Unbalanced Multistage Heat Conduction and Mass Diffusion Algorithm in an Educational Digital Library

Discovering valuable and diverse resources in an educational digital library can be difficult with the existing large number of content collections. Researchers in information retrieval and related domains have moved from considering only keyword-based matching to modeling the underlying behavior pattern of users. To achieve the purpose of both precision and diversity while providing online educational resource services, in this paper, we propose a weighted network-based information filtering framework that models user usage as a bipartite user-resource network; users and resources are treated as nodes in this network, each edge from a user to a resource means usage, and the weight represents the accumulation of multiple usage scenarios. Under this framework, we propose two individual algorithms, the unbalanced heat conduction algorithm and the unbalanced mass diffusion algorithm, and one hybrid multistage heat conduction and mass diffusion algorithm. There are two stages in these algorithms. In stage 1, an initial energy is assigned to each resource that has been visited by the target and passes to users according to a specific strategy. In stage 2, the energy is similarly transferred from users to resources, and a sorted resource list ranked by energy is presented to the target user. Experiments on a real-world dataset of one year of academic search logs showed improved performance from multiple indicators compared to existing algorithms.