Evolutionary Algorithms to Semantic Role Labeling

• Grant number: 12/07926-3
• Support Opportunities: Scholarships in Brazil - Doctorate
• Effective date (Start): June 01, 2012
• Effective date (End): January 31, 2014
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Zhao Liang
• Grantee: Murillo Guimarães Carneiro
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil

Abstract

Semantic roles represent the logical relationships between an event and its participants. Semantic role labeling (SRL) is the process of automatically extracting semantic roles structures that allow analysis of the meaning of sentences and provide information useful in solving many tasks of natural language processing (NLP), such as information extraction, documents categorization and classification, machine translation, among others. Most models in SRL are developed for the English language because there are few labeled sources to other languages. So, the exploitation of SRL in these languages is a great challenge for NLP. To reduce the difficulty in building models to the Portuguese language, PropBank.br was recently developed. However, another challenge is that the methods from literature for SRL have presented limitations related to low generalization ability, low portability between different labeled sources, as well as computational costs increasing. Evolutionary Algorithms (EAs) are stochastic techniques of search and optimization guided by simulation of mechanisms of the natural selection and genetics. Among other qualities, they are able to take advantage of parallel architectures, working on problems with little information and have obtained good results in many applications. Thus, this project includes research and development of methods based on evolutionary algorithms, such as differential evolution, for the automatic semantic role labeling. Some applications of EAs in other NLP tasks provide evidences that this is promising way. In principle, they can contribute to SRL in terms of results and computational performance. EAs can also obtain generalization and portability more effective than existing methods due to their adaptive characteristics. In this context, to effectively evaluate the performance of new models will be carried out tests well known in literature such as precision, coverage and F1, considering mainly PropBank.br and also analyzing the results of the methods developed in different applications. As final result is expected that the thesis elaborated assist the computational study of semantic roles for the Portuguese language and the models developed can be used in various systems of the NLP.

Scientific publications

(References retrieved automatically from Web of Science and SciELO through information on FAPESP grants and their corresponding numbers as mentioned in the publications by the authors)

CARNEIRO, MURILLO GUIMARAES; ZHAO, LIANG. Organizational Data Classification Based on the Importance Concept of Complex Networks. IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS, v. 29, n. 8, p. 3361-3373, AUG 2018. (13/07375-0, 12/07926-3, 15/50122-0)

CARNEIRO, MURILLO G.; CHENG, RAN; ZHAO, LIANG; JIN, YAOCHU. Particle swarm optimization for network-based data classification. NEURAL NETWORKS, v. 110, p. 243-255, FEB 2019. (15/50122-0, 13/07375-0, 12/07926-3)

Academic Publications

(References retrieved automatically from State of São Paulo Research Institutions)

CARNEIRO, Murillo Guimarães. Data classification in complex networks via pattern conformation, data importance and structural optimization. 2016. Doctoral Thesis - Universidade de São Paulo (USP). Instituto de Ciências Matemáticas e de Computação (ICMC/SB) São Carlos.