Difference between revisions of "Paraphrase Identification (State of the art)"
| Line 23: | Line 23: | ||
! Reference | ! Reference | ||
! Description | ! Description | ||
| + | ! Supervision | ||
! Accuracy | ! Accuracy | ||
! F | ! F | ||
|- | |- | ||
| Vector Based Similarity (Baseline) | | Vector Based Similarity (Baseline) | ||
| − | | Mihalcea et al. (2006 | + | | Mihalcea et al. (2006) |
| − | | | + | | cosine similarity with tf-idf weighting |
| + | | unsupervised | ||
| 65.4% | | 65.4% | ||
| 75.3% | | 75.3% | ||
| Line 34: | Line 36: | ||
| KM | | KM | ||
| Kozareva and Montoyo (2006) | | Kozareva and Montoyo (2006) | ||
| − | | | + | | combination of lexical and semantic features |
| + | | supervised | ||
| 76.6% | | 76.6% | ||
| 79.6% | | 79.6% | ||
| Line 40: | Line 43: | ||
| RMLMG | | RMLMG | ||
| Rus et al. (2008) | | Rus et al. (2008) | ||
| − | | | + | | graph subsumption |
| + | | unsupervised | ||
| 70.6% | | 70.6% | ||
| 80.5% | | 80.5% | ||
| Line 46: | Line 50: | ||
| MCS | | MCS | ||
| Mihalcea et al. (2006) | | Mihalcea et al. (2006) | ||
| − | | | + | | combination of several word similarity measures |
| + | | unsupervised | ||
| 70.3% | | 70.3% | ||
| 81.3% | | 81.3% | ||
| Line 52: | Line 57: | ||
| STS | | STS | ||
| Islam and Inkpen (2007) | | Islam and Inkpen (2007) | ||
| − | | | + | | combination of semantic and string similarity |
| + | | unsupervised | ||
| 72.6% | | 72.6% | ||
| 81.3% | | 81.3% | ||
| Line 58: | Line 64: | ||
| QKC | | QKC | ||
| Qiu et al. (2006) | | Qiu et al. (2006) | ||
| − | | | + | | sentence dissimilarity classification |
| + | | supervised | ||
| 72.0% | | 72.0% | ||
| 81.6% | | 81.6% | ||
| Line 64: | Line 71: | ||
| matrixJcn | | matrixJcn | ||
| Fernando and Stevenson (2008) | | Fernando and Stevenson (2008) | ||
| − | | | + | | JCN WordNet similarity with matrix |
| + | | unsupervised | ||
| 74.1% | | 74.1% | ||
| 82.4% | | 82.4% | ||
| Line 70: | Line 78: | ||
| FHS | | FHS | ||
| Finch et al. (2005) | | Finch et al. (2005) | ||
| − | | | + | | combination of MT evaluation measures as features |
| + | | supervised | ||
| 75.0% | | 75.0% | ||
| 82.7% | | 82.7% | ||
| Line 76: | Line 85: | ||
| WDDP | | WDDP | ||
| Wan et al. (2006) | | Wan et al. (2006) | ||
| − | | | + | | dependency-based features |
| + | | supervised | ||
| 75.6% | | 75.6% | ||
| 83.0% | | 83.0% | ||
| Line 82: | Line 92: | ||
| SHPNM | | SHPNM | ||
| Socher et al. (2011) | | Socher et al. (2011) | ||
| − | | | + | | recursive autoencoder with dynamic pooling |
| + | | supervised | ||
| 76.8% | | 76.8% | ||
| 83.6% | | 83.6% | ||
| Line 88: | Line 99: | ||
| MTMETRICS | | MTMETRICS | ||
| Madnani et al. (2012) | | Madnani et al. (2012) | ||
| − | | | + | | combination of eight machine translation metrics |
| + | | supervised | ||
| 77.4% | | 77.4% | ||
| 84.1% | | 84.1% | ||
|- | |- | ||
|} | |} | ||
| − | |||
== References == | == References == | ||
Revision as of 11:33, 26 September 2012
- source: Microsoft Research Paraphrase Corpus (MSRP)
- task: given a pair of sentences, classify them as paraphrases or not paraphrases
- see: Dolan et al. (2004)
- train: 4,076 sentence pairs (2,753 positive: 67.5%)
- test: 1,725 sentence pairs (1,147 positive: 66.5%)
Sample data
- Sentence 1: Amrozi accused his brother, whom he called "the witness", of deliberately distorting his evidence.
- Sentence 2: Referring to him as only "the witness", Amrozi accused his brother of deliberately distorting his evidence.
- Class: 1 (true paraphrase)
Table of results
- Listed in order of increasing F score.
| Algorithm | Reference | Description | Supervision | Accuracy | F |
|---|---|---|---|---|---|
| Vector Based Similarity (Baseline) | Mihalcea et al. (2006) | cosine similarity with tf-idf weighting | unsupervised | 65.4% | 75.3% |
| KM | Kozareva and Montoyo (2006) | combination of lexical and semantic features | supervised | 76.6% | 79.6% |
| RMLMG | Rus et al. (2008) | graph subsumption | unsupervised | 70.6% | 80.5% |
| MCS | Mihalcea et al. (2006) | combination of several word similarity measures | unsupervised | 70.3% | 81.3% |
| STS | Islam and Inkpen (2007) | combination of semantic and string similarity | unsupervised | 72.6% | 81.3% |
| QKC | Qiu et al. (2006) | sentence dissimilarity classification | supervised | 72.0% | 81.6% |
| matrixJcn | Fernando and Stevenson (2008) | JCN WordNet similarity with matrix | unsupervised | 74.1% | 82.4% |
| FHS | Finch et al. (2005) | combination of MT evaluation measures as features | supervised | 75.0% | 82.7% |
| WDDP | Wan et al. (2006) | dependency-based features | supervised | 75.6% | 83.0% |
| SHPNM | Socher et al. (2011) | recursive autoencoder with dynamic pooling | supervised | 76.8% | 83.6% |
| MTMETRICS | Madnani et al. (2012) | combination of eight machine translation metrics | supervised | 77.4% | 84.1% |
References
- Listed alphabetically.
Dolan, B., Quirk, C., and Brockett, C. (2004). Unsupervised construction of large paraphrase corpora: Exploiting massively parallel news sources, Proceedings of the 20th international conference on Computational Linguistics (COLING 2004), Geneva, Switzerland, pp. 350-356.
Fernando, S., and Stevenson, M. (2008). A semantic similarity approach to paraphrase detection, Computational Linguistics UK (CLUK 2008) 11th Annual Research Colloquium.
Finch, A., and H, Y.S., and Sumita, E. (2005). Using machine translation evaluation techniques to determine sentence-level semantic equivalence, "Proceedings of the Third International Workshop on Paraphrasing (IWP 2005)", Jeju Island, South Korea, pp. 17-24.
Islam, A., and Inkpen, D. (2007). Semantic similarity of short texts, Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2007), Borovets, Bulgaria, pp. 291-297.
Kozareva, Z., and Montoyo, A. (2006). Paraphrase identification on the basis of supervised machine learning techniques, Advances in Natural Language Processing: 5th International Conference on NLP (FinTAL 2006), Turku, Finland, 524-533.
Madnani, N., Tetreault, J., and Chodorow, M. (2012). Re-examining Machine Translation Metrics for Paraphrase Identification, Proceedings of 2012 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL 2012), pp. 182-190.
Mihalcea, R., Corley, C., and Strapparava, C. (2006). Corpus-based and knowledge-based measures of text semantic similarity, Proceedings of the National Conference on Artificial Intelligence (AAAI 2006), Boston, Massachusetts, pp. 775-780.
Qiu, L. and Kan, M.Y. and Chua, T.S. (2006). Paraphrase recognition via dissimilarity significance classification, Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP 2006), pp. 18-26.
Rus, V. and McCarthy, P.M. and Lintean, M.C. and McNamara, D.S. and Graesser, A.C. (2008). Paraphrase identification with lexico-syntactic graph subsumption, FLAIRS 2008, pp. 201-206.
Socher, R. and Huang, E.H., and Pennington, J. and Ng, A.Y., and Manning, C.D. (2011). Dynamic pooling and unfolding recursive autoencoders for paraphrase detection, "Advances in Neural Information Processing Systems 24"
Wan, S., Dras, M., Dale, R., and Paris, C. (2006). Using dependency-based features to take the "para-farce" out of paraphrase, Proceedings of the Australasian Language Technology Workshop (ALTW 2006), pp. 131-138.