Difference between revisions of "SimLex-999 (State of the art)"

From ACL Wiki
Jump to navigation Jump to search
m
 
(2 intermediate revisions by 2 users not shown)
Line 8: Line 8:
 
! Algorithm !! Reference for algorithm !! Reference for reported results  !! Type !! Spearman's rho !! Pearson's r !! Notes
 
! Algorithm !! Reference for algorithm !! Reference for reported results  !! Type !! Spearman's rho !! Pearson's r !! Notes
 
|-
 
|-
| Neural language model
+
| Re16
| Collobert & Weston (2008)<ref>R. Collobert and J. Weston. 2008. A unified architecture for natural language pro- cessing: Deep neural networks with multitask learning. In International Conference on Machine Learn- ing, ICML.</ref>
+
| Recski et al. (2016)<ref name=recski16>Recski, G., Iklódi, E., Pajkossy, K., & Kornai, A. (2016). [https://www.aclweb.org/anthology/W16-1622 Measuring semantic similarity of words using concept networks]. In: Proceedings of the 1st Workshop on Representation Learning for NLP, pp. 193-200.</ref>
| Hill et al. (2014a)<ref name=simlex/>  
+
| Recski et al. (2016)<ref name=recski16/>
| Distributional || 0.268 || - || Trained on Wikipedia
+
| Hybrid || 0.76 || -
 +
|-
 +
| SVR4
 +
| Banjade et al. (2015)<ref name=lemontea/>
 +
| Banjade et al. (2015)<ref name=lemontea/>
 +
| Combined || 0.642 || 0.658
 +
|-
 +
| Do19-hybrid
 +
| Dobó (2019)<ref name=dobo19>Dobó, A. (2019). [http://doktori.bibl.u-szeged.hu/10120/1/AndrasDoboThesis2019.pdf A comprehensive analysis of the parameters in the creation and comparison of feature vectors in distributional semantic models for multiple languages]. University of Szeged. [https://github.com/doboandras/dsm-parameter-analysis GitHub repository]</ref>
 +
| Dobó (2019)<ref name=dobo19/>
 +
| Hybrid || 0.621 || 0.481
 +
|-
 +
| Sp17
 +
| Speer et al. (2017)<ref name=speer17>Speer, R., Chin, J., and Havasi, C. (2017). [https://www.aaai.org/ocs/index.php/AAAI/AAAI17/paper/download/14972/14051 Conceptnet 5.5: An open multilingual graph of general knowledge]. AAAI-17, pp. 4444-4451.</ref>
 +
| Dobó (2019)<ref name=dobo19/>
 +
| Hybrid || 0.616 || 0.634
 +
|-
 +
| joint(SP+,skip-gram)
 +
| Schwartz et al. (2015)<ref name=spplus>Schwartz, R., Reichart, Roi, Rappoport, A. (2015). Symmetric Pattern Based Word Embeddings for Improved Word Similarity Prediction, CoNLL 2015.</ref>
 +
| Schwartz et al. (2015)<ref name=spplus/>
 +
| Distributional || 0.56 || - || Trained on word2vec corpus, best results for pure distributional model.
 +
|-
 +
| UMBC
 +
| Han et al. (2013)<ref>Han, L., Kashyap, A., Finin, T., Mayfield, J., Weese, J.: UMBC EBIQUITY-CORE: Semantic textual similarity systems. In: Proceedings of the Second Joint Conference on Lexical and Computational Semantics, vol. 1, pp. 44–52 (2013)</ref>
 +
| Banjade et al. (2015)<ref name=lemontea/>
 +
| || 0.558 || 0.557 || without using POS information
 +
|-
 +
| SP+
 +
| Schwartz et al. (2015)<ref name=spplus>Schwartz, R., Reichart, Roi, Rappoport, A. (2015). Symmetric Pattern Based Word Embeddings for Improved Word Similarity Prediction, CoNLL 2015.</ref>
 +
| Schwartz et al. (2015)<ref name=spplus/>
 +
| Distributional || 0.52 || -
 +
|-
 +
| RNNenc
 +
| Hill et al. (2014b)<ref name=rnnenc>Hill, F., Cho, K., Jean, S., Devin, C., & Bengio, Y. (2014b). Not All Neural Embeddings are Born Equal, 1–5.</ref>
 +
| Hill et al. (2014b)<ref name=rnnenc/>
 +
| Distributional, multilingual || 0.52 || -
 
|-
 
|-
| Neural language model with global context
+
| Sa18
| Huang et al. (2012)<ref>Eric H Huang, Richard Socher, Christopher D Manning, and Andrew Y Ng. 2012. Improving word representations via global context and multiple word prototypes. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 873–882. Association for Computational Linguistics.</ref>
+
| Salle et al. (2018)<ref name=salle18>Salle A., Idiart M., and Villavicencio A. (2018). [https://github.com/alexandres/lexvec/blob/master/README.md LexVec]</ref>  
| Hill et al. (2014a)<ref name=simlex/>  
+
| Dobó (2019)<ref name=dobo19/>
| Distributional || 0.098 || - || Trained on Wikipedia
+
| Distributional || 0.417 || 0.426
 
|-
 
|-
 
| Word2vec
 
| Word2vec
Line 22: Line 57:
 
| Hill et al. (2014a)<ref name=simlex>Hill, F., Reichart, R., & Korhonen, A. (2014a). SimLex-999: Evaluating Semantic Models with (Genuine) Similarity Estimation. Computation and Language.</ref>
 
| Hill et al. (2014a)<ref name=simlex>Hill, F., Reichart, R., & Korhonen, A. (2014a). SimLex-999: Evaluating Semantic Models with (Genuine) Similarity Estimation. Computation and Language.</ref>
 
| Distributional || 0.414 || - || Trained on Wikipedia
 
| Distributional || 0.414 || - || Trained on Wikipedia
 +
|-
 +
| Pe14
 +
| Pennington et al. (2014)<ref name=pennington14>Pennington, J., Socher, R., and Manning, C. (2014). [https://www.aclweb.org/anthology/D14-1162 Glove: Global vectors for word representation]. EMNLP 2014, pp. 1532-1543.</ref>
 +
| Dobó (2019)<ref name=dobo19/>
 +
| Distributional || 0.406 || 0.433
 
|-
 
|-
 
| Lesk
 
| Lesk
Line 28: Line 68:
 
| || 0.404 || 0.347
 
| || 0.404 || 0.347
 
|-
 
|-
| UMBC
+
| Do19-corpus
| Han et al. (2013)<ref>Han, L., Kashyap, A., Finin, T., Mayfield, J., Weese, J.: UMBC EBIQUITY-CORE: Semantic textual similarity systems. In: Proceedings of the Second Joint Conference on Lexical and Computational Semantics, vol. 1, pp. 44–52 (2013)</ref>
+
| Dobó (2019)<ref name=dobo19/>
| Banjade et al. (2015)<ref name=lemontea/>
+
| Dobó (2019)<ref name=dobo19/>
| || 0.558 || 0.557 || without using POS information
+
| Distributional || 0.393 || 0.401
|-
 
| SVR4
 
| Banjade et al. (2015)<ref name=lemontea/>
 
| Banjade et al. (2015)<ref name=lemontea/>
 
| Combined || 0.642 || 0.658
 
 
|-
 
|-
 
| ESA
 
| ESA
Line 43: Line 78:
 
| || 0.271 || 0.145
 
| || 0.271 || 0.145
 
|-
 
|-
| RNNenc
+
| Neural language model
| Hill et al. (2014b)<ref name=rnnenc>Hill, F., Cho, K., Jean, S., Devin, C., & Bengio, Y. (2014b). Not All Neural Embeddings are Born Equal, 1–5.</ref>
+
| Collobert & Weston (2008)<ref>R. Collobert and J. Weston. 2008. A unified architecture for natural language pro- cessing: Deep neural networks with multitask learning. In International Conference on Machine Learn- ing, ICML.</ref>
| Hill et al. (2014b)<ref name=rnnenc/>
+
| Hill et al. (2014a)<ref name=simlex/>  
| Distributional, multilingual || 0.52 || -
+
| Distributional || 0.268 || - || Trained on Wikipedia
 +
|-
 +
| Neural language model with global context
 +
| Huang et al. (2012)<ref>Eric H Huang, Richard Socher, Christopher D Manning, and Andrew Y Ng. 2012. Improving word representations via global context and multiple word prototypes. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 873–882. Association for Computational Linguistics.</ref>
 +
| Hill et al. (2014a)<ref name=simlex/>  
 +
| Distributional || 0.098 || - || Trained on Wikipedia
 
|}
 
|}
  

Latest revision as of 18:34, 15 September 2019

SimLex-999 aims at a cleaner benchmark of similarity (but not relatedness). Pairs of words were chosen to represent different ranges of similarity and with either high or low association. Subjects were instructed to differentiate between similarity and relatedness and rate regarding the former only.

See also: Similarity (State of the art), Similar-Associated-Both Test Collection (State of the art).


Algorithm Reference for algorithm Reference for reported results Type Spearman's rho Pearson's r Notes
Re16 Recski et al. (2016)[1] Recski et al. (2016)[1] Hybrid 0.76 -
SVR4 Banjade et al. (2015)[2] Banjade et al. (2015)[2] Combined 0.642 0.658
Do19-hybrid Dobó (2019)[3] Dobó (2019)[3] Hybrid 0.621 0.481
Sp17 Speer et al. (2017)[4] Dobó (2019)[3] Hybrid 0.616 0.634
joint(SP+,skip-gram) Schwartz et al. (2015)[5] Schwartz et al. (2015)[5] Distributional 0.56 - Trained on word2vec corpus, best results for pure distributional model.
UMBC Han et al. (2013)[6] Banjade et al. (2015)[2] 0.558 0.557 without using POS information
SP+ Schwartz et al. (2015)[5] Schwartz et al. (2015)[5] Distributional 0.52 -
RNNenc Hill et al. (2014b)[7] Hill et al. (2014b)[7] Distributional, multilingual 0.52 -
Sa18 Salle et al. (2018)[8] Dobó (2019)[3] Distributional 0.417 0.426
Word2vec Mikolov et al. (2013)[9] Hill et al. (2014a)[10] Distributional 0.414 - Trained on Wikipedia
Pe14 Pennington et al. (2014)[11] Dobó (2019)[3] Distributional 0.406 0.433
Lesk Banjade et al. (2015)[2] 0.404 0.347
Do19-corpus Dobó (2019)[3] Dobó (2019)[3] Distributional 0.393 0.401
ESA Banjade et al. (2015)[2] 0.271 0.145
Neural language model Collobert & Weston (2008)[12] Hill et al. (2014a)[10] Distributional 0.268 - Trained on Wikipedia
Neural language model with global context Huang et al. (2012)[13] Hill et al. (2014a)[10] Distributional 0.098 - Trained on Wikipedia


References

  1. 1.0 1.1 Recski, G., Iklódi, E., Pajkossy, K., & Kornai, A. (2016). Measuring semantic similarity of words using concept networks. In: Proceedings of the 1st Workshop on Representation Learning for NLP, pp. 193-200.
  2. 2.0 2.1 2.2 2.3 2.4 Banjade, R., Maharjan, N., Niraula, N., Rus, V., & Gautam, D. (2015). Lemon and Tea Are Not Similar: Measuring Word-to-Word Similarity by Combining Different Methods. Computational Linguistics and Intelligent Text Processing, 9041, 335–346. doi:10.1007/978-3-319-18111-0_25
  3. 3.0 3.1 3.2 3.3 3.4 3.5 3.6 Dobó, A. (2019). A comprehensive analysis of the parameters in the creation and comparison of feature vectors in distributional semantic models for multiple languages. University of Szeged. GitHub repository
  4. Speer, R., Chin, J., and Havasi, C. (2017). Conceptnet 5.5: An open multilingual graph of general knowledge. AAAI-17, pp. 4444-4451.
  5. 5.0 5.1 5.2 5.3 Schwartz, R., Reichart, Roi, Rappoport, A. (2015). Symmetric Pattern Based Word Embeddings for Improved Word Similarity Prediction, CoNLL 2015.
  6. Han, L., Kashyap, A., Finin, T., Mayfield, J., Weese, J.: UMBC EBIQUITY-CORE: Semantic textual similarity systems. In: Proceedings of the Second Joint Conference on Lexical and Computational Semantics, vol. 1, pp. 44–52 (2013)
  7. 7.0 7.1 Hill, F., Cho, K., Jean, S., Devin, C., & Bengio, Y. (2014b). Not All Neural Embeddings are Born Equal, 1–5.
  8. Salle A., Idiart M., and Villavicencio A. (2018). LexVec
  9. Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. 2013. Efficient estimation of word representations in vector space. In Proceedings of International Conference of Learning Representations, Scottsdale, Arizona, USA.
  10. 10.0 10.1 10.2 Hill, F., Reichart, R., & Korhonen, A. (2014a). SimLex-999: Evaluating Semantic Models with (Genuine) Similarity Estimation. Computation and Language.
  11. Pennington, J., Socher, R., and Manning, C. (2014). Glove: Global vectors for word representation. EMNLP 2014, pp. 1532-1543.
  12. R. Collobert and J. Weston. 2008. A unified architecture for natural language pro- cessing: Deep neural networks with multitask learning. In International Conference on Machine Learn- ing, ICML.
  13. Eric H Huang, Richard Socher, Christopher D Manning, and Andrew Y Ng. 2012. Improving word representations via global context and multiple word prototypes. In Proceedings of the 50th Annual Meeting of the Association for Computational Linguistics: Long Papers-Volume 1, pages 873–882. Association for Computational Linguistics.