Video game recommendation system
Students Name: Sartiukova Anastasiia Olehivna
Qualification Level: magister
Speciality: Systems and Methods of Decision Making
Institute: Institute of Computer Science and Information Technologies
Mode of Study: full
Academic Year: 2023-2024 н.р.
Language of Defence: ukrainian
Abstract: Recommender systems are software algorithms, usually based on machine learning, that use the processing of large amounts of data to provide and recommend additional products to consumers. These systems can rely on various factors, such as purchase history, search history, de-mographic data, and others, to help users find products and services that they might not have discovered without this assistance. The object of the study is the process of video game recommendation based on user preferences and video game characteristics. The subject of the study is the methods, technologies and software tools used to analyze user preferences and generate personalized video game recommendations. The purpose of the study is to increase the efficiency of the process of choosing video games by the user by introducing intelligent recommendation algorithms. Although there are a huge number of recommendation algorithms and techniques, most of them fall into the following broad categories: collaborative filtering, content filtering, and contextual filtering. Collaborative filtering algorithms recommend items based on information about the preferences of many users. This approach uses the similarity of user behavior, taking into account previous interactions between users and objects, recommendation algorithms learn to predict future interactions. These recommender systems build a model based on the user’s past behavior, such as items purchased previously or ratings given to those items, as well as similar decisions by other users. In contrast, content filtering uses attributes or features of an object to recommend other objects that are similar to the user’s preferences. This approach is based on the similarity of the characteristics of the item and the user, taking into account information about the user and the items with which he or she has interacted (for example, the user’s age, the average review of a movie), and models the probability of a new interaction. Hybrid recommender systems combine the advantages of the above types to create a more comprehensive recommender system. Contextual filtering incorporates users’ contextual information into the recommendation process. This approach uses the sequence of contextual actions of the user, as well as the current context, to predict the probability of the next action. Recommender systems are an important component of personalized user experiences, deeper customer engagement, and powerful decision support tools in retail, entertainment, healthcare, finance, and other industries. On some of the largest commercial platforms, recommendations generate up to 30% of revenue. Improving the quality of recommendations by 1% can generate billions of dollars in revenue. In the practical part of the work, three different recommender system algorithms were implemented, one content-based and two collaborative filtering algorithms (one with the ALS algorithm and the other with the EM and SVD algorithms). As shown in the Evaluation and Analysis section, for this project, the recommender system with the ALS algorithm provides the best recommendations based on the evaluation. Creating this project helped us better understand how a collaborative filtering system works. It really doesn’t use any information about objects, but relies entirely on user interaction with objects and matrix operations to generate recommendations. We needed to find an approach to work with the dataset (only user data for the collaborative recommender), since it contains only implicit data. Both approaches described in this paper handle implicit data in different ways to generate recommendations. For example, singular value decomposition (SVD) is used to remove some noise data as a dimensionality reduction method to facilitate the handling of a large dataset. On the other hand, it is clearly seen that the content-oriented approach requires the description of items to generate recommendations. Some problems were found in the implementation of the content-based recommender because it uses two different datasets. When the project started using two datasets (user and game), it was expected to find all the games available in the user dataset in the game dataset, since they both come from Steam. This creates a serious problem for a content-based recommender, as it relies on the assumption that all games available in the user dataset have information in the game dataset. Because of this, it is impossible to create recommendations for every game that the user has purchased. Keywords: recommendation system, collaborative filtering, content filtering, video game, Steam, matrix factorization. References. 1. Melville P., Sindhwani V. (2017) Recommender Systems. In: Sammut C., Webb G.I. (eds) Encyclopedia of Machine Learning and Data Mining. Springer, Boston, MA. https://doi.org/10.1007/978-1-4899-7687-1_964 2. Sohail, Shahab & Siddiqui, Jamshed & Ali, Rashid. (2017). Classifications of Recommender Systems: A review. JOURNAL OF ENGINEERING SCIENCE AND TECHNOLOGY REVIEW. 10. 132-153. 10.25103/jestr.104.18. 3. Ko, H., Lee, S., Park, Y., & Choi, A. (2022). A Survey of Recommendation Systems: Recommendation Models, Techniques, and Application Fields. Electronics, 11(1), 141. https://doi.org/10.3390/electronics11010141 4. Fayyaz, Z., Ebrahimian, M., Nawara, D., Ibrahim, A., & Kashef, R. (2020). Recommendation systems: Algorithms, challenges, metrics, and business opportunities. applied sciences, 10(21), 7748. https://doi.org/10.3390/app10217748 5. A. Konstan Recommender systems: from algorithms to user experience / J. A. Konstan J. A. // User Modeling and User-Adapted Interaction. – 2012 – Vol. 22. – No. 1–2. – P. 101–123.