New Deep Neural Network Methods and Collaborative Filtering Algorithms for Recommender Systems
Project Reference :
National University of Singapore (NUS)
Principal Investigator :
Dr Huang Zhiyong
Technology Readiness :
4 (Technology validated in lab)
Technology Categories :
The recommendation engine market was valued at USD 2.12 billion in 2020, and it is expected to reach USD 15.13 billion by 2026, registering a CAGR of 37.46% during the period of 2021-2026, with Asia-Pacific identified as the fastest-growing market.
Item recommendation is a personalised ranking task. To this end, many recommender systems optimise models with pairwise ranking objectives such as the Bayesian Personalised Ranking (BPR).
Optimising matrix factorisation (MF), the most widely used model in recommendation, with BPR leads to a recommender model that is not robust. In particular, the resultant model is highly vulnerable to adversarial perturbations on its model parameters, which implies a possibly large error in generalisation.
Incorporating knowledge graph into recommender systems: By exploring the interlinks within a knowledge graph, the connectivity between users and items can be discovered as paths, which provide rich and complementary information to user-item interactions. Such connectivity not only reveals the semantics of entities and relations, but also helps to comprehend a user’s interest. However, existing efforts have not fully explored this connectivity to infer user preferences, especially in terms of modeling the sequential dependencies within and holistic semantics of a path.
Two state-of-the-art recommender systems have been developed to address the above two problem statements.
- Adversarial Personalized Ranking (APR), is a new optimisation framework to enhance the robustness of a recommender model and thus improve its generalization performance. APR enhances the pairwise ranking method BPR, by performing adversarial training. The minimisation of the BPR objective function defends an adversary, which adds adversarial perturbations on model parameters to maximise the BPR objective function. To illustrate how it works, APR has been implemented on MF by adding adversarial perturbations on the embedding vectors of users and items. Extensive experiments on three public real-world datasets demonstrate the effectiveness of APR. By optimising MF with APR, it outperforms BPR with a relative improvement of 11.2% on average and achieves state-of-the-art performance for item recommendation. The implementation is available here.
- A new model named Knowledge-aware Path Recurrent Network (KPRN) has been developed to exploit knowledge graph for recommendation. KPRN can generate path representations by composing the semantics of both entities and relations. By leveraging the sequential dependencies within a path, we allow effective reasoning on paths to infer the underlying rationale of a user-item interaction. Furthermore, a new weighted pooling operation to discriminate the strengths of different paths in connecting a user with an item has been designed, endowing the model with a certain level of explainability. Extensive experiments have been conducted on two datasets (movie and music), demonstrating significant improvements over state-of-the-art solutions.
NUS welcomes interest from the industry in the co-development / customisation of the technology into a new product/ service.
- Better predictive CTR (click-through rate) than the existing methods
- Reasonable explanations on why the user clicks the ads in the form of feature attributions
- Design of effective and efficient tools to preprocess large, unstructured data
The work has created an impact in fundamental research in personalised recommendation and can be directly used in online customer-oriented services such as healthcare, fintech and e-commerce.
We welcome interest from the industry for collaboration/ co-development / customisation of the technology into a new product or service. If you have any enquiries or are keen to collaborate, please contact us.