Artur Zygadło
supervisor: dr hab. inż. Artur Janicki
The goal of this research is to create an intelligent therapeutic dialogue agent, based on artificial intelligence methods, capable of understanding the user’s inputs and responding in Polish.
In the initial phase, the research focus is on semi-automatic creation of a relevant training dataset, a so-called corpus of dialogues in Polish. It is planned to prepare sentences containing so-called slots, which can then be filled with dictionary values, e.g. names, numbers, temporal expressions, items from a given category etc.
The dialogue system will be implemented within the Chatterbot framework. Such an agent responds by selecting the sentence that best matches the user input in terms of a particular sentence similarity metric. Experiments with different sentence similarity metrics are planned, including those using so-called word embeddings, neural network-based methods like word2vec or BERT, for which models for Polish are available.
Within this research, it is planned to evaluate the developed system in terms of its correctness and user acceptance. One of the research goals is to create a system, with which the user will be willing to interact and develop a relationship. The conclusions from the evaluation will lead to the further improvement of the dialogue system, making it an innovative solution.