Gradual Learning of Matrix-Space Models of Language for Sentiment Analysis

Shima AsaadiShima Asaadi, Sebastian RudolphSebastian Rudolph

Download

Shima Asaadi, Sebastian Rudolph
Gradual Learning of Matrix-Space Models of Language for Sentiment Analysis
, August 2017

Details
Bibtex

KurzfassungAbstract
Learning word representations to capture the semantics and compositionality of language has received much research interest in natural language processing. Beyond the popular vector space models, matrix representations for words have been proposed, since then, matrix multiplication can serve as natural composition operation. In this work, we investigate the problem of learning matrix representations of words. We present a learning approach for compositional matrix-space models for the task of sentiment analysis. We show that our approach, which learns the matrices gradually in two steps, outperforms other approaches and a gradient-descent baseline in terms of quality and computational cost.
Projekt:Project: QuantLA
Forschungsgruppe:Research Group: Computational LogicComputational Logic

@InProceedings{asaadi-rudolph:2017:RepL4NLP,
  author    = {Asaadi, Shima  and  Rudolph, Sebastian},
  title     = {Gradual Learning of Matrix-Space Models of Language for Sentiment Analysis},
  booktitle = {Proceedings of the 2nd Workshop on Representation Learning for NLP},
  month     = {August},
  year      = {2017},
  address   = {Vancouver, Canada},
  publisher = {Association for Computational Linguistics},
  pages     = {178--185},
  abstract  = {Learning word representations to capture the semantics and compositionality of
	language has received much research interest in natural language processing.
	Beyond the popular vector space models, matrix representations for words have
	been proposed, since then, matrix multiplication can serve as natural
	composition operation. In this work, we investigate the problem of learning
	matrix representations of words. We present a learning approach for
	compositional matrix-space models for the task of sentiment analysis. We show
	that our approach, which learns the matrices gradually in two steps,
	outperforms other approaches and a gradient-descent baseline in terms of
	quality and computational cost.},
  url       = {http://www.aclweb.org/anthology/W17-2621}
}