Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Modeling Behavioral Patterns in News Recommendations Using Fuzzy Neural Networks

Kevin Innerebner, Stephan Bartl, Markus Reiter-Haas, Elisabeth Lex

TL;DR

This work tackles the transparency gap in news recommender systems by introducing a fuzzy neural network (FNN) that learns human‑readable rules from behavioral data to predict article clicks, with rule complexity controlled via a threshold $t$. The approach combines multi‑layer fuzzy logic (AND/OR/NOT) with learned weights in $[-1,1]$ (via a $\tanh$ transform) and end‑to‑end training, enabling negations and rule extraction. Evaluations on EB‑NeRD and MIND show competitive predictive performance, especially on EB‑NeRD, while yielding interpretable rules that reveal recency and topical patterns in news consumption; ablation and sensitivity analyses demonstrate the impact of components on rule complexity and predictive power. The paper also presents a Theory of Change, positioning the model as a decision‑support tool for editors, and outlines future work toward integrating manual editorial guidelines and enabling tractable auditing through decomposability and determinism.

Abstract

News recommender systems are increasingly driven by black-box models, offering little transparency for editorial decision-making. In this work, we introduce a transparent recommender system that uses fuzzy neural networks to learn human-readable rules from behavioral data for predicting article clicks. By extracting the rules at configurable thresholds, we can control rule complexity and thus, the level of interpretability. We evaluate our approach on two publicly available news datasets (i.e., MIND and EB-NeRD) and show that we can accurately predict click behavior compared to several established baselines, while learning human-readable rules. Furthermore, we show that the learned rules reveal news consumption patterns, enabling editors to align content curation goals with target audience behavior.

Modeling Behavioral Patterns in News Recommendations Using Fuzzy Neural Networks

TL;DR

This work tackles the transparency gap in news recommender systems by introducing a fuzzy neural network (FNN) that learns human‑readable rules from behavioral data to predict article clicks, with rule complexity controlled via a threshold . The approach combines multi‑layer fuzzy logic (AND/OR/NOT) with learned weights in (via a transform) and end‑to‑end training, enabling negations and rule extraction. Evaluations on EB‑NeRD and MIND show competitive predictive performance, especially on EB‑NeRD, while yielding interpretable rules that reveal recency and topical patterns in news consumption; ablation and sensitivity analyses demonstrate the impact of components on rule complexity and predictive power. The paper also presents a Theory of Change, positioning the model as a decision‑support tool for editors, and outlines future work toward integrating manual editorial guidelines and enabling tractable auditing through decomposability and determinism.

Abstract

News recommender systems are increasingly driven by black-box models, offering little transparency for editorial decision-making. In this work, we introduce a transparent recommender system that uses fuzzy neural networks to learn human-readable rules from behavioral data for predicting article clicks. By extracting the rules at configurable thresholds, we can control rule complexity and thus, the level of interpretability. We evaluate our approach on two publicly available news datasets (i.e., MIND and EB-NeRD) and show that we can accurately predict click behavior compared to several established baselines, while learning human-readable rules. Furthermore, we show that the learned rules reveal news consumption patterns, enabling editors to align content curation goals with target audience behavior.
Paper Structure (15 sections, 1 equation, 3 figures, 3 tables, 1 algorithm)

This paper contains 15 sections, 1 equation, 3 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Interpretable Rule Learning with Fuzzy NNs
  3. Learning News Behavior for Editorial Oversight
  4. Article Age Atoms
  5. EB-NeRD Atoms
  6. MIND Atoms
  7. Results
  8. Quantitative Results
  9. Ablation Study
  10. Sensitivity Analysis
  11. Interpretation of Rules
  12. Qualitative Analysis of the Extracted Rules
  13. Theory of Change: A transparent, fuzzy neural news recommender for editorial oversight
  14. Conclusion and Future Work
  15. Acknowledgments.

Figures (3)

  • Figure 1: Architecture of our model design. User and article features are converted to fuzzy sets, serving as the input. Our model features multiple layers of logic nodes that perform product t-norm operations ($\land$ or $\lor$) on their inputs. Each input to the logic node is weighted so that high weights result in a strong influence of the input, while low weights mean the input has little influence on the computation. Inputs are negated ($\lnot$) when the corresponding weight is negative.
  • Figure 2: Analysis of parameter sensitivity (training) and threshold (prediction).
  • Figure : A rule extraction algorithm. Removes all connections with a weight below the threshold $t$, thus mapping the fuzzy logical operations to crisp logical operators.