Vývoj chatbota pro doporučování videoher

The video game market grows every year, and navigating an extensive catalogue places increasing cognitive demands on players. Conversational approaches to game recommendation remain underexplored in practice. This thesis therefore focuses on the development of a chatbot that elicits player preferences through a guided dialogue and uses them to select a suitable title from a large game catalogue. A trait-based model of player preferences was adopted as the theoretical framework for structuring the recommendation logic. Ten dimensions were selected from the model based on their predictive power and extended to form the foundation of the algorithm. A survey of 128 active players showed that genre is by far the most important factor in game selection and revealed three dominant search patterns – by reference title, by genre, and by detailed description – which served as a direct basis for the chatbot's design. The findings on player motivations provided empirical support for mapping game tags onto the individual dimensions of the model. The system is built as a three-layer architecture separating a conversational layer implemented in the Botpress v12 platform, an application layer containing the recommendation logic built on the NestJS framework, and a data layer holding a catalogue of 900,000 titles in PostgreSQL. The chatbot offers three recommendation paths corresponding to the identified search patterns. Usability was evaluated using the think-aloud method across two rounds with five participants. Recommendations were rated as relevant, and the second round confirmed improvements in onboarding, questionnaire flow, and intent recognition. A lower reliability of the NLU model when processing longer or less direct user inputs emerged as a systemic limitation.