four runs with different combination of topic coverage and link information are submitted. On video track, We take part in both of the sub-tasks. In the task of shot boundary detection, we have submitted two runs with different parameters. In the task of video retrieval, we have submitted the results of 17 topics among all the topics.
[1]
Wessel Kraaij,et al.
TNO-UT at TREC-9: How Different are Web Documents?
,
2000,
TREC.
[2]
Dekang Lin,et al.
A dependency-based method for evaluating broad-coverage parsers
,
1995,
Natural Language Engineering.
[3]
Marco Ceccarelli,et al.
Automation of systems enabling search on stored video data
,
1997,
Electronic Imaging.
[4]
Sanda M. Harabagiu,et al.
FALCON: Boosting Knowledge for Answer Engines
,
2000,
TREC.
[5]
Zhu Xing.
AN AUTOMATIC THRESHOLD DETECTION METHOD IN VIDEO SHOT SEGMENTATION
,
2000
.
[6]
Sanda M. Harabagiu,et al.
A Parallel System for Textual Inference
,
1999
.
[7]
Charles L. A. Clarke,et al.
Deriving Very Short Queries for High Precision and Recall (MultiText Experiments for TREC-7)
,
1998,
TREC.