Análisis de Temas Utilizando Twitter: Una Aplicación del Modelo LDA al Caso Colombiano

Abstract

According to (Alvino, 2021), Twitter is one of the most used social networks in the world, with more than 322 million users has figures for the month of July 2021, and is based on the publication of microblogs in which people mainly publish in writing what they are thinking at the moment. It is a perfect social network for those people who, due to their recognition or fame, can easily influence their opinions, thanks to their large number of fans or followers.

Twitter in recent years has designed an API which allows you to connect to the information that is produced every second on its platform, from the more than 322 million users. To access this information, 2 main programming languages are used: R and Python.

In the current work, the Twitter API is used, because thanks to this tool, you can download the data, the opinions that are being posted in real time every minute on the social network. With this data, analyzes can be carried out, for example, exploratory analyzes on people's opinions about local elections in capital cities, since different types of location filters can be applied, and data can be obtained from specific cities, which is done in this job. In addition, the classification will be made between whether what users post has a positive or negative feeling.

With the above, different Natural Language Processing (NLP) techniques will also be used, in order to obtain conclusions corresponding to the topics of interest, and through the findings, obtain results. The main technique that will be used in the following work corresponds to topic modeling techniques, especially the LDA (\textit{Latent Dirichlet Allocation} model).

The Latent Dirichlet Allocation (LDA) is a generative statistical model that allows explaining sets of observations by means of unobserved groups that explain why some parts of the data are similar (Blei et.al 2003).

It can be said that there are two kinds of topic modeling, which are linear topic models and probabilistic topic models. An example of the linear model is the Latent Semantic Assignment (LSA) model and an example of the probabilistic model is the Dirichlet Latent Assignment model, which is the main model of real work. Similar works have been developed, one of which is Guarnizo, E. \& Monroy, A. (2021), in which public acceptability on the subject of the special jurisdiction for peace was tried to be inferred, using opinions on this specific topic, taken from social networks.

These results can be implemented in matters of public policy, or political influence, taking into account the pre-electoral panorama that Colombia faces, and even more so with events such as the Covid 19 pandemic, the National Strike, or the resurgence of violence. (Haselmayer & Jenny, 2016)