What does the media tell us about a destination? The Cancun case, seen from the USA, Canada, and Mexico

PurposeThis study aims to present a framework for automatically collecting, cleaning and analyzing text (news articles, in this case) to provide valuable decision-making information to destination management organizations. Keeping a record of certain aspects of the projected destination image of an attraction (Cancun in this study) will grant the design of better strategies for the promotion and administration of destinations without the time-consuming effort of manually evaluating high quantities of textual information. Design/methodology/approachUsing Web scraping, news articles were collected from the USA, Mexico and Canada over an interval of one year. The documents were analyzed using an automatic topic modeling method known as Latent Dirichlet Allocation and a coherence analysis to determine the number of themes present in each collection. With the data provided, the authors were able to extract valuable information to understand how Cancun is presented to the countries. FindingsIt was found that in all countries, Cancun is an important destination to travel and vacation;however, given the period defined for this study (from July 2021 to July 2022), an important part of the articles analyzed was concerned with the sanitary measures derived from the COVID-19 pandemic. Besides, given the rise of violence and the threat of organized crime, many articles from the three countries are focused on warning potential tourists about the risks of traveling to Cancun. Originality/valueThe examination of the relevant literature revealed that similar analyses are manually performed by the experts on a set of predefined categories. Although those approaches are methodologically sound, the logistic effort and the time used could become prohibitively expensive, precluding carrying out this analysis frequently. Additionally, the preestablished categories to be studied in press articles may distort the results. For these reasons, the proposed framework automatically allows for gathering valuable information for decision-making in an unbiased manner.

Overview of Rest-Mex at IberLEF 2022: Recommendation System, Sentiment Analysis and Covid Semaphore Prediction for Mexican Tourist Texts

This paper presents the framework and results from the Rest-Mex task at IberLEF 2022. This task considered three tracks: Recommendation System, Sentiment Analysis and Covid Semaphore Prediction, using texts from Mexican touristic places. The Recommendation System task consists in predicting the degree of satisfaction that a tourist may have when recommending a destination of Nayarit, Mexico, based on places visited by the tourists and their opinions. On the other hand, the Sentiment Analysis task predicts the polarity of an opinion issued and the attraction by a tourist who traveled to the most representative places in Mexico. We have built corpora for both tasks considering Spanish opinions from the TripAdvisor website. As a novelty, the Covid Semaphore Prediction task aims to predict the color of the Mexican Semaphore for each state, according to the Covid news in the state, using data from the Mexican Ministry of Health. This paper compares and discusses the participants' results for all three tacks.

Mexican Epidemiological Semaphore Color Prediction by Means of Mutual Information Features

In this paper is presented a proposed solution to predict the Mexican Epidemiological Semaphore (MES) color from a set of online news. This problem was presented in Rest-Mex 2022: Recommendation System, Sentiment Analysis and Covid Semaphore Prediction for Mexican Tourist Texts. This task consists of determining the MES color through the COVID-19 news in Mexico until 8 weeks in advance. The MES system is crucial because it indicates which kind of activities are allowed to the population, for example, tourism activities. Thus, our approach is based on the Mutual Information (MI) measure. In the training stage, by using the training data, our approach first clusters every word from every news by the respective class. Then, for each word in each class, we compute its MI value. In this way, the set of words (trained words) with their normalized MI value is used as class features. In the classification stage, when a new instance is given, each word is intersected with the trained words for each class, and the corresponding MI values of the intersected words are summed. The predicted class is assigned to the class with the highest sum value. The final ranking value on the testing data was 0.175716016. We think that the obtained results are because the data has many noise words (tokens), and our approach does not deal with that issue. © 2022 Copyright for this paper by its authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).