Deep Learning Model Based on a Transformers Network for Sentiment Analysis Using NLP in Sports Worldwide

Baca L.; Ardiles N.; Cruz J.; Mamani W.; Capcha J. · Communications in Computer and Information Science (2023) · DOI: 10.1007/978-3-031-37940-6_27

Resumen

Currently, the amount of information processed through the Internet reaches 396 exabytes per month, almost triple what was generated in 2017, and the number of users has gone from 18 billion to almost 29 billion in 2022. This information, generally extracted from unstructured sources, must be processed to obtain data of interest. Users of social networks in 2022 reached a number of 4.7 billion, increasing by 5% compared to last year. Artificial Intelligence, specifically Natural Language Processing, offers an alternative to information processing. In this sense, the model based on Transformer DistilBERT carried out based on BERT allowed the analysis of sentiments of the tweets issued on the Twitter platform related to the Qatar 2022 World Cup. For this, the Google platform located in the cloud was used, as well as Vertex IA, Storage, and BigQuery services. Another resource used was the Tweeter API to collect the data and the CRISP-DM methodology. The results showed the relationship between the variables, with a normal distribution and few outliers. To improve the work in future research, it is proposed to eliminate the outliers and tune some hyperparameters of the model.

Instituciones: Universidad Nacional del Altiplano; Universidad César Vallejo