Project of an Information System for Creating Personalized Game Recommendations
Students Name: Hlovatskyi Rostyslav Vasylovych
Qualification Level: magister
Speciality: IT Project Management
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2024-2025 н.р.
Language of Defence: ukrainian
Abstract: This work is devoted to the development of an innovative personalized game recommendation system that utilizes a hybrid algorithm combining collaborative filtering and content-based analysis. A distinctive feature of the system is its deep integration with the Steam platform and the use of semantic tag analysis for precise determination of users’ gaming preferences. The recommendation algorithm is based on a hybrid approach that combines collaborative filtering and content-based analysis. For semantic tag analysis, the Word2Vec model was used, which allowed representing tags as vector representations and calculating their semantic similarity. To further improve the quality of recommendations, a tag combination mechanism was introduced. This allowed for the identification of more nuanced user preferences and the suggestion of games that match not only individual tags but also their combinations. For example, if a user likes RPG games with stealth elements, the system can suggest games that combine both of these elements. The tag system was built based on tags provided by the Steam platform and supplemented with custom tags, allowing for a more detailed description of games. When a user has a large number of tags, the system employs a selection mechanism, analyzing which tags the user has the most games with or spends the most time on. This allows for a more accurate determination of their key interests. The system architecture is based on a microservices architecture, allowing for system scalability and easy addition of new features. For development, the Unity framework and C# programming language were used. The MongoDB database was used to store information about users, games, ratings, and recommendations. The SteamKit SDK was used to interact with the Steam platform. Experiments and results. To evaluate the system’s effectiveness, testing was conducted on a small sample of users. Metrics such as Precision@k and Recall@k were calculated to assess the accuracy of recommendations. The results showed that the proposed approach achieves high recommendation accuracy, especially for users with well-formed gaming preferences. The use of tag combinations significantly improved the relevance of recommendations and satisfied a wider range of user requests. Comparison with baseline models. To evaluate the effectiveness of the proposed approach, a comparison was made with baseline recommendation models such as random selection and popularity-based recommendations. To assess the quality of recommendations, Precision@k and Recall@k metrics were used. The results showed that the proposed hybrid algorithm achieved significantly higher values of Precision@10 and Recall@10 compared to baseline models, indicating higher accuracy and completeness of recommendations. Impact of the size of the tag vocabulary. Increasing the size of the tag vocabulary allows for a more accurate description of games and user preferences. However, an excessive number of tags can lead to a sparse interaction matrix and complicate the model training process. Therefore, it is important to choose the optimal size of the tag vocabulary. Future research directions: ? Personalization of the interface: It is planned to develop a personalized recommendation system interface that will adapt to the individual needs of each user. ? Integration with other services: It is planned to integrate the system with other gaming platforms and social networks to expand data sources and improve the quality of recommendations. ? Gamification: It is planned to introduce gamification elements to increase user engagement with the system. ? Development of a web interface: It is planned to develop an intuitive web interface that will provide convenient access to the recommendation system from any device with Internet access. Key words: recommendation system, games, personalization, semantic analysis, tags, machine learning, MongoDB, Flask, Steam, integration, collaborative filtering, content-based analysis, Word2Vec, transformers, contextual recommendations, tag combinations.