Deep multi-domain representations for analyzing social media posts

• Grant number: 21/03830-0
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): June 01, 2021
• Effective date (End): May 31, 2022
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Moacir Antonelli Ponti
• Grantee: Guilherme Amaral Hiromoto
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Associated research grant: 19/07316-0 - Singularity theory and its applications to differential geometry, differential equations and computer vision, AP.TEM

Abstract

With the huge amount of data circulating on social networks, the relationships between the content of posts (context) and the hashtags used as a way of engaging in multiple social spheres become increasingly relevant. Such interactions can be modeled from relationships between data from posts and hashtags that are generated from techniques of learning joint representations: involving textual and visual data, in a vector cross-domain format, which allows for correlation and recognition analysis of patterns. The project aims to investigate methods to learn abstract representations from datasets and to relate those representations to other groups of distinct domains, but with similar semantics. We will consider feature learning methods, particularly convolutional neural networks, and word embedding, and explore architectures capable of achieving a match between the learned representations. In addition, each solution will be studied from the point of view of its guarantee of learning. (AU)