Doctoral Researcher Hồ Thị Hoàng Vy successfully defended a doctoral thesis in Information Systems entitled “Research and Proposals for Neighbourhood-based Collaborative Filtering Methods in Recommender Systems” at VNUHCM–University of Science (HCMUS) on 5 November 2026. The thesis was supervised by Assoc. Prof. Lê Nguyễn Hoài Nam and Dr. Phạm Nguyễn Cương.

Neighbourhood-based collaborative filtering is one of the widely used approaches to developing recommender systems. The method predicts user preferences based on similarities with other users, offering an intuitive and interpretable approach. However, existing methods still face several limitations concerning the aggregation of preferences from neighbouring users, similarity measurement, sparse data handling and scalability when deployed in large-scale systems.

In response to these challenges, the thesis proposes solutions to improve preference prediction and user similarity measurement. For preference aggregation, the research develops objective functions to learn the influence weights of neighbouring users on a target user. Notably, the degree of influence is determined for each individual item, rather than being based solely on the overall similarity between two users. The proposed methods also draw on both rating scores and review content to support preference prediction.

For similarity measurement, the thesis investigates both explicit and implicit feedback data. Explicit feedback is approached through a multi-factor framework, allowing different dimensions of product evaluation to be taken into account. For implicit feedback, different types of user interactions are distinguished and analysed separately according to the extent to which each reflects user preferences. In addition, the research proposes a mechanism for transforming user representation spaces to reduce data sparsity, thereby improving the effectiveness of similarity measurement.

Another research direction explored in the thesis is enhancing the scalability of recommender systems through a user clustering mechanism with an informed initialisation method. This approach narrows the computational search space during training, reduces the number of user pairs requiring consideration, and maintains the quality of neighbourhood selection and recommendation performance.

The research findings contribute new approaches to preference aggregation, similarity measurement and computational optimisation in neighbourhood-based collaborative filtering recommender systems. The proposed methods have potential applications in developing recommender systems across various fields. The thesis also identifies further research directions, including extending the methods to group recommendation and integrating contextual and temporal data to further enhance recommender system performance.