Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

UIPC-MF: User-Item Prototype Connection Matrix Factorization for Explainable Collaborative Filtering

Lei Pan, Von-Wun Soo

TL;DR

UIPC-MF addresses the dual challenges of accuracy and explainability in collaborative filtering by learning two prototype banks for users and items and a global User-Item Prototypes Connections Matrix. The model computes a decomposable logit score as a linear combination of user-prototype similarities and item-prototype similarities, weighted by learned connections, enabling transparent rationales for recommendations under implicit feedback. It combines flexible base losses (BCE, BPR, SSM) with four interpretability terms and an L1 constraint on user preferences, achieving state-of-the-art performance among prototype-based methods across three datasets while enhancing transparency. The work demonstrates practical impact by enabling explainable recommendations and offering a pathway to integrate content-based signals and explicit feedback in future work.

Abstract

Recommending items to potentially interested users has been an important commercial task that faces two main challenges: accuracy and explainability. While most collaborative filtering models rely on statistical computations on a large scale of interaction data between users and items and can achieve high performance, they often lack clear explanatory power. We propose UIPC-MF, a prototype-based matrix factorization method for explainable collaborative filtering recommendations. In UIPC-MF, both users and items are associated with sets of prototypes, capturing general collaborative attributes. To enhance explainability, UIPC-MF learns connection weights that reflect the associative relations between user and item prototypes for recommendations. UIPC-MF outperforms other prototype-based baseline methods in terms of Hit Ratio and Normalized Discounted Cumulative Gain on three datasets, while also providing better transparency.

UIPC-MF: User-Item Prototype Connection Matrix Factorization for Explainable Collaborative Filtering

TL;DR

UIPC-MF addresses the dual challenges of accuracy and explainability in collaborative filtering by learning two prototype banks for users and items and a global User-Item Prototypes Connections Matrix. The model computes a decomposable logit score as a linear combination of user-prototype similarities and item-prototype similarities, weighted by learned connections, enabling transparent rationales for recommendations under implicit feedback. It combines flexible base losses (BCE, BPR, SSM) with four interpretability terms and an L1 constraint on user preferences, achieving state-of-the-art performance among prototype-based methods across three datasets while enhancing transparency. The work demonstrates practical impact by enabling explainable recommendations and offering a pathway to integrate content-based signals and explicit feedback in future work.

Abstract

Recommending items to potentially interested users has been an important commercial task that faces two main challenges: accuracy and explainability. While most collaborative filtering models rely on statistical computations on a large scale of interaction data between users and items and can achieve high performance, they often lack clear explanatory power. We propose UIPC-MF, a prototype-based matrix factorization method for explainable collaborative filtering recommendations. In UIPC-MF, both users and items are associated with sets of prototypes, capturing general collaborative attributes. To enhance explainability, UIPC-MF learns connection weights that reflect the associative relations between user and item prototypes for recommendations. UIPC-MF outperforms other prototype-based baseline methods in terms of Hit Ratio and Normalized Discounted Cumulative Gain on three datasets, while also providing better transparency.
Paper Structure (28 sections, 14 equations, 2 figures, 8 tables)

This paper contains 28 sections, 14 equations, 2 figures, 8 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Collaborative filtering
  4. Explainable Recommender Models
  5. The Prototype-based Collaborative Filtering
  6. Methodology
  7. User-Item Prototypes Connections Matrix Factorization (UIPC-MF)
  8. Loss function
  9. Binary Cross-Entropy Loss
  10. Bayesian Personalized Ranking Loss
  11. Sampled Softmax Loss
  12. Interpretability Terms
  13. $L^{1}$-Norm for Users' Preference Scores
  14. Experiments and Discussion
  15. Evaluation Metrics
  16. ...and 13 more sections

Figures (2)

  • Figure 1: UIPC-MF Model. The red lines show the used connection weights to obtain user preference value for a particular item prototype 1, $r_{1}$.
  • Figure 2: The left side shows the visualization of the user preference value $r$, $\text{sim}(t,p^t)$, and the item prototype score $s$, while the right side presents the boxplots with the range of user preference values in UIPC-MF (top) and UIPC-MF-$L^1$ (bottom).