Abstract
The performance of trust models highly depend on the characteristics of the environments where they are applied. Thus, it becomes challenging to choose a suitable trust model for a given e-marketplace environment, especially when ground truth about the agent (buyer and seller) behavior is unknown (called unknown environment). We propose a case-based reasoning framework to choose suitable trust models for unknown environments, based on the intuition that if a trust model performs well in one environment, it will do so in another similar environment. Firstly, we build a case base with a number of simulated environments (with known ground truth) along with the trust models most suitable for each of them. Given an unknown environment, case-based retrieval algorithms retrieve the most similar case(s), and the trust model of the most similar case(s) is chosen as the most suitable model for the unknown environment. Evaluation results confirm the effectiveness of our framework in choosing suitable trust models for different e-marketplace environments.
Introduction
Trust models are critical for identifying malicious sellers and unfair ratings in e-marketplaces. However, their effectiveness varies significantly depending on the specific environment. For instance, a model like BRS might excel when most advisors are honest, while BLADE performs better in sparse environments with subjective rating differences.
The primary challenge in real-world "unknown" environments is the lack of "ground truth"—we do not know which sellers are malicious or which ratings are deceptive. Without this information, selecting the optimal trust model is difficult. This paper addresses this gap by proposing a Case-Based Reasoning (CBR) framework that uses environmental similarity as a proxy for performance: if a trust model works well in a specific simulated environment, it is likely to perform well in a similar real-world one.
The CBR Framework
The proposed framework utilizes a CBR cycle consisting of four stages: Retrieve, Reuse, Revise, and Retain. The process is divided into two main phases: building the case base and case retrieval.
1. Case Base Construction
The authors developed a library of 2,268 simulated environments. For each simulation, they evaluated a pool of 45 candidate trust models (variations of 7 major models, including BRS, TRAVOS, and iCLUB) to identify the top performer.
2. Feature Extraction and Selection
To compare environments, 18 statistical features were initially identified (e.g., variance of ratings per seller, ratio of buyers to sellers). Through correlation and regression analysis, this set was refined to the 13 most influential features.
3. Retrieval and Weighting
Three algorithms were compared for finding the "most similar" case:
- k-Nearest Neighbors (k-NN): A flat search offering high accuracy.
- K-dimensional (K-d) Trees: A hierarchical structure that significantly speeds up retrieval.
- Decision Trees: An inductive approach that learns classification rules.
Additionally, a linear adaptive filter was implemented to assign weights to features, recognizing that certain characteristics (like the percentage of satisfactory sellers) are more predictive of model performance than others.
Experimental Setup
The framework was evaluated using both simulated and real-world data:
- Simulated Environments: 2,268 unique scenarios varying parameters like agent count, honesty levels, and attack types (e.g., RepBad, RepTrap).
- Real-World Data (IMDB): Movie rating data (188 users, 40 directors, 1,142 movies) was used to test real-world applicability. Unfair rating attacks were simulated on top of these real rating distributions.
Key Results and Insights
The evaluation demonstrated that no single trust model is universally superior; the "best" model is a moving target depending on the attack pattern. Key findings include:
- High Selection Accuracy: In unknown random environments, the framework achieved 97% accuracy in choosing the correct trust model using k-NN/K-dT with $k=3$.
- Real-World Robustness: On the IMDB dataset, the framework maintained over 86% accuracy, outperforming any single fixed trust model.
- Increased Buyer Utility: Buyers using framework-selected models achieved higher rewards for successful transactions compared to using a simple average or a sub-optimal model.
- Computational Efficiency: K-d trees reduced retrieval time by 60-70% compared to standard k-NN without sacrificing accuracy, making the system viable for large-scale applications.
- Algorithm Performance: k-NN and K-d trees outperformed Decision Trees, suggesting that the boundaries for optimal model selection are too complex for simple decision rules.
Conclusion and Future Work
This research demonstrates that similarity to known simulated cases is a powerful proxy for selecting trust models in environments where ground truth is unavailable. However, the framework's success depends on the diversity of the simulated case base.
Future work will focus on handling dynamic environments where behavior changes rapidly over time and identifying new features to counter emerging types of e-marketplace attacks.
Summary updated