Metodología para el análisis de la similitud entre marcas mediante técnicas de aprendizaje automático

Abstract

Trademarks consist of the symbols and words that businesses use to identify their products and services. They are often one of the most valuable assets of a company and therefore there are regulations for their registration and protection. When a trademark is registered, it gives its holder the right to prevent third parties from marketing similar products with identical or similar symbols. In trademark registration and protection processes it is necessary to determine the similarity between 2 trademarks to detect potential confusion that may mislead consumers. Traditionally, this similarity has been established through a qualitative human assessment, but given the increasing number of trademarks registration, the need to automate this task is configured. This research evaluates different techniques of Natural Language Processing (NLP), Computer Vision and phonology, applied in the context of trademark matching, to obtain a system of models that can measure visual, spelling, and phonetic similarity between trademarks. The proposed method is evaluated on a dataset of trademark registration oppositions in applications filed with the Colombian Trademark Office (Superintendencia de Industria y Comercio).