Abstract:This paper investigates the notion of learning user and item representations in non-Euclidean space. Specifically, we study the connection between metric learning in hyperbolic space and collaborative filtering by exploring Mobius gyrovector spaces where the formalism of the spaces could be utilized to generalize the most common Euclidean vector operations. Overall, this work aims to bridge the gap between Euclidean and hyperbolic geometry in recommender systems through metric learning approach. We propose HyperML (Hyperbolic Metric Learning), a conceptually simple but highly effective model for boosting the performance. Via a series of extensive experiments, we show that our proposed HyperML not only outperforms their Euclidean counterparts, but also achieves state-of-the-art performance on multiple benchmark datasets, demonstrating the effectiveness of personalized recommendation in hyperbolic geometry.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is how to use the metric learning method in hyperbolic space to improve the performance of personalized recommendation in the recommendation system. Specifically, the authors explored the possibility of learning user and item representations in non - Euclidean spaces (especially hyperbolic spaces), and by studying the connection between metric learning in hyperbolic space and collaborative filtering, proposed a new model named HyperML (Hyperbolic Metric Learning). This model aims to bridge the gap between Euclidean geometry and hyperbolic geometry in the recommendation system through the method of metric learning, thereby improving the performance of the recommendation system. ### Main Contributions: 1. **Explore Hyperbolic Space**: For the first time, attempt to use hyperbolic space for metric learning in the recommendation system, and use the Riemannian geometry of the Poincaré model and the Möbius gyrovector space to explore the application of hyperbolic space in the recommendation system. 2. **Propose the HyperML Model**: Design a new metric learning model HyperML for one - class collaborative filtering (i.e., personalized ranking). Different from previous metric learning models, HyperML introduces a penalty term called "distortion" to balance accuracy and distance preservation. 3. **Experimental Verification**: Verify the effectiveness of the HyperML model through extensive experiments on ten publicly available benchmark datasets. The experimental results show that HyperML not only significantly outperforms its Euclidean counterpart models, but also exceeds multiple competing baseline models, especially on standard ranking metrics, with a performance improvement of up to 32.32%. ### Method Overview: - **Hyperbolic Geometry and Poincaré Embedding**: Introduce the basic concepts of hyperbolic space and the Poincaré ball model, and explain the exponential expansion property of hyperbolic space and its advantages in the recommendation system. - **Möbius Gyrovector Space**: Describe in detail operations such as Möbius addition, scalar multiplication, exponential mapping, and logarithmic mapping, which make metric learning in hyperbolic space possible. - **Model Formulation**: Define the loss function for pulling positive sample pairs closer and pushing negative sample pairs apart, and introduce a distortion optimization function to maintain the global structure quality. Integrate these two parts through a multi - task learning framework to form the final optimization objective. - **Gradient Transformation**: Use the projected Riemannian stochastic gradient descent (RSGD) method for parameter learning, and explain in detail the transformation process from the Euclidean gradient to the Riemannian gradient. ### Experimental Results: - **Performance Improvement**: On multiple benchmark datasets, the HyperML model significantly outperforms other baseline models on both HR@10 and nDCG@10 evaluation metrics. - **Convergence Analysis**: By visualizing the embedding space, show the convergence behavior of the HyperML model in hyperbolic space, and verify the effectiveness and self - organizing ability of the model. In conclusion, this paper provides a new perspective and an effective solution for the recommendation system by introducing the metric learning method in hyperbolic space, significantly improving the performance of personalized recommendation.

HyperML: A Boosting Metric Learning Approach in Hyperbolic Space for Recommender Systems