AutoML Strategy Based on Grammatical Evolution: A Case Study about Knowledge Discovery from Text

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4356–4365 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 4356 AutoML strategy based on grammatical evolution: A case study about knowledge discovery from text Suilan Estevez-Velarde1, Yoan Guti´errez2, Andr´es Montoyo3, and Yudivi´an Almeida-Cruz1 1School of Math and Computer Science, University of Havana, Cuba {sestevez,yudy}@matcom.uh.cu 2University Institute for Computing Research (IUII), University of Alicante, Spain 3Department of Languages and Computing Systems, University of Alicante, Spain {ygutierrez,montoyo}@dlsi.ua.es Abstract The process of extracting knowledge from natural language text poses a complex prob- lem that requires both a combination of ma- chine learning techniques and proper feature selection. Recent advances in Automatic Ma- chine Learning (AutoML) provide effective tools to explore large sets of algorithms, hyper- parameters and features to find out the most suitable combination of them. This paper proposes a novel AutoML strategy based on probabilistic grammatical evolution, which is evaluated on the health domain by facing the knowledge discovery challenge in Spanish text documents. Our approach achieves state-of- the-art results and provides interesting insights into the best combination of parameters and algorithms to use when dealing with this chal- lenge. Source code is provided for the research community. 1 Introduction In recent years there has been a large increase in the amount of technical documents produced by the scientific community. These documents form part of a large corpora of knowledge (e.g., re- search papers and encyclopedias) mostly accessi- ble through text-based search engines. However, to better exploit this knowledge (i.e., not just re- covering text fragments or URLs) in automated processes, it is necessary to process the text and extract the pieces of information in a semantic for- mat useful for machine analysis. The challenge of automatically extracting useful knowledge from text sources is studied by research areas such as knowledge discovery (Gonzalez et al., 2015), on- tology learning (Cimiano et al., 2009) and learning by reading (Barker et al., 2007). One of the most useful representation for dis- covering knowledge in natural language sentences is using Subject-Verb-Object triplets (Estevez- Velarde et al., 2018). This structure is ubiqui- tous in many human languages (Crystal, 1997), and as such, it has been used as the base for knowledge extraction in several systems (Mitchell et al., 2018). In a similar line, Giunchiglia and Fumagalli (2017) propose the use of objects, ac- tions and functions as main components for repre- senting common knowledge. Additionally, other relations with a specific meaning are often used in ontologies to represent taxonomies (i.e., is-a) or composition (i.e., part-of). Hence, combining objects and actions with a small set of specific semantic relations allows the representation of a broad range of domains with a simple computa- tional structure. Specifically in the health domain, research in knowledge discovery techniques has steadily in- creased motivated by the potential impact in the quality of human life. Even though the amount of medical knowledge published in natural lan- guage is considerable, it is still an open …

Content truncated to 500 words