Mitigating Data Sparsity via Neuro-Symbolic Knowledge Transfer

Data sparsity is a well-known historical limitation of recommender systems that still impacts the performance of state-of-the-art approaches. The literature proposed various ways to mitigate the problem by providing additional information to the model (e.g., hybrid recommendation, knowledge graph-based systems). In particular, one promising technique involves transferring information from other domains or tasks to compensate for sparsity in the target domain, where the recommendations must be performed. Following this idea, we propose a novel approach based on Neuro-Symbolic computing designed for the knowledge transfer task in recommender systems. In particular, we use a Logic Tensor Network (LTN) to train vanilla Latent Factor Models (LFMs) for rating prediction. We show how the LTN can be used to regularize the LFMs using axiomatic knowledge that permits injecting pre-trained information learned by Collaborative Filtering on a different task or domain. Extensive experiments comparing our models with different baselines on two versions of a novel real-world dataset prove our proposal's potential in the knowledge transfer task. In particular, our models outperform the baselines, including those that can encode additional information, suggesting that the knowledge is effectively transferred to the target domain via logical reasoning. Moreover, an experiment that drastically decreases the density of user-item ratings shows that the benefits of the acquired knowledge increase with the sparsity of the dataset, showing the importance of exploiting knowledge from a denser source of information when training data is scarce in the target domain.