2.англ. Ph.D. thesis: 296 p., 8 tables, 97 figures, 6 appendixes, 77 references.
RECOMMENDATION SYSTEM, COLLABORATIVE FILTERING, ALGORITHM, DATABASE, API, QUALITY, REVIEW, PARSING, MICRO-MARKUP, DATA, QUALIMETRY, EVALUATION, "ITEM TO ITEM", "USER TO USER", CYCLOGRAM, METRIC, INDICATOR, GOOGLE ANALYTICS, ATTRACTOR, PHASE PORTRAIT, DETERMINED CHAOS. The aim of the research is to improve the quality of recommendation systems with collaborative filtering based on qualimetric methods of measurement. The scientific justification is to improve quality by introducing additional parameters into the methods of forming recommendations by analyzing the “opinions” of users, which ensures an improvement in the quality of recommendation systems. To achieve the outlined goal, the following tasks need to be accomplished:
• Analyze recommendation systems and scientific publications regarding the set goal;
• Formulate principles for determining the quality of recommendation systems;
• Investigate the main approaches to structuring characteristics for evaluating the quality of recommendation systems;
• Implement a simple recommendation system and obtain results;
• Implement a recommendation system based on collaborative filtering algorithms;
• Formulate principles for determining the quality of reviews based on qualimetric methods;
• Implement a recommendation system based on collaborative filtering algorithms and improve it with qualimetric quality parameters that take into account the user's "opinion" - "The Value of Opinion";
• Propose a qualimetric method for evaluating the quality of recommendation systems based on research;
• Improve the quality of the collaborative filtering recommendation system based on the proposed qualimetric quality metrics;
• Propose qualimetric quality metrics based on the theory of deterministic chaos. The object of the study is the quality parameters of recommendation systems on the car search aggregator website Automoto.ua in the process of searching for car, motorcycle, special equipment, and other vehicle sales ads in Ukraine. Automoto.ua enables search for car sale offers throughout Ukraine, providing the most comprehensive and up-to-date results. Currently, the site processes information from over 100 Ukrainian car websites. Each day, the database contains more than 500,000 ads, 9-16 thousand of which are fresh arrivals for the current day. The subject of the research is the qualimetric method of quality evaluation in recommendation systems. Research methods - during the dissertation work, research methods based on system analysis, qualimetric measurement methods, quality evaluation methods of products, fuzzy logic theory, and deterministic chaos theory were used. Google Analytics tool was utilized. The scientific novelty of the obtained results:
1. For the first time, a qualimetric method for evaluating the quality of collaborative filtering recommendation systems in the sphere of provided internet services was developed, which, unlike known methods, covers business needs in terms of quality of recommendation and evaluation of result deviations simultaneously, allowing to form a comprehensive approach to quality assessment considering "opinions" based on review analysis.
2. The qualimetric method for evaluating the quality of recommendation systems has further developed by expanding the set of indicators compared to known ones, ultimately improving the quality of collaborative filtering recommendations in service provision to users.
3. For the first time in recommendation systems, the theory of deterministic chaos was used, which increased the accuracy of predicting user interaction with the recommendation system based on time series analysis using ARIMA and LSTM models.
4. The obtained results have not only been practically applied but also bring benefits to the business and its clients. During the research period since 2016, this product has been used over 2,000,000 times. The proposed collaborative filtering algorithm based on qualimetric measurement methods and the "Cyclogram of Quality of Recommendation Systems" can be used for any business tasks on the internet and for any goods, therefore they are universal for use and scalable. Based on the theory of deterministic chaos, the accuracy of predicting user interaction with the recommendation system was increased by 1.5% using the LSTM model compared to ARIMA.