Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A First Principles Approach to Trust-Based Recommendation Systems in Social Networks

Paras Stefanopoulos, Sourin Chatterjee, Ahad N. Zehmakan

TL;DR

This work presents a first-principles analysis of trust-based recommender systems in social networks by separating three information forms: trust graphs, intra-item similarities, and item-rating data. It systematically builds single-information and multi-information models using a Weighted Average framework, showing that item-rating information dominates accuracy while trust signals confer robustness against adversarial attacks; intra-item information offers stability and constructive, additiveValue when combined. The strongest overall performer across datasets is the Jaccard Item-Jaccard WA method, which integrates trust and item-rating signals in a simple, scalable way. The study highlights the value of transparent, modular designs for understanding information contributions in recommendations and points to practical directions for robust, cold-start friendly systems.

Abstract

This paper explores recommender systems in social networks which leverage information such as item rating, intra-item similarities, and trust graph. We demonstrate that item-rating information is more influential than other information types in a collaborative filtering approach. The trust graph-based approaches were found to be more robust to network adversarial attacks due to hard-to-manipulate trust structures. Intra-item information, although sub-optimal in isolation, enhances the consistency of predictions and lower-end performance when fused with other information forms. Additionally, the Weighted Average framework is introduced, enabling the construction of recommendation systems around any user-to-user similarity metric. All the codes are publicly available on GitHub.

A First Principles Approach to Trust-Based Recommendation Systems in Social Networks

TL;DR

This work presents a first-principles analysis of trust-based recommender systems in social networks by separating three information forms: trust graphs, intra-item similarities, and item-rating data. It systematically builds single-information and multi-information models using a Weighted Average framework, showing that item-rating information dominates accuracy while trust signals confer robustness against adversarial attacks; intra-item information offers stability and constructive, additiveValue when combined. The strongest overall performer across datasets is the Jaccard Item-Jaccard WA method, which integrates trust and item-rating signals in a simple, scalable way. The study highlights the value of transparent, modular designs for understanding information contributions in recommendations and points to practical directions for robust, cold-start friendly systems.

Abstract

This paper explores recommender systems in social networks which leverage information such as item rating, intra-item similarities, and trust graph. We demonstrate that item-rating information is more influential than other information types in a collaborative filtering approach. The trust graph-based approaches were found to be more robust to network adversarial attacks due to hard-to-manipulate trust structures. Intra-item information, although sub-optimal in isolation, enhances the consistency of predictions and lower-end performance when fused with other information forms. Additionally, the Weighted Average framework is introduced, enabling the construction of recommendation systems around any user-to-user similarity metric. All the codes are publicly available on GitHub.
Paper Structure (34 sections, 10 equations, 5 figures, 13 tables, 2 algorithms)

This paper contains 34 sections, 10 equations, 5 figures, 13 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Traditional Approaches
  4. Cold-Start Approaches
  5. Machine Learning Approaches
  6. Adversarial Defense Approaches
  7. Our Contribution
  8. Methodology
  9. Primary Training Data Set
  10. Similarity Metrics
  11. Jaccard Index
  12. Jaccard-Item Index
  13. Item-Rating Difference
  14. Pearson Correlation
  15. Intra-Item Jaccard
  16. ...and 19 more sections

Figures (5)

  • Figure 1: An example of three types of information. 4 individuals whose social structure is shown via edges. Rating corresponding to each individual and overall similarity metric among items provided using tables.
  • Figure 2: Schematic diagram of building a recommendation system based on different subsets of data and combining them. In particular, cold-start compatible approaches are outlined in red.
  • Figure 3: Distribution of ratings for all items
  • Figure 4: Distribution of MAE for Control Recommenders
  • Figure 5: MAE distributions for different recommender algorithms. The dotted lines depict the performance with an adversarial network and the solid lines show the same recommender under normal conditions.