Use of Machine Learning Models of the “Transformers” Type in the Construction of Services in a Gamified Web app

Abstract

Abstract The purpose of this document is to describe the use of a natural language processing model in the multiplatform system “Gamivity” by means of a sentence similarity algorithm to offer a personalized experience module based on the conceptual relationship between questions. For the selection process, certain criteria were chosen that will allow several pre-trained models under the “Transformers" architecture for evaluation, later. These criteria were the language with which the model was altered; Python was the programming language used for the implementation. Regarding the evaluation phase of the selected models, the “Sentence Transformers” library of the Python programming language was used. In addition, a work environment analogous to the module present in the “Gamivity” system was built, in which the development platform “Google Colab” was used to test these models. The criteria for choosing the candidate model were based on its effectiveness in relation to questions as well as the computational cost involved while performing the operations in the said model Based on the applied methodology, the model that yielded the best results was “paraphrase-multilingual-MiniLM-L12-v2,” modified with a large corpus of text in Spanish and 50 other languages, which showed a degree of precision. When it comes to conceptually relating the questions provided it was found to be optimal, having relatively low computational cost when performing these operations.

Keywords

• sentence transformers, sentence similarity, relate questions, personalized learning.