Late Breaking Advances on Conceptual Structures

Actualizada 9 de septiembre de 2024

Publicada 3 de junio de 2024

Organizers:

Roberto G. Aragón, Univ. of Cádiz
Manuel Ojeda-Hernández, Univ. of Málaga
David Lobo, Univ. of Cádiz

Keywords and Topics

Formal concept analysis
Fuzzy sets and fuzzy logic
Rough sets
Fuzzy rough sets
Interval-valued fuzzy sets
Operators in relational data analysis
Fuzzy relation equations
Mathematical morphology
Aggregation operators in relational data analysis

Content

There are many approaches for obtaining information from relational datasets and representing it by means of conceptual structures. Formal Concept Analysis (FCA) is a mathematical tool used for extracting and handling information from databases and it has been related to other strategies, such as Rough Set Theory, Possibility Theory, Mathematical Morphology, Fuzzy Relation Equations, fuzzy logic, etc. The combination of these methods has resulted in the creation of robust and efficient mechanisms that take advantage of the main properties of each of them. Furthermore, some of them have been developed with the intention of enabling the treatment and management of information with uncertainty. This Special Session focuses on both theoretical and applied tools developed within the aforementioned topics, which have been recently published or describe new unpublished research.

Workshop keynote speaker

Peter Eklund is program lead for technology internships at Uniting, the community services not-for-profit of the Uniting Church, Honorary Professor at Deakin University, and an elected fellow of Australian Computer Society. Peter is a senior editor at Expert Systems with Applications and in 2022, he and his co-authors won the Andrew P. Sage award for best IEEE transactions paper. Co-founder of three tech start-ups, Peter is advisor to ZTLment, a Copenhagen-based blockchain fintech.

Title: Enhanced natural language processing using semantic triples

Abstract: Semantic Information Extraction is the task of extracting important features from unstructured text. Semantic Information Extraction involves several tasks, examples of which may be Named Entity Recognition (NER), Relation Extraction, Topic Modelling, Entity linking, Open Information Extraction (OpenIE), Term extraction, and key-phrase extraction. With the advent of large language models and the availability of massive amounts of text data, natural language processing (NLP) tasks such as document classification, question answering (QA), and passage ranking have improved dramatically in recent years. One approach that has gained traction in the NLP community is OpenIE, a technique that automatically extracts semantic triples of the form <subject, verb, object> from natural language text without relying on pre-defined schemas or knowledge bases. In this talk, I explore the potential of semantic triples extracted using OpenIE in natural language tasks, these are document relevance classification, unsupervised question generation, and passage re-ranking. Specifically, I demonstrate how triples can be used to enhance the performance of these tasks by providing more accurate and comprehensive information about text.

Política de Cookies

Personalización de Cookies

UniversidaddeCádiz

UniversidaddeCádiz