|
|
|
                                          
|
Return to Session IIIa: XML retrieval
Indexing and ranking are two key factors for efficient and effective XML information retrieval. Inappropriate indexing may result in false negatives and false positives, and improper ranking may lead to low precisions. In this paper, we propose a configurable XML information retrieval system, in which users can configure appropriate index types for XML tags and text contents. Based on users' index configurations, the system transforms XML structures into a compact tree representation, Ctree, and indexes XML text contents. To support XML ranking, we propose the concepts of "weighted term frequency" and "inverted element frequency," where the weight of a term depends on its frequency and location within an XML element as well as its popularity among similar elements in an XML dataset. We evaluate the effectiveness of our system through extensive experiments on the INEX 03 dataset and 30 content and structure (CAS) topics. The experimental results reveal that our system has significantly high precision at low recall regions and achieves the highest average precision (0.3309) as compared with 38 official INEX 03 submissions using the strict evaluation metric.
@inproceedings{1009010, author = {Shaorong Liu and Qinghua Zou and Wesley W. Chu}, title = {Configurable indexing and ranking for XML information retrieval}, booktitle = {SIGIR '04: Proceedings of the 27th annual international conference on Research and development in information retrieval}, year = {2004}, isbn = {1-58113-881-4}, pages = {88--95}, location = {Sheffield, United Kingdom}, doi = {http://doi.acm.org/10.1145/1008992.1009010}, publisher = {ACM Press}, }  ©2005 Association for Computing Machinery |
