Paper Information | SBC BOOKS

Fill in your paper information

Language (*)

Title (*)

Keywords (*)

Abstract (*)

Junção por similaridade retorna todos os pares de objetos cuja similaridade não é menor que um limite especificado. Essa operação é de fundamental importância para limpeza e integração de dados. Uma abordagem popular é adotar uma representação em espaço vetorial e usar cosseno para medir a similaridade de dois vetores. Calcular a similaridade de cosseno em todos os pares de vetores é proibitivamente custoso para grandes conjuntos de dados. Algoritmos de junção de similaridade exploram a esparsidade do espaço vetorial gerado por métodos tradicionais para derivar filtros e, assim, reduzir o espaço de comparação. Avanços recentes no processamento de linguagem natural possibilitaram representações baseadas em vetores semanticamente mais ricos, aumentando assim a qualidade dos resultados. No entanto, tais vetores apresentam características diferentes daquelas geradas por métodos tradicionais. Em particular, esses vetores são densos e possuem menor dimensionalidade. Nesse contexto, apresentamos uma proposta de pesquisa centrada em duas questões principais: 1) como essas características afetam o desempenho dos algoritmos de junção por similaridade existentes?; 2) é possível obter desempenho nesta nova representação comparável aos métodos anteriores mantendo a qualidade dos resultados? Este artigo descreve ainda nossa metodologia de pesquisa, apresenta alguns resultados iniciais e delineia trabalhos futuros.

Pages (*)

File Link

English Information

Title

Keywords

Abstract

Similarity join returns all pairs of objects whose similarity is not less than a speciﬁed threshold. This operation is of fundamental importance for data cleaning and integration. A popular approach is to adopt a vector-space representation and use cosine to measure the similarity of two vectors. Calculating cosine similarity over all vector pairs is prohibitively expensive for large datasets. State-of-the-art similarity join algorithms exploit the sparsity of the vector space generated by traditional methods to derive filters and, thus, reduce the comparison space. Recent advances in natural language processing have enabled representations based on semantically richer vectors, thereby increasing the quality of the results. However, such vectors exhibit different characteristics from those generated by traditional methods. In particular, these vectors are dense and have lower dimensionality. In this context, we present a research proposal centered around two main questions: 1) how do these characteristics affect the performance of existing similarity join algorithms?; 2) is it possible to achieve performance in this new representation comparable to previous methods while maintaining the quality of the results? This paper further describes our research methodology, presents some initial results, and outlines future work.

(*) To change the order drag the item to the new position.

Authors

#	Name
1	Leonardo Ribeiro (laribeiro@inf.ufg.br)
2	Douglas Santana(douglasrolins@discente.ufg.br)

(*) To change the order drag the item to the new position.

Reference

#	Reference
1	Anastasiu, D. C. and Karypis, G. (2014). L2AP: Fast Cosine Similarity Search with Prefix L-2 Norm Bounds. In Proceedings of the ICDE Conference, pages 784–795.
2	Chaudhuri, S., Ganti, V., and Kaushik, R. (2006). A Primitive Operator for Similarity Joins in Data Cleaning. In Proceedings of the ICDE Conference, page 5.
3	Devlin, J., Chang, M., Lee, K., and Toutanova, K. (2019). BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. In Proceedings of the Confe- rence of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 4171–4186
4	Johnson, J., Douze, M., and J ́egou, H. (2019). Billion-scale similarity search with GPUs. IEEE Transactions on Big Data, 7(3):535–547.
5	Li, Y., Li, J., Suhara, Y., Doan, A., and Tan, W. (2020). Deep Entity Matching with Pre-Trained Language Models. Proceedings of the VLDB Endowment, 14(1):50–60.
6	Mikolov, T., Chen, K., Corrado, G., and Dean, J. (2013). Efficient estimation of word representations in vector space.
7	Mudgal, S., Li, H., Rekatsinas, T., Doan, A., Park, Y., Krishnan, G., Deep, R., Arcaute, E., and Raghavendra, V. (2018). Deep Learning for Entity Matching: A Design Space Exploration. In Proceedings of the SIGMOD Conference, pages 19–34.
8	Reimers, N. and Gurevych, I. (2019). Sentence-bert: Sentence embeddings using siamese bert-networks. CoRR, abs/1908.10084.
9	Ribeiro, L. A. and H ̈arder, T. (2011). Generalizing Prefix Filtering to Improve Set Simi- larity Joins. Information Systems, 36(1):62–78.
10	Ribeiro-J ́unior, S., Quirino, R. D., Ribeiro, L. A., and Martins, W. S. (2017). Fast Parallel Set Similarity Joins on Many-core Architectures. Journal of Information and Data Management, 8(3):255–270.
11	Salton, G., Wong, A., and Yang, C. S. (1975). A vector space model for automatic inde- xing. Commun. ACM, 18(11):613–620.
12	Suri, S., Ilyas, I. F., R ́e, C., and Rekatsinas, T. (2021). Ember: No-Code Context En- richment via Similarity-Based Keyless Joins. Proceedings of the VLDB Endowment, 15(3):699–712.
13	Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, L., and Polosukhin, I. (2017). Attention is All you Need. In Annual Conference on Neural Information Processing Systems, pages 5998–6008.

Paper Registration

Fill in your paper information

English Information

Authors

Reference