Realization and testing of intelligent behaviour in chatbots

Within the past few decades, the chatbot technology has developed in a way that gave birth to various types of these communication systems. There are chatbots which communicate through rules entered by their human creator, where the bot recognises a certain intent or entity as a part of the text and returns a predefined answer. However, chatbots of a higher quality help themselves by using elements of artificial intelligence, for example they can detect emotion and tones, recognise the content of received images or generate new text with a pre-trained model. These very chatbots, which can manifest intelligent behaviour, are the topic of this thesis. The theoretical goals of this work are to research the topic of chatbots and and chatbot construction platforms, intelligent components in chatbots and testing of the (rational, intelligent) behaviour of chatbots. Three practical goals follow; first of all comes design, realisation and deployment of specific chatbots with a differing degree of "intelligence", next is creation of a questionnaire for chatbot testing evaluation and testing with several groups of users and finally evaluation, conclusions and recommendation. As a part of the thesis, three chatbot variants were created that represent three different chatbot groups regarding their usage of intelligent elements. The goal of each of them was to recommend a music album to the user after a short conversation. These variants were tested by groups of users. The testing was finished by completing the questionnaire and the collected questions were used to evaluate the chatbots. The variants which utilised elements of artificial intelligence were proven to achieve better results based on the rationality of the dialogue, rationality of the final recommendation as well as user friendliness and fuctionality.