Abstract:

Sentiment analysis in NLP (Natural Language Process) is a technique used to identify, extract and quantify the emotional polarity or subjectivity of a text. In other words, sentiment analysis makes it possible to determine whether a text has a positive, negative or neutral connotation, based on the identification of emotions, opinions, attitudes and value judgments expressed in the text. Sentiment analysis is used to make important assessments about a customer's attitude towards a company.

Objective:
Perform sentiment analysis on the opinions expressed by students at the end of their internships using three artificial intelligence tools and compare the results.

Methodology:
The results obtained from the application of a previously validated instrument for monitoring the performance of professional practices during the period from 2016 to 2022 were taken. All the comments expressed by the 1520 students of the different degrees taught at the Faculty were extracted and the MeaningCloud platform API, the Python programming language making use of the Pandas and NLTK (Natural Language Toolkit) libraries were applied in parallel, and subsequently the chatGPT ver 4 tool was used. This analysis was conducted using TextBlob as a proxy for a more complex model like BERT, and the sentiments were classified as positive, negative, or neutral based on the polarity of the sentences by heuristic sentiment technic. The results were then compared to determine the effectiveness of these tools.

Results and Discussion:
The results of the first tool had to be homogenized by grouping the polarities N+ with N, neutral with none and P+ with P. In the case of the MeaningCloud tool, most of the comments emitted were classified with positive polarities, while with the Python tool it was the opposite, the minority obtained a positive polarity. On the other hand, the results emitted by the chatGPT tool were those based on the heuristic sentiment analysis that groups them into three polarities positive, negative and neutral.

Conclusions:
The comments expressed by the students at the end of their internship in the company can be considered very satisfactory since most of them had a POSITIVE polarity; however, there is a discrepancy in the results obtained between both tools, so it can be concluded that the traditional text analysis tools still present problems in semantics; proof of this are the results obtained in the neutral comments that contain words that should be qualified with positive polarity if reviewed manually. Another factor that influences the results is the language, since it does not have a library that correctly recognizes Spanish, as was the case with Phyton. For the case of chatGPT, it suggests using algorithms that handle sentiment analysis in Spanish, however, in this study we only compared the results based on the heuristic analysis, it also allows us to perform a basic, textual and mixed analysis that would be convenient to review if there are significant differences in the results. For the purposes of this research, the results were not segmented, but it is recommended that in a subsequent analysis we look for correlations or patterns to identify if some factors such as the area where the internships were performed or the economic stimulus or some other variable influences the polarity of the feelings.

Keywords:

Sentiment analysis, artificial intelligence, chatGPT, Natural Language Process.