1 Introduction

This work is focused on identifying posts and opinions of people in the Twitter social network by using a tagging algorithm in the natural language processing. The algorithm consists in separating two types of comments from the Tweets: praises and insults. The dynamic consists in downloading the information and group it into two well defined classes. In this case, the ”Tweets” that take our interest are downloaded with all the related information (id, author, message, date, language, etc.). The algorithm developed utilizes a clips library from Python that allows the extraction of comments in a customized way. With this it is possible to mark from one word to a sentence with the needed content.

The comments that are taken from Twitter have a special variable called Unicode, intractable as a string type variable. In this point, the information is stored into a csv (comma separated values) file and it is then codified as UTF8 in order to support special characters including accent marks.

In this article we have focused our attention to the opinions that Twitter users write regarding the Mexican presidential elections, and from the different descriptive terms that are identified surrounding the presidential candidates, the tweets are separated into opposite descriptive terms about praises and disapproval. The information processed comprises 2000 random tweets per day, and in a similar manner, 500 comments are selected for each candidate.

The present work is organized as follows: This introduction as section 1. The section 2 presents the programming reference framework. Section 3 makes use of the previous section describing the modules with a practical end for the final users and at the same time with the purpose of helping programmers understand the implementation logic and allow its modification.

In section 4 we explain a basic example about the implementation of the algorithm. In this case we show how a news piece has many associated comments, negative or positive. Finally, in section 5 we discuss the results and future work.

2 Program Strategy

In this section we describe the necessary computational terms in this software:

3 Modular Programming Methodology

To download the tweets that the algorithm requires, the first step consists in identifying the relevant information of the tweet. In this point, the topic in the corpus without a specific word is enough. When enough tweets are collected, these must be concentrated to the least amount of words due to the ambiguity of the relations between comments and words of the tweets downloaded. This control is measured to avoid useless information in amounts that are impossible to treat.

For this work, inside the central topic, the users that support political parties were used as keywords.

Once a considerable amount of tweets has been downloaded, the information is cleaned with the UTF-8 codification to be used as strings. On the other hand, the Python split function has to be used. The split function, divides a chain of characters into sub-chains, according to a delimiter, to fragment the strings in at least two parts.

4 Proposed Method of Analysis

Our method consists of the following steps.

5 Results

In this section, we present a review of the results of the dictionary algorithm. Descriptive statistics are the focus of this section.

The variability of praises and insults mainly consisted of identifying the amount of words that reflected a sentiment regarding a praise or an insult. In this work, the frequency of the qualifying terms has been the main observation from the tweets: those that were repeated at least 10 times, were included in the dictionary.

The similarity between the words of the tweets, with respect to a general dictionary of praises and insults initially established, constitutes the final dictionary of the algorithm developed. For this, each word in the tweets is checked, making an exact analogy with respect to the words of the dictionary proposed at the beginning. With a linear search to each individual tweet, the identical qualifying terms are identified and the dictionary that the algorithm needs is created.

The majority of the praises collected in the tweets were downloaded during 10 days. In praises (Fig. 1), we can see that the Anaya (blue color) had more points than others candidates (29%). In insults (Fig. 2), we observe that Obrador had more number of offenses (28%). Nevertheless, the candidates had more or less similar number of insults and praises.

Fig. 1 Praises of candidates in Twitter 

Fig. 2 Insults of candidates in Twitter 

6 Conclusions

In this work, an algorithm was developed to identify the sentiments during a political event in twitter [⁵]. The goal was focused on recognizing two variants of qualifying terms associated to the presidential candidates for the 2018 Mexican election. To initiate this algorithm, we assumed that the polls done by different companies had to have support in additional analyses.

The results that we present reveal the percentages of praises and insults before the second debate, which are not actually revealing [⁶].

However, after the second debate, the praises leaned towards the candidate of the MORENA party, though without significant difference with respect to the others.