A Re-ranking Method Based on Tag-Topic Model

DOI

10.2991/eeic-13.2013.36How to use a DOI?

Keywords

Information Retrieval; Re-ranking; LDA;Tag-topic model

Abstract

We describe a novel approach to improve the accuracy of the IR which combines the tag-topic model with natural language processing. Tag-topic model extended the Latent Dirichlet We describe a novel approach to improve the accuracy of the IR which combines the tag-topic model with natural language processing. We get tag though semantic fingerprint. Tag-topic model extended the Latent Dirichlet Allocation(LDA) model by adding the data set with 901446 documents, and train the model with different number of topics, we can obtain three important distributions: the document-tag distribution, the tag-topic distribution and the topic-word distribution by using the Tag-topic model. Then through the matrix operation we can get the tag-word distribution which quantify the importance of each word of the document. Finally, based on this distribution these documents are re-ranked. Experiments on NTCIR-5 document collection for SLIR(Single Language IR) show that this method achieves an 13.6% and 19.6% improvement comparing to the initial retrieval method without any re-ranking.

Copyright

© 2013, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - Maoyuan Zhang
AU  - Shuiyin Chen
AU  - Fanli He
PY  - 2013/12
DA  - 2013/12
TI  - A Re-ranking Method Based on Tag-Topic Model
BT  - Proceedings of the 3rd International Conference on Electric and Electronics
PB  - Atlantis Press
SP  - 154
EP  - 157
SN  - 1951-6851
UR  - https://doi.org/10.2991/eeic-13.2013.36
DO  - 10.2991/eeic-13.2013.36
ID  - Zhang2013/12
ER  -