Paper: Clustered Sub-Matrix Singular Value Decomposition

ACL ID N07-2018
Title Clustered Sub-Matrix Singular Value Decomposition
Venue Human Language Technologies
Session Short Paper
Year 2007

This paper presents an alternative algo- rithm based on the singular value decom- position (SVD) that creates vector rep- resentation for linguistic units with re- duced dimensionality. The work was mo- tivated by an application aimed to repre- sent text segments for further processing in a multi-document summarization sys- tem. The algorithm tries to compensate for SVD’s bias towards dominant-topic documents. Our experiments on measur- ing document similarities have shown that the algorithm achieves higher average pre- cision with lower number of dimensions than the baseline algorithms - the SVD and the vector space model.