Journal article
From patents to partners: Recommending organizations’ technological partners via representation learning from temporal heterogeneous graphs
Information processing & management, Vol.63(2 Part B), 104458
03/2026
DOI: 10.1016/j.ipm.2025.104458
Abstract
Due to high search costs and the risks of technological innovation, organizations urgently need efficient methods for recommending technological partners. However, existing models for technological partner recommendations overlooked multiple types of relationships among multiple types of entities and treated complex relationships among these entities as discrete temporal graphs, leading to potential information loss. To address these issues, we propose a novel technological partner recommendation model based on learning node embeddings from a continuously evolving temporal heterogeneous graph among organizations, patents, and technological fields. Multiple meta-paths are designed to reflect characteristics of technological collaboration scenarios. Also, two novel constructs-time-constrained paths and distance-constrained edges-are proposed in this paper and selected through a Hawkes process. Attention mechanisms further helped the model incorporate graph structures, patent semantics, and temporal patterns into the learning of node embeddings. Experiments on patent datasets from the new energy vehicle sector in China and the biomedical engineering sector in the United States show that, for cut-off thresholds of 3–20, our model improves nDCG over the strongest baseline by an average of 3.30% and 1.18%, respectively. This study has important implications for organizations to identify potential partners for technological collaboration in the ever-changing technological landscape.
Details
- Title: Subtitle
- From patents to partners: Recommending organizations’ technological partners via representation learning from temporal heterogeneous graphs
- Creators
- Yang YuJiang WuKang Zhao
- Resource Type
- Journal article
- Publication Details
- Information processing & management, Vol.63(2 Part B), 104458
- DOI
- 10.1016/j.ipm.2025.104458
- ISSN
- 0306-4573
- eISSN
- 1873-5371
- Publisher
- Elsevier
- Grant note
- National Natural Science Foundation of China: 72204189 Wuhan University Digital Intelligence Humanities Foundation: 2024SZWK023
This research work was supported by the National Natural Science Foundation of China (Grant No. 72232006) , the National Natural Science Foundation of China (Grant No. 72204189) , the Wuhan University Digital Intelligence Humanities Foundation (No. 2024SZWK023) .
- Language
- English
- Date published
- 03/2026
- Academic Unit
- Bus Admin College; Business Analytics
- Record Identifier
- 9985024249802771
Metrics
1 Record Views