MIARec: Mutual-influence-aware Heterogeneous Network Embedding for Scientific Paper Recommendation

TL;DR

MIARec tackles the challenge of recommending scientific papers by modeling asymmetric academic influence within heterogeneous scholarly networks. It introduces a gravity-based mutual-influence factor and a multi-channel embedding framework with independent and interdependent channels to capture both within-relational and cross-relational signals, fused by attention. Paper content is integrated via Doc2Vec and aligned to scholar embeddings with an MLP, trained using Bayesian Personalized Ranking to optimize ranking of relevant papers. Empirical results on Aminer DBLP-V12 and ACM-V10 show consistent improvements over strong baselines, with ablations confirming the contribution of each module, including the interdependent channel, mutual-influence factor, and textual content. The approach offers a principled, scalable way to encode asymmetric scholarly influence for improved recommendation in real-world scientific networks.

Abstract

With the rapid expansion of scientific literature, scholars increasingly demand precise and high-quality paper recommendations. Among various recommendation methodologies, graph-based approaches have garnered attention by effectively exploiting the structural characteristics inherent in scholarly networks. However, these methods often overlook the asymmetric academic influence that is prevalent in scholarly networks when learning graph representations. To address this limitation, this study proposes the Mutual-Influence-Aware Recommendation (MIARec) model, which employs a gravity-based approach to measure the mutual academic influence between scholars and incorporates this influence into the feature aggregation process during message propagation in graph representation learning. Additionally, the model utilizes a multi-channel aggregation method to capture both individual embeddings of distinct single relational sub-networks and their interdependent embeddings, thereby enabling a more comprehensive understanding of the heterogeneous scholarly network. Extensive experiments conducted on real-world datasets demonstrate that the MIARec model outperforms baseline models across three primary evaluation metrics, indicating its effectiveness in scientific paper recommendation tasks.

Figures (5)

• Figure 1: The framework of proposed MIARec model. (a) The scholar representation learning part. The heterogeneous network is decomposed into single-relational sub-networks, each embedded in an independent channel via a mutual-influence-aware embedding module. These sub-networks share embedding parameters within a common interdependent embedding channel and are then aggregated. An attention module fuses these aggregated embeddings into a unified representation for scholars $U$. (b) The paper representation learning part. Scientific papers' embeddings are learned through Doc2Vec. (c) The recommendation prediction part. The model employs a multi-layer perceptron (MLP) to align the two types of embedding representations and performs recommendations based on their inner product similarity.
• Figure 2: Performance comparison of network embedding methods within MIARec framework.
• Figure 3: Ablation study of the impact of individual modules on MIARec Performance.
• Figure 4: Ablation study of the impact of different scholar relationships on MIARec Performance.
• Figure 5: Performance of MIARec models with different batch sizes.