Artificial intelligence to answer challenges of Mexican society
These challenges are organized by IberLEF, a campaign for the comparative evaluation of automatic processing systems for Iberian natural languages. The objective is, through artificial intelligence, to promote research related to Spanish languages and related dialects. The topics of the challenges were selected for their concrete impact on Mexican society. To achieve this, all participants had access to the same data to propose the best methodologies, processes and ideas.
Boosting tourism
The first challenge, called RestMex, aimed to solve problems related to recommending and predicting the level of users’ satisfaction when visiting a Mexican tourist place. Tourism is vital in Mexico, representing 8.7% of the national GDP and generating about 4.5 million direct jobs. However, this sector has been one of the most affected by the COVID pandemic. Through this challenge, the tourism sector is trying to recover by improving the quality and safety of offerings and services.
As part of RestMex, Esau Villatoro and CIMAT team implemented an efficient matching system between a user's profile and a set of tourist destinations. The proposed method can leverage contributors’ profiles to the TripAdvisor platform and the reviews they wrote. This information is then used to recommend the most appropriate places for a particular user. This is made possible by predicting the user's satisfaction when visiting the recommended location.
Improving security through social networks analysis
"When they witness a violent event, in my country people are really used to informing and sharing such acts on social networks," explains Esau Villatoro. The second challenge his team won, the DA-VINCIS challenge, aimed to develop a method to identify and classify violent acts or incidents reported on Twitter. To achieve this, Esau and the CIMAT team used pre-trained language models and combined them with specific multi-task learning strategies. The proposed methodology is capable of detecting whether (or not) a tweet is reporting a violent incident, while at the same time, is able to distinguish the subtypes of incidents cited in the tweet (e.g. accident, robbery, kidnapping, etc.). "The Mexican youth wants to get involved and provide solutions to concerning problems among the Mexican society. This motivates me to participate again for the next edition next year in 2023". His work was presented at the IberLEF-SEPLN 2022 conference (Conference of the Spanish Society for Natural Language Processing) held in A Coruña, Spain in September 2022.
More information