APPLICATION OF DATA-COLLECTING CHATBOT FOR POLISH TEXT EMOTION ANALYSIS

The main purpose of this article is to prepare a tool that will allow the preparation of a set of training data for emotion analysis models in Polish texts. An important aspect in social robotics is maintaining a high level of human contact not only by understanding spoken words but also by recognizing emotions. There are solutions to analyze the sentiment of statements, but taking a step further and analyzing emotions in more detail, we encounter fewer and fewer data sets. The article aims to present the architecture of the system for the preparation of specific data sets in the field of text emotion analysis, based on ChatGPT chatbot. In addition, based on the generated data, a model for analyzing emotions in Polish texts was created based on LSTM neural networks and distribution models. The paper compares a trained model on a dataset proposed by artificial intelligence (AI) and an expert set assessed by a human. It can be seen that the proposed model copes well with a set of data evaluated by a human and the accuracy of such a model is comparable to the operation of the model on a set generated by AI. The proposed issues and solutions can help improve social robots that will recognize emotions more accurately. Codes and datasets are available at https://github.com/SkowRon96/Polish-texts-emotionanalysis-chatbot.