Publication Search

The exponential growth of digital music catalogs on streaming platforms such as Spotify has made personalized recommendation systems crucial for enhancing user experience. This study develops a hybrid music recommendation system that addresses both warm-user and cold-user scenarios by combining Alternating Least Squares (ALS) collaborative filtering with content-based filtering (CBF) augmented by a popularity component. The dataset consists of 8,549,544 user-track interactions and a master file of 1,204,025 tracks with ten audio features. After preprocessing, users were segmented into 14,880 warm users and 723 cold users based on a five-interaction threshold. The ALS model was trained on the user-item implicit feedback matrix and tuned through grid search over factors, alpha, and regularization. CBF was implemented using cosine similarity on normalized audio features, while popularity scores were applied for new users with insufficient history. Evaluation used Precision@10, Recall@10, and NDCG@10. The final ALS configuration achieved NDCG@10 of 0.1116, representing a 30% improvement over baseline, while the hybrid CBF improved NDCG@10 for cold users from 0.0070 to 0.0201. Findings indicate that adaptive routing among ALS, CBF, and popularity reliably handles different user states, providing a practical foundation for production-grade music recommendation systems.

The rapid development of digital technology has significantly transformed commercial activities, particularly through the emergence of marketplaces as platforms for online transactions. The vast number of products available in a marketplace often creates difficulties for users in finding items that suit their needs and preferences. To address this challenge, a recommendation system is required to provide personalized and relevant product suggestions. This study discusses the implementation of a product recommendation system in a marketplace using the Collaborative Filtering method. This method works by leveraging information from users’ previous behavior, such as purchase history, ratings, and similarity of preferences with other users, to generate more accurate product recommendations. The Collaborative Filtering approach has proven effective in identifying user preference patterns based on relationships between users as well as between items. This study employs user interaction data such as ratings and shopping activities as the processing foundation. The process involves data collection, preprocessing, calculation of similarity between users or products, and generating recommendation lists. The results indicate that this method enhances the shopping experience by providing relevant product suggestions tailored to user interests, thereby increasing customer satisfaction and potentially improving sales performance in the marketplace. Thus, the application of a Collaborative Filtering-based recommendation system not only simplifies product discovery for users but also offers strategic advantages for marketplace operators in digital business competition

This research presents a personalized, context-aware recommender system to suggest Places of Interest (POIs) using a hybrid approach combining Bayesian inference and collaborative filtering. The system explicitly addresses the cold-start problem that new users face and improves recommendation accuracy by considering contextual variables such as user mood, budget, companion, and location. The system collects real-time contextual inputs for new users with no historical data and applies Bayesian inference to generate relevant POI suggestions. As users begin to interact and provide ratings, the system progressively shifts to a collaborative filtering mechanism, leveraging cosine similarity to identify similar users within comparable contexts. The recommender system focuses on three categories of POIs: restaurants, hotels, and landmarks. These locations are retrieved through the Google Maps API, and only mapped locations are considered. The system was implemented on Android devices and evaluated through a user study involving 25 participants from diverse backgrounds, including software developers, IT students, and general users. Evaluation metrics such as normalized Discounted Cumulative Gain (nDCG) and classification accuracy were used to assess recommendation quality. Results demonstrate that the system performs better than traditional methods, with nDCG improvements reaching up to 83 percent. Users reported high satisfaction regarding the recommendations' accuracy, ease of use, and contextual relevance. While the system offers significant improvements, it also has certain limitations. Its dependency on Google Maps data may restrict its scope, and using only four contextual factors limits the system’s adaptability to more complex user preferences. Future enhancements could include additional dynamic contexts such as weather, POI popularity, and time-related trends, as well as integrating more advanced models to increase personalization and flexibility in real-world applications.

Item-based collaborative filtering is a popular technique in recommendation systems that aims to provide suggestions for films to watch or services to users based on similarities between items. In this approach, the similarity between items is calculated using metrics such as cosine similarity, allowing the prediction of user preferences for items that have never been rated. This research implements Item-based collaborative filtering using datasets from Kaggle. Experimental results show that the resulting model is able to provide recommendations with significant improvements in evaluation metrics such as Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) of 3.05 and 3.26. This shows that the smaller the value, the better.

The objective of this research is to devise a personalized recommendation system for a freelancing platform to optimize the freelancer project matching process. This enhancement is intended to improve user experience and increase the success rate of projects. The system will recommend projects to freelancers based on their skills and preferences by employing data analysis and machine learning methodologies. The research methodology adheres to the Cross Industry Standard Process for Data Mining (CRISP-DM), incorporating six stages: Business Understanding, Data Understanding, Data Preparation, Modeling, Evaluation, and Deployment. The proposed project employs a hybrid recommendation strategy, integrating Content-Based Filtering through KNearest Neighbors (K-NN) and Cosine Similarity, Collaborative Filtering via Singular Value Decomposition (SVD), and recommendations derived from Word2vec. Evaluation metrics such as precision, recall, F1 score, MAP, and MRR are utilized to assess model performance. The results, including precision scores of 0.80 for KNN and 0.728 for SVD, recall scores of 0.60 for KNN and 0.623 for SVD, and F1 scores of 0.69 for KNN and 0.671 for SVD, as well as a MAP of 0.75 and MRR of 0.80 for Word2vec, demonstrate the efficacy of the hybrid recommendation system in delivering accurate and varied project suggestions to freelancers, with a weighted average ensemble learning model emerging as the most effective solution.