Bias mitigation of multimodal datasets in an urban-social category classifier

ABSTRACT

This research project is based on the relational implications of the sociomoral development of Piaget’s psychogenetic theory on the cognition construction of ethics in personal biases as in references of discursive dialectics in linguistics. Functional data from training and testing were parameterized in an urban-social category classifier in a textual analytical approach by Natural Language Processing (NLP) and based on the Transformers adapted attention mechanism. In this perspective, a bias mitigation methodology was developed to restructure the convergence criteria in which multimodal datasets were retrained, retested, and reevaluated. Finally, the heterogeneity of the common collective human ethics was verified and validated, over interpretive inferences, insights, and real social trends, whereby the city/citizen relation addresses the “social sensing” in the identification of public-social problems.

KEYWORDS:
Bias mitigation; Social sensing; Transformers; NLP text analysis; Text classification