DMCA
Learning Discriminative Projections for Text Similarity Measures
Cached
Download Links
| Citations: | 19 - 5 self |
BibTeX
@MISC{Yih_learningdiscriminative,
author = {Wen-tau Yih and Kristina Toutanova and John C. Platt and Christopher Meek},
title = {Learning Discriminative Projections for Text Similarity Measures},
year = {}
}
Share
 |
 |
 |
 |
||
OpenURL
Abstract
Traditional text similarity measures consider each term similar only to itself and do not model semantic relatedness of terms. We propose a novel discriminative training method that projects the raw term vectors into a common, low-dimensional vector space. Our approach operates by finding the optimal matrix to minimize the loss of the pre-selected similarity function (e.g., cosine) of the projected vectors, and is able to efficiently handle a large number of training examples in the highdimensional space. Evaluated on two very different tasks, cross-lingual document retrieval and ad relevance measure, our method not only outperforms existing state-of-the-art approaches, but also achieves high accuracy at low dimensions and is thus more efficient. 1
Keyphrases
discriminative projection text similarity measure projected vector raw term vector ad relevance measure novel discriminative training method pre-selected similarity function cross-lingual document retrieval traditional text similarity measure state-of-the-art approach low dimension low-dimensional vector space large number high accuracy optimal matrix different task highdimensional space semantic relatedness
