The objective of this research is to introduce the first application of Belief Propagation (BP), an iterative probabilistic algorithm, to both reputation and recommender systems. The research departs from the traditional approaches to solve for reputation and recommender problems in one major aspect: it formulates both problems as the inference problem while maintaining the dependency of the statistical data. This problem, however, cannot be solved in a large-scale reputation (and recommender) systems, because the number of terms grow exponentially with the number of nodes (i.e., users/items in reputation and recommender systems). The key benefit of the BP algorithm is that it solves the problem with the complexity that grows only linearly with the number of nodes.

 

The research views both the reputation management and recommendation problems as the inference problem involving the computation of the marginal probability distribution functions (of the reputation values or ratings to be predicted) from complicated global functions of many variables. This view maintains the statistical dependency of the data, and hence, results in robust and accurate systems. To compute these marginal distributions, the research then employs BP whose computational efficiency is derived by exploring the way in which the statistical data encoded on some forms of large graphical models, e.g., factor graphs. In particular, the project includes research to:
(1) study the general theories of BP-based reputation management and recommender systems on various graphical models and develop novel algorithms;
(2) study the convergence, scalability, and robustness of the developed algorithms via the mathematical analysis and intensive simulations;
(3) evaluate the developed algorithms and compare them with the current state of art using real-life datasets and conducting user studies; and
(4) adaptively learn various attack strategies against the reputation and recommender systems, determine the impacts of such attacks, and decrease their impact.