@inproceedings{gabor-etal-2017-exploring,
title = "Exploring Vector Spaces for Semantic Relations",
author = {G{\'a}bor, Kata and
Zargayouna, Ha{\"\i}fa and
Tellier, Isabelle and
Buscaldi, Davide and
Charnois, Thierry},
booktitle = "Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing",
month = sep,
year = "2017",
address = "Copenhagen, Denmark",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/D17-1193",
doi = "10.18653/v1/D17-1193",
pages = "1814--1823",
abstract = "Word embeddings are used with success for a variety of tasks involving lexical semantic similarities between individual words. Using unsupervised methods and just cosine similarity, encouraging results were obtained for analogical similarities. In this paper, we explore the potential of pre-trained word embeddings to identify generic types of semantic relations in an unsupervised experiment. We propose a new relational similarity measure based on the combination of word2vec{'}s CBOW input and output vectors which outperforms concurrent vector representations, when used for unsupervised clustering on SemEval 2010 Relation Classification data.",
}
1. What is it?
The authors proposed a method for computing similarity.
2. What is amazing compared to previous works?
They proposed to add, multiplicate, and use context vectors for computing similarities.
3. Where is the key to technologies and techniques?
In computing a similarity between words (a1, a2) and (b1, b2) they proposed some methods.
Baseline: only use a1 and b1 vectors.
concatenate: concate a1, a2 and b1, b2.
addition: add sim(a1, b1) and sim(a2, b2).
multiplication: multiplicate sim(a1, b1) and sim(a2, b2).
difference:
additive in-out: use not only word vectors a1, a2, b1, and b2 but also their context vectors.
multiplicative in-out:
4. How did evaluate it?
They used their similarity methods(3.) as the input for the clustering model.
training word2vec CBoW model: ukWaC corpus
clustering: Cluto's clustering, and UPGMA criterion function.
a1da4
commented
4 years ago
0. Paper
@inproceedings{gabor-etal-2017-exploring, title = "Exploring Vector Spaces for Semantic Relations", author = {G{\'a}bor, Kata and Zargayouna, Ha{\"\i}fa and Tellier, Isabelle and Buscaldi, Davide and Charnois, Thierry}, booktitle = "Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing", month = sep, year = "2017", address = "Copenhagen, Denmark", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/D17-1193", doi = "10.18653/v1/D17-1193", pages = "1814--1823", abstract = "Word embeddings are used with success for a variety of tasks involving lexical semantic similarities between individual words. Using unsupervised methods and just cosine similarity, encouraging results were obtained for analogical similarities. In this paper, we explore the potential of pre-trained word embeddings to identify generic types of semantic relations in an unsupervised experiment. We propose a new relational similarity measure based on the combination of word2vec{'}s CBOW input and output vectors which outperforms concurrent vector representations, when used for unsupervised clustering on SemEval 2010 Relation Classification data.", }
1. What is it?
The authors proposed a method for computing similarity.
2. What is amazing compared to previous works?
They proposed to add, multiplicate, and use context vectors for computing similarities.
3. Where is the key to technologies and techniques?
In computing a similarity between words (a1, a2) and (b1, b2) they proposed some methods.
Baseline: only use a1 and b1 vectors.
concatenate: concate a1, a2 and b1, b2.
addition: add sim(a1, b1) and sim(a2, b2).
multiplication: multiplicate sim(a1, b1) and sim(a2, b2).
difference:
additive in-out: use not only word vectors a1, a2, b1, and b2 but also their context vectors.
multiplicative in-out:
4. How did evaluate it?
They used their similarity methods(3.) as the input for the clustering model.
As below, in-out methods outperform the baseline.
5. Is there a discussion?
6. Which paper should read next?
Using context vectors is based on this paper: [A Simple Word Embedding Model for Lexical Substitution]