Node importance estimation (NIE) in heterogeneous knowledge graphs is a critical yet challenging task, essential for applications such as recommendation, knowledge reasoning, and question answering. Existing methods often rely on pairwise connections, neglecting high-order dependencies among multiple entities and relations, and they treat structural and semantic signals independently, hindering effective cross-modal integration. To address these challenges, we propose MetaHGNIE, a meta-path induced hypergraph contrastive learning framework for disentangling and aligning structural and semantic information. MetaHGNIE constructs a higher-order knowledge graph via meta-path sequences, where typed hyperedges capture multi-entity relational contexts. Structural dependencies are aggregated with local attention, while semantic representations are encoded through a hypergraph transformer equipped with sparse chunking to reduce redundancy. Finally, a multimodal fusion module integrates structural and semantic embeddings under contrastive learning with auxiliary supervision, ensuring robust cross-modal alignment. Extensive experiments on benchmark NIE datasets demonstrate that MetaHGNIE consistently outperforms state-of-the-art baselines. These results highlight the effectiveness of explicitly modeling higher-order interactions and cross-modal alignment in heterogeneous knowledge graphs. Our code is available at https://github.com/SEU-WENJIA/DualHNIE
Knowledge Graphs (KGs) have become a fundamental paradigm for representing complex, multi-typed real-world data [41,47]. Unlike homogeneous graphs with a single type of node and edge, heterogeneous KGs incorporate diverse entities, such as users, items, authors, and movies connected by rich semantic relations [14,39].
Within this context, a central problem is node importance estimation, which assigns importance scores to entities based on their structural and semantic roles in the graph [31,32], such as PageRank [30] assigns importance scores to visited pages on the World Wide Web. Accurate and efficient NIE is essential for numerous downstream applications [1], including recommendation systems [42], expert finding [45] and information retrieval [17]. Therefore, heterogeneous information in KGs represents a relevance and diverse resource to advance recommendation system for effective ranking and retrieval.
In HKGs, entity importance often depends not only on pairwise relations but also on complex correlations between user behaviors and semantic contexts. Traditional HKG models [15,29], which typically rely on meta-path-based neighbors in Figure 1, which captures only sequential relations between pairs of entities. This pairwise view overlooks collective interactions that naturally arise in real-world scenarios, such as a group of users jointly rating a movie or multiple authors co-authoring a paper [24,43]. These interactions involve more than two entities and are inherently higher-order interactions. To formally capture such multi-entity relations, hypergraph modeling provides a principled way to represent higher-order structures [21], enabling more faithful estimation of entity influence. Nevertheless, existing graph neural networks (GNNs), which rely on message passing across pairwise neighbors, are fundamentally limited in encoding these multi-ary relations. The absence of unified and scalable architectures for modeling higher-order dependencies in HKGs has thus become a key bottleneck, especially the development of foundation models for knowledge-aware applications.
Existing methods have been proposed to encode structural or semantic information in HKGs [21,29]. Techniques such as importance estimation with structural embedding of multiple entities adjusted via in-degree and similarity metrics [29,31]. Others emphasize semantic attributes [48] integrating semantics into the representation. However, node importance depends jointly on both topological position and semantic context. Moreover, some efforts to merge the two signals usually adopt attention-based mechanisms [2,29,37], but these approaches often treat heterogeneous information as a homogeneous flow, without explicitly disentangling structural and semantic modalities. This leads to insufficient utilization and suboptimal extraction of heterogeneous rich information.
To address these challenges, we introduce MetaHGNIE, a novel framework for node importance estimation in heterogeneous knowledge graphs. First, to model higher-order dependencies among multiple entities, We construct a meta-path-induced hypergraph, where typed hyperedges connect all entities within a meta-path. This representation explicitly captures multi-entity interactions that conventional pairwise GNNs overlook. Second, to disentangle and integrate heterogeneous information, MetaHGNIE employs a dual-channel architecture: a structure-aware hypergraph attention network aggregates local higher-order topological features, while a semantic hypergraph transformer encodes global contextual semantics. To reduce the computational cost of dense attention, we adopt a sparse-chunked attention mechanism that efficiently processes non-zero entries. Finally, structural and semantic embeddings are fused via contrastive alignment and auxiliary supervision to ensure robust cross-modal representations. Our main contributions are summarized as follows:
• We propose a Heterogeneous Higher-Order Knowledge Graph, a meta-path-induced hypergraph framework that explicitly models multi-entity interactions. Hyperedges connect all participating entities within a meta-path, with hyperedge types encoded as ordered relation tuples, addressing the limitation of pairwise-only methods. • We introduce a dual-channel encoding architecture that disentangles structural and semantic information: a structure-aware hypergraph attention network captures higher-order relations between multi-entities, while a contextual hypergraph transformer models semantic dependencies. • We develop a contrastive alignment and fusion mechanism that jointly optimizes structural and semantic embeddings with auxiliary supervision, ensuring effective integration of structural and semantic embeddings for accurate node importance estimation.
Estimating Node Importance on Graph. Estimating node importance in KGs is a fundamental task in graph representation learning. Existing methods can be categorized into two main paradigms: single-vie
This content is AI-processed based on open access ArXiv data.
