A fast combination method in DSmT and its application to recommender system.

Area of Science:

• Artificial Intelligence
• Information Fusion
• Decision Theory

Background:

• Epistemic uncertainties are often modeled using belief functions, requiring approximation to Bayesian BBAs for probabilistic systems.
• This approximation is crucial for integrating belief function outputs into probabilistic frameworks like Bayesian inference and decision-making processes.

Purpose of the Study:

• To introduce a novel, fast combination method, Modified Rigid Coarsening (MRC), for approximating general basic belief assignments (BBAs) to Bayesian BBAs.
• To reduce computational complexity in belief function combination through hierarchical decomposition and efficient coarsening of focal elements.

Main Methods:

• The Modified Rigid Coarsening (MRC) method utilizes hierarchical decomposition of the frame of discernment.
• It employs a disagreement vector and a dichotomous approach for efficient coarsening of focal elements.
• The method's practicality is demonstrated by applying it to combine user preferences in recommender systems (RSs).

Main Results:

• Experiments show that MRC is more effective in improving recommendation accuracy compared to the original Rigid Coarsening (RC) method.
• MRC demonstrates comparable computational time to existing methods.
• Performance was benchmarked against the proportional conflict redistribution rule #6 (PCR6).

Conclusions:

• The Modified Rigid Coarsening (MRC) method provides an efficient and accurate approach for converting general BBAs to Bayesian BBAs.
• MRC offers a practical solution for applications requiring integration with probabilistic frameworks, such as recommender systems.
• The method effectively balances accuracy and computational efficiency in information fusion tasks.