Organizations: National Key Laboratory of Big Data and Decision, National University of Defense Technology, Changsha, China · The Center for machine learning research, Peking University, Beijing, China · Tsinghua University, Beijing, China · The Hong Kong University of Science and Technology, Clear Water Bay, Hong Kong, China
Domain-specific knowledge graphs (DKGs) are critical yet often suffer from limited coverage compared to General Knowledge Graphs (GKGs). Existing tasks to enrich DKGs rely primarily on extracting knowledge from external unstructured data or completing KGs through internal reasoning, but the scope and quality of such integration remain limited. This highlights a critical gap: little systematic exploration has been conducted on how comprehensive, high-quality GKGs can be effectively leveraged to supplement DKGs. To address this gap, we propose a new and practical task: domain-specific knowledge graph fusion (DKGF), which aims to mine and integrate relevant facts from general knowledge graphs into domain-specific knowledge graphs to enhance their completeness and utility. Unlike previous research, this new task faces two key challenges: (1) high ambiguity of domain relevance, i.e., difficulty in determining whether knowledge from a GKG is truly relevant to the target domain , and (2) cross-domain knowledge granularity misalignment, i.e., GKG facts are typically abstract and coarse-grained, whereas DKGs frequently require more contextualized, fine-grained representations aligned with particular domain scenarios. To address these, we present ExeFuse, a neuro-symbolic framework based on a novel Fact-as-Program paradigm. ExeFuse treats fusion as an executable process, utilizing neuro-symbolic execution to infer logical relevance beyond surface similarity and employing target space grounding to calibrate granularity. We construct new datasets to establish the first standardized evaluation suite for this task. Extensive experiments demonstrate that ExeFuse effectively overcomes domain barriers to achieve superior fusion performance.
Figures & tables
Figure 1. Examples of domain-specific knowledge graph fusion. Figure 1 (b) demonstrates the fusion of highly relevant entities from GKG into DKG. Figure 1 (c) shows the augmented DKG.
Figure 2. The framework of ExeFuse for DKGF.
Figure 3. Comprehensive evaluation across diverse domain-specific knowledge graphs on DKGF(W-Drug) , DKGF(W-UMLS) , DKGF(D-Alex) and DKGF(D-Crunch) .
Models
Split
Seen
Unseen
All
ACC
F1
ACC
F1
ACC
F1
DistMult-F
S1
0.563
0.623
0.491
0.340
0.554
0.585
S2
0.548
0.602
0.485
0.321
0.536
0.570
Avg.
0.556
0.613
0.488
0.331
0.545
0.578
Self-RAG-F
S1
0.565
0.570
0.594
0.372
0.576
0.544
S2
0.552
0.565
0.581
0.366
0.544
0.545
Table 4. Transferability and resource robustness on DKGF(W-I) . We report performance on Seen and Unseen entities across different resource splits. “ Avg. ” denotes the average performance across S1 and S2.
Variant
DKGF(W-I) -S1
Δ
ACC
F1
Avg.
ExeFuse (Full Model)
0.680
0.690
0.685
-
I. Complexity Validation (Is the complex pipeline necessary?)
w/o Rules L (Single Affine Projection)
0.605
0.612
0.609
-11.1%
w/o Reasoning (Direct Classification)
0.577
0.583
0.580
-15.3%
II. Module Necessity (Do all components contribute?)
Table 5. Ablation study and design verification on DKGF(W-I) -S1 . “ Avg. ”: average of ACC and F1. “ Δ ”: relative performance drop in Avg.
Settings
DKGF(W-I) -S1 (Accuracy)
Cost (Efficiency and Effectiveness)
ACC
F1
Avg.
Avg. tokens
Avg. time (seconds)
Self-Consistency-F
0.592
0.561
0.577
41,561,586
178,041.6
ChatEA-F
0.596
0.566
0.581
48,162,276
254,417.4
TransH-F
0.640
0.591
0.583
0
55.7
SimpleHHEA-F
0.490
0.499
0.495
0
2628.2
ExeFuse (Ours)
0.680
0.690
0.685
0
42.6
Table 6. Accuracy and cost analysis of ExeFuse and benchmark configurations on DKGF(W-I) -S1. Avg.tokens / Avg.time : the average tokens/time cost of the model.
Appendix figures & tables4 assets
Supplementary material from the paper’s appendix.
Appendix
Models
Split
Seen
Unseen
All
ACC
F1
ACC
F1
ACC
F1
DistMult-F
S1
0.531
0.564
0.472
0.315
0.525
0.536
S2
0.582
0.591
0.490
0.342
0.575
0.562
Avg.
0.557
0.578
0.481
0.329
0.550
0.549
Self-RAG-F
S1
0.602
0.585
0.582
0.334
0.594
0.540
S2
0.585
0.550
0.601
0.355
0.577
0.532
Appendix
Table 7. Transferability and resource robustness on DKGF(Y-I) . Metrics are averaged across splits to demonstrate fusion stability under varying resource constraints.
Figure 4. Transferability results (F1) across multiple domains (Biomedical, Academic, and Business).
Figure 5. The venn diagram of correct predictions on DKGF(W-I) -S1. Intersections show shared same predictions.
Models
DKGF(W-I) -S1
DKGF(Y-I) -S1
ACC
P
R
F1
Avg.
ACC
P
R
F1
Avg.
DistMult-F
0.975
0.830
0.840
0.835
0.905
0.946
0.853
0.730
0.787
0.866
Self-RAG-F
0.970
0.849
0.731
0.785
0.878
0.936
0.897
0.603
0.721
0.829
ChatEA-F
0.973
0.888
0.737
0.806
0.889
0.938
0.937
0.583
0.719
0.828
ExeFuse
0.982
0.968
0.791
0.871
0.927
0.958
0.958
0.721
0.823
0.890
Appendix
Table 8. Performance analysis of the new task “Relevant Entity Finding (REF)”.
Multi-domain knowledge graph completion (MKGC) aims to improve missing triple prediction in a target KG by transferring knowledge from other support KGs. Existing methods typically enforce consistency constraints on equivalent entities across KGs to transfer knowledge, which risks suppressing domain-specific contextual information of entities. This design can also compromise entity representation information from all KG domains, impeding performance improvements, especially in low-resource data scenarios. To address this, we pioneer a generation-based paradigm for MKGC and propose DMKGC, a conditional diffusion-guided knowledge transfer framework. Our key insight is to treat each KG as a partial view of the entity entire information, and generate informative domain-general entity embeddings through diffusion models conditioned on support KGs. Particularly, we first initialize domain-agnostic entity embeddings as prior entity embeddings, and then encode them within individual KGs. Afterward, we fuse equivalent entities from support KGs as the conditional diffusion generation guidance. We leverage the prior entity embeddings as the proxy generation objective, which ensures this conditional generation to be unbiased towards any conditioned KGs. Simultaneously, we also train the generated embeddings to be predictive across KGs, thus preserving domain-specific information. Extensive experiments on 14 KGs in 3 benchmarks demonstrate a 4.3% average MRR improvement in tail entity prediction over state-of-the-art methods, with sustained gains in low-resource data settings.
Jiawei Sheng, Taoyu Su, Xixun Lin +2
Institute of Information Engineering, Chinese Academy of Sciences Beijing, China · Institute of Information Engineering, Chinese Academy of Sciences School of Cyber Security, UCAS Beijing, China
Leveraging Large Language Models (LLMs) for Knowledge Graph Completion (KGC) is promising but hindered by a fundamental granularity mismatch. LLMs operate on fragmented token sequences, whereas entities are the fundamental units in knowledge graphs (KGs) scenarios. Existing approaches typically constrain predictions to limited candidate sets or align entities with the LLM's vocabulary by pooling multiple tokens or decomposing entities into fixed-length token sequences, which fail to capture both the semantic meaning of the text and the structural integrity of the graph. To address this, we propose KGT, a novel framework that uses dedicated entity tokens to enable efficient, full-space prediction. Specifically, we first introduce specialized tokenization to construct feature representations at the level of dedicated entity tokens. We then fuse pre-trained structural and textual features into these unified embeddings via a relation-guided gating mechanism, avoiding training from scratch. Finally, we implement decoupled prediction by leveraging independent heads to separate and combine semantic and structural reasoning. Experimental results show that KGT consistently outperforms state-of-the-art methods across multiple benchmarks.
Combining multiple knowledge graphs (KGs) across linguistic boundaries is a persistent challenge due to semantic heterogeneity and the complexity of graph environments. We propose a framework for cross-lingual graph fusion, leveraging the in-context reasoning and multilingual semantic priors of Large Language Models (LLMs). The framework implements structural linearization by mapping triplets directly into natural language sequences (e.g., [head] [relation] [tail]), enabling the LLM to map relations and reconcile entities between an evolving fused graph (Gc(t−1)) and a new candidate graph (Gt). Evaluated on the DBP15K dataset, this exploratory study demonstrates that LLMs can serve as a universal semantic bridge to resolve cross-lingual discrepancies. Results show the successful sequential agglomeration of multiple heterogeneous graphs, offering a scalable, modular solution for continuous knowledge synthesis in multi-source, multilingual environments.
Kaung Myat Kyaw, Khush Agarwal, Jonathan Chan
Innovative Cognitive Computing Research Center (IC2) School of Information Technology, KMUTT