Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Differentiable Fuzzy Neural Networks for Recommender Systems

Stephan Bartl, Kevin Innerebner, Elisabeth Lex

TL;DR

This work tackles the transparency gap in recommender systems by proposing differentiable fuzzy neural networks (FNNs) that learn weighted logic rules over fixed, human-readable atoms. The method uses differentiable fuzzy operators based on product t-norms to implement AND/OR/NOT, with a parallel-rule architecture that aggregates rule outputs into a final prediction, under a closed-world Horn-clause interpretation. Empirical results on synthetic data show exact recovery of ground-truth rules and perfect top-k metrics, while MovieLens 1M demonstrates competitive performance and clear interpretability, though SVD remains stronger overall. The approach advances human-centered explainability by delivering transparent reasoning paths with potential for hybrid integration with other neural models and future improvements in ranking-focused objectives and atom learning.

Abstract

As recommender systems become increasingly complex, transparency is essential to increase user trust, accountability, and regulatory compliance. Neuro-symbolic approaches that integrate symbolic reasoning with sub-symbolic learning offer a promising approach toward transparent and user-centric systems. In this work-in-progress, we investigate using fuzzy neural networks (FNNs) as a neuro-symbolic approach for recommendations that learn logic-based rules over predefined, human-readable atoms. Each rule corresponds to a fuzzy logic expression, making the recommender's decision process inherently transparent. In contrast to black-box machine learning methods, our approach reveals the reasoning behind a recommendation while maintaining competitive performance. We evaluate our method on a synthetic and MovieLens 1M datasets and compare it to state-of-the-art recommendation algorithms. Our results demonstrate that our approach accurately captures user behavior while providing a transparent decision-making process. Finally, the differentiable nature of this approach facilitates an integration with other neural models, enabling the development of hybrid, transparent recommender systems.

Differentiable Fuzzy Neural Networks for Recommender Systems

TL;DR

This work tackles the transparency gap in recommender systems by proposing differentiable fuzzy neural networks (FNNs) that learn weighted logic rules over fixed, human-readable atoms. The method uses differentiable fuzzy operators based on product t-norms to implement AND/OR/NOT, with a parallel-rule architecture that aggregates rule outputs into a final prediction, under a closed-world Horn-clause interpretation. Empirical results on synthetic data show exact recovery of ground-truth rules and perfect top-k metrics, while MovieLens 1M demonstrates competitive performance and clear interpretability, though SVD remains stronger overall. The approach advances human-centered explainability by delivering transparent reasoning paths with potential for hybrid integration with other neural models and future improvements in ranking-focused objectives and atom learning.

Abstract

As recommender systems become increasingly complex, transparency is essential to increase user trust, accountability, and regulatory compliance. Neuro-symbolic approaches that integrate symbolic reasoning with sub-symbolic learning offer a promising approach toward transparent and user-centric systems. In this work-in-progress, we investigate using fuzzy neural networks (FNNs) as a neuro-symbolic approach for recommendations that learn logic-based rules over predefined, human-readable atoms. Each rule corresponds to a fuzzy logic expression, making the recommender's decision process inherently transparent. In contrast to black-box machine learning methods, our approach reveals the reasoning behind a recommendation while maintaining competitive performance. We evaluate our method on a synthetic and MovieLens 1M datasets and compare it to state-of-the-art recommendation algorithms. Our results demonstrate that our approach accurately captures user behavior while providing a transparent decision-making process. Finally, the differentiable nature of this approach facilitates an integration with other neural models, enabling the development of hybrid, transparent recommender systems.
Paper Structure (15 sections, 6 equations, 2 figures, 4 tables)

This paper contains 15 sections, 6 equations, 2 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Methodology
  3. Fuzzy Operators
  4. Model Architecture
  5. Datasets
  6. Synthetic Dataset
  7. MovieLens 1M
  8. Evaluation
  9. Experimental Setup
  10. Preliminary Results
  11. Synthetic Dataset
  12. MovieLens 1M Dataset
  13. Conclusion
  14. Limitations.
  15. Future Work.

Figures (2)

  • Figure 1: Model architecture: Given a user-item pair as input, the model first computes predefined propositional atoms. These atoms are then fed into parallel layers, implementing fuzzy decision rules based on fuzzy neurons. The outputs of the layers are then aggregated via a fuzzy OR neuron, leading to the model's final output. This output classifies the item as either relevant or not relevant to the user.
  • Figure 2: Box plot of the values in the fuzzy weight matrix $\mathbf{W'}$ on the MovieLens 1M dataset. We observe a clear difference in the weights of the three most important atoms - HIGH AVG MOVIE RATING, HIGH AVG RATING PER USER, and MOVIE RATED OFTEN - compared to the remaining 77 atoms.