Complete Technical Paper Content — Knowledge Graph Ai
1. Executive Abstract, Problem Formulation & Background
In knowledge representation and reasoning, a knowledge graph is a knowledge base that uses a graph-structured data model or topology to represent and operate on data. Knowledge graphs are often used to store interlinked descriptions of entities – objects, events, situations or abstract concepts – while also encoding the free-form semantics or relationships underlying these entities. Since the development of the Semantic Web, knowledge graphs have often been associated with linked open data projects, focusing on the connections between concepts and entities. They are also historically associated with and used by search engines such as Google, Bing, and Yahoo; knowledge engines and question-answering services such as WolframAlpha, Apple's Siri, and Amazon Alexa; and social networks such as LinkedIn and Facebook. Recent developments in data science and machine learning, particularly in graph neural networks, representation learning, and machine learning, have broadened the scope of knowledge graphs beyond their traditional use in search engines and recommender systems. They are increasingly used in scientific research, with notable applications in fields such as genomics, proteomics, and systems biology.
Flexible relations among knowledge in topical domains: A knowledge graph (i) defines abstract classes and relations of entities in a schema, (ii) mainly describes real world entities and their interrelations, organized in a graph, (iii) allows for potentially interrelating arbitrary entities with each other, and (iv) covers various topical domains. General structure: A network of entities, their semantic types, properties, and relationships. To represent properties, categorical or numerical values are often used. Supporting reasoning over inferred ontologies: A knowledge graph acquires and integrates information into an ontology and applies a reasoner to derive new knowledge. There are, however, many knowledge graph representations for which some of these features are not relevant. For those knowledge graphs, this simpler definition may be more useful:
A digital structure that represents knowledge as concepts and the relationships between them (facts). A knowledge graph can include an ontology that allows both humans and machines to understand and reason about its contents.
A knowledge graph formally represents semantics by describing entities and their relationships. Knowledge graphs may make use of ontologies as a schema layer. By doing this, they allow logical inference for retrieving implicit knowledge rather than only allowing queries requesting explicit knowledge. In order to allow the use of knowledge graphs in various machine learning tasks, several methods for deriving latent feature representations of entities and relations have been devised. These knowledge graph embeddings allow them to be connected to machine learning methods that require feature vectors like word embeddings. This can complement other estimates of conceptual similarity. Models for generating useful knowledge graph embeddings are commonly the domain of graph neural networks (GNNs). GNNs are deep learning architectures that comprise edges and nodes, which correspond well to the entities and relationships of knowledge graphs. The topology and data structures afforded by GNNs provide a convenient domain for semi-supervised learning, wherein the network is trained to predict the value of a node embedding (provided a group of adjacent nodes and their edges) or edge (provided a pair of nodes). These tasks serve as fundamental abstractions for more complex tasks such as knowledge graph reasoning and alignment.
As new knowledge graphs are produced across a variety of fields and contexts, the same entity will inevitably be represented in multiple graphs. However, because no single standard for the construction or representation of knowledge graph exists, resolving which entities from disparate graphs correspond to the same real world subject is a non-trivial task. This task is known as knowledge graph entity alignment, and is an active area of research. Strategies for entity alignment generally seek to identify similar substructures, semantic relationships, shared attributes, or combinations of all three between two distinct knowledge graphs. Entity alignment methods use these structural similarities between generally non-isomorphic graphs to predict which nodes correspond to the same entity. In 2023, researchers found success in using large language models (LLMs) in the task of entity alignment. This was in particular thanks to their effectiveness at producing syntactically meaningful embeddings. As the amount of data stored in knowledge graphs grows, developing dependable methods for knowledge graph entity alignment becomes an increasingly crucial step in the integration and cohesion of knowledge graph data.
Will Douglas Heaven (4 September 2020). "This know-it-all AI learns by reading the entire web nonstop". MIT Technology Review. Retrieved 5 September 2020. Diffbot is building the biggest-ever knowledge graph by applying image recognition and natural-language processing to billions of web pages.
In representation learning, knowledge graph embedding (KGE), also called knowledge representation learning (KRL), or multi-relation learning, is a machine learning task of learning a low-dimensional representation of a knowledge graph's entities and relations while preserving their meaning. Leveraging their embedded representation, knowledge graphs can be used for various applications such as link prediction, triple classification, entity recognition, clustering, and relation extraction.
Representation space: The low-dimensional space in which the entities and relations are represented. Scoring function: A measure of the goodness of a triple-embedded representation. Encoding models: The modality in which the embedded representation of the entities and relations interact with each other. Additional information: Any additional information coming from the knowledge graph that can enrich the embedded representation. Usually, an ad hoc scoring function is integrated into the general scoring function for each additional piece of information.
2. Related Work, Literature Survey & Algorithmic Foundations
algorithm Compute entity and relation embeddings input: The training set S = { ( h , r , t ) } {\displaystyle S=\{(h,r,t)\}} , entity set E {\displaystyle E} , relation set R {\displaystyle R} , embedding dimension k {\displaystyle k}
initialization: the entities e {\displaystyle e} and relations r {\displaystyle r} embeddings (vectors) are randomly initialized
while stop condition do S b a t c h ← s a m p l e ( S , b ) {\displaystyle S_{batch}\leftarrow sample(S,b)} // Sample a batch from the training set for each ( h , r , t ) {\displaystyle (h,r,t)} in S b a t c h {\displaystyle S_{batch}} do ( h ′ , r , t ′ ) ← s a m p l e ( S ′ ) {\displaystyle (h',r,t')\leftarrow sample(S')} // Sample a corrupted fact T b a t c h ← T b a t c h ∪ { ( ( h , r , t ) , ( h ′ , r , t ′ ) ) } {\displaystyle T_{batch}\leftarrow T_{batch}\cup \{((h,r,t),(h',r,t'))\}} end for Update embeddings by minimizing the loss function end while
M R = 1 | Q | ∑ q ∈ Q q {\displaystyle MR={\frac {1}{|Q|}}\sum _{q\in Q}{q}}
M R R = 1 | Q | ∑ q ∈ Q 1 q ∈ [ 0 , 1 ] {\displaystyle MRR={\frac {1}{|Q|}}\sum _{q\in Q}{\frac {1}{q}}\in [0,1]}
Given a collection of triples (or facts) F = { ⟨ head , relation , tail ⟩ } {\displaystyle {\mathcal {F}}=\{\langle {\text{head}},{\text{relation}},{\text{tail}}\rangle \}} , the knowledge graph embedding model produces, for each entity and relation present in the knowledge graph a continuous vector representation. ( h , r , t ) {\displaystyle (h,r,t)} is the corresponding embedding of a triple with h , t ∈ I R d {\displaystyle h,t\in {\rm {I\!R}}^{d}} and r ∈ I R k {\displaystyle r\in {\rm {I\!R}}^{k}} , where d {\displaystyle d} is the embedding dimension for the entities, and k {\displaystyle k} for the relations. The score function of a given model is denoted by f r ( h , t ) {\displaystyle {\mathcal {f}}_{r}(h,t)} and measures the distance of the embedding of the head from the embedding of tail given the embedding of the relation. In other words, it quantifies the plausibility of the embedded representation of a given fact. Rossi et al. propose a taxonomy of the embedding models and identifies three main families of models: tensor decomposition models, geometric models, and deep learning models.
DistMult: Since the embedding matrix of the relation is a diagonal matrix, the scoring function can not distinguish asymmetric facts. ComplEx: As DistMult uses a diagonal matrix to represent the relations embedding but adds a representation in the complex vector space and the hermitian product, it can distinguish symmetric and asymmetric facts. This approach is scalable to a large knowledge graph in terms of time and space cost. ANALOGY: This model encodes in the embedding the analogical structure of the knowledge graph to simulate inductive reasoning. Using a differentiable objective function, ANALOGY has good theoretical generality and computational scalability. It is proven that the embedding produced by ANALOGY fully recovers the embedding of DistMult, ComplEx, and HolE. SimplE: This model is the improvement of canonical polyadic decomposition (CP), in which an embedding vector for the relation and two independent embedding vectors for each entity are learned, depending on whether it is a head or a tail in the knowledge graph fact. SimplE resolves the problem of independent learning of the two entity embeddings using an inverse relation and average the CP score of ( h , r , t ) {\displaystyle (h,r,t)} and ( t , r − 1 , h ) {\displaystyle (t,r^{-1},h)} . In this way, SimplE collects the relation between entities while they appear in the role of subject or object inside a fact, and it is able to embed asymmetric relations.
f r ( h , t ) = δ ( τ ( h , r ) , t ) {\displaystyle {\mathcal {f}}_{r}(h,t)=\delta (\tau (h,r),t)}
3. Dataset Ingestion, Preprocessing & Stochastic Augmentations
Geometric models are similar to the tensor decomposition model, but the main difference between the two is that they have to preserve the applicability of the transformation τ {\displaystyle \tau } in the geometric space in which it is defined.
STransE: This model is the result of the combination of TransE and of the structure embedding in such a way it is able to better represent the one-to-many, many-to-one, and many-to-many relations. To do so, the model involves two additional independent matrix W r h {\displaystyle W_{r}^{h}} and W r t {\displaystyle W_{r}^{t}} for each embedded relation r {\displaystyle r} in the KG. Each additional matrix is used based on the fact the specific relation interact with the head or the tail of the fact. In other words, given a fact ( h , r , t ) {\displaystyle (h,r,t)} , before applying the vector translation, the head h {\displaystyle h} is multiplied by W r h {\displaystyle W_{r}^{h}} and the tail is multiplied by W r t {\displaystyle W_{r}^{t}} . CrossE: Crossover interactions can be used for related information selection, and could be very useful for the embedding procedure. Crossover interactions provide two distinct contributions in the information selection: interactions from relations to entities and interactions from entities to relations. This means that a relation, e.g.'president_of' automatically selects the types of entities that are connecting the subject to the object of a fact. In a similar way, the entity of a fact inderectly determine which is inference path that has to be chosen to predict the object of a related triple. CrossE, to do so, learns an additional interaction matrix C {\displaystyle C} , uses the element-wise product to compute the interaction between h {\displaystyle h} and r {\displaystyle r} . Even if, CrossE, does not rely on a neural network architecture, it is shown that this methodology can be encoded in such architecture.
TorusE: The regularization term of TransE makes the entity embedding to build a spheric space, and consequently loses the translation properties of the geometric space. To address this problem, TorusE leverages the use of a compact Lie group that in this specific case is n-dimensional torus space, and avoid the use of regularization. TorusE defines the distance functions to substitute the L1 and L2 norm of TransE. RotatE: RotatE is inspired by the Euler's identity and involves the use of Hadamard product to represent a relation r {\displaystyle r} as a rotation from the head h {\displaystyle h} to the tail t {\displaystyle t} in the complex space. For each element of the triple, the complex part of the embedding describes a counterclockwise rotation respect to an axis, that can be describe with the Euler's identity, whereas the modulus of the relation vector is 1. It is shown that the model is capable of embedding symmetric, asymmetric, inversion, and composition relations from the knowledge graph.
ConvE: ConvE is an embedding model that represents a good tradeoff expressiveness of deep learning models and computational expensiveness, in fact it is shown that it used 8x less parameters, when compared to DistMult. ConvE uses a one-dimensional d {\displaystyle d} -sized embedding to represent the entities and relations of a knowledge graph. To compute the score function of a triple, ConvE apply a simple procedure: first concatenes and merge the embeddings of the head of the triple and the relation in a single data [ h ; r ] {\displaystyle {\ce {[h;{\mathcal {r}}]}}} , then this matrix is used as input for the 2D convolutional layer. The result is then passed through a dense layer that apply a linear transformation parameterized by the matrix W {\displaystyle {\mathcal {W}}} and at the end, with the inner product is linked to the tail triple. ConvE is also particularly efficient in the evaluation procedure: using a 1-N scoring, the model matches, given a head and a relation, all the tails at the same time, saving a lot of evaluation time when compared to the 1-1 evaluation program of the other models. ConvR: ConvR is an adaptive convolutional network aimed to deeply represent all the possible interactions between the entities and the relations. For this task, ConvR, computes convolutional filter for each relation, and, when required, applies these filters to the entity of interest to extract convoluted features. The procedure to compute the score of triple is the same as ConvE. ConvKB: ConvKB, to compute score function of a given triple ( h , r , t ) {\displaystyle (h,r,t)} , it produces an input [ h ; r ; t ] {\displaystyle {\ce {[h;{\mathcal {r}};t]}}} of dimension d × 3 {\displaystyle d\times 3} without reshaping and passes it to series of convolutional filter of size 1 × 3 {\displaystyle 1\times 3} . This result feeds a dense layer with only one neuron that produces the final score. The single final neuron makes this architecture as a binary classifier in which the fact could be true or false. A difference with ConvE is that the dimensionality of the entities is not changed.
CapsE: CapsE implements a capsule network to model a fact ( h , r , t ) {\displaystyle (h,r,t)} . As in ConvKB, each triple element is concatenated to build a matrix [ h ; r ; t ] {\displaystyle {\ce {[h;{\mathcal {r}};t]}}} and is used to feed to a convolutional layer to extract the convolutional features. These features are then redirected to a capsule to produce a continuous vector, more the vector is long, more the fact is true.
RSN: During the embedding procedure is commonly assumed that, similar entities has similar relations. In practice, this type of information is not leveraged, because the embedding is computed just on the undergoing fact rather than a history of facts. Recurrent skipping networks (RSN) uses a recurrent neural network to learn relational path using a random walk sampling.
Research Topics in Knowledge Graph Papers and Materials from All Areas Note 1: The papers from the database/data science communities are marked with 🌟. Note 2: From Nov 2023, I started to add papers related to LLM in each session, marked with 🔥. 0. Overview Survey of Knowledge Graphs - General Concepts, Scientific Problems, Applications, Challenges, and Directions) [ link ] 1. Knowledge Extraction and Integration to Construct KGs Knowledge Graph Construction (Demo or System) [link] About Domain-Specific Knowledge Bases [link] About Multi-Modal Knowledge Graph (MMKG) [ link ] Named Entity Recognition, Entity Extraction and Entity Typing [link] Coreference Resolution [link] Entity Linking and Entity Disambiguation [link] Entity Resolution, Entity Matching and Entity Alignment [link] Relation Extraction [link] General Information Extraction, Open Information Extraction, Open KG Construction, and Open KG Canonicalization [link] Relation Linking and Relation Disambiguation [link] 2. Mining and Refinement of KGs Knowledge Graph Embedding, Learning, Reasoning, Rule Mining, and Path Finding [link] Knowledge Base Refinement (Incompleteness, Incorrectness, and Freshness) [link] Knowledge Fusion, Cleaning, Evaluation and Truth Discovery [link] 3. Applications Supported by KGs Knowledge Graph Question Answering (KGQA) [link] Knowledge Graph Recommendation [link] Knowledge Graph Enhanced Machine Learning [link] Knowledge Graphs and Large Language Models (LLMs) [link] 🔥🔥🔥 Knowledge Graphs in RAG [ Link ] Knowledge Graphs in Agent Memory [ Link ] 4. Schema and Query of KGs Knowledge Graph Representation (RDF and Property Graph), Schema and Query [link] Knowledge Graph Taxonomy/Ontology Construction and Improvement [ link ] 5. Others Graph Databases [ Link ] Other Interesting Research Topics Related to KGs (e.g., KG Security and Evaluation) [link] Good DB Papers [link] Tutorials and Notes of Other Related Insightful Topics [ Link ] Papers and Materials from the Database/Data Science Communities Note: Papers from SIGMOD/VLDB/ICDE/KDD/TKDE/VLDBJ [2018] [2019] [2020] [2021] [2022] [2023] [ 2024 ] [ 2025 ] [ 2026 ] Tutorials and Notes from Talented People Note: We only list the general tutorials here. For the surveys or communities of specific topics such as KG Construction, KG4LLM or KG Embedding, please refer to the above detailed sections. Tutorials, Discussion, Courses and Communities An introduction to knowledge graph and knowledge extraction from unstructured text. [ Link ] Information Extraction by Niranjan Balasubramanian {Slides in my Mac} CS 520 - Knowledge Graphs (seminar) - provided by Stanford OpenKG.cn GitHub Repos that Summarize the Papers/Projects/Data related to Knowledge Graphs A Collection of KG Surveys, Papers (WWW+ACL+AAAI) and Data [ GitHub ] KG SOTA [ GitHub ] Awesome KG tutorials/papers/projects/communities [ GitHub ] KG SOTA (Chinese) [ Zhihu ] Tracking Progress in Natural Language Processing [ GitHub ] KG Embedding SOTA [ GitHub ] Entity Related Papers [ GitHub ] Information Extraction Resources [ GitHub ] KGQA [ Giters ] Open-Environment Knowledge Graph Construction and Reasoning: Challenges, Approaches, and Opportunities [ GitHub ] KG-LLM-Papers [ Link ] Awesome LLM-KGs [ Link ] Research Groups You may refer to this collection for more details: [ Github ] Knowledge Graph Datasets Note: We only list the general KGs here. For the datasets or benchmarks for specific topics such as KGQA or KG Completion, please refer to the above detailed sections. WikiData [ Link ] DBPedia [ Link ] Yago [ Link ] Google Knowledge Graph (Freebase) [ Link ] Chinese OwnThink Knowledge Graph [ Link ] Useful Tools or APIs From Freebase to Wikidata: The Great Migration [ Paper and useful links ] SPARQL tutorial [ Link ] Installing and running ElasticSearch [ Link ] Open KG on COVID-19 [ Link ] BOOKNLP [ Link ] (Pronominal Coreference Resolution, a natural language processing pipeline that scales to books and other long documents (in English)) Wikidata Integrator [ GitHub ] OpenTapioca [ Link ] Grakn KGLIB (Knowledge Graph Library) [ GitHub ] SPARQL server on Freebase [ GitHub ] [ About VOS ] LATEX Code Search [ Link ] I have listed more tools/benchmarks/APIs/demos in each sub-page of different topics in this repo. Feel free to check!
General Information Extraction, Open Information Extraction, Open KG Construction, and Open KG Canonicalization [link]
4. Neural Network Backbone, Layer Specifications & Loss Formulations
Other Interesting Research Topics Related to KGs (e.g., KG Security and Evaluation) [link]
Note: We only list the general tutorials here. For the surveys or communities of specific topics such as KG Construction, KG4LLM or KG Embedding, please refer to the above detailed sections.
Note: We only list the general KGs here. For the datasets or benchmarks for specific topics such as KGQA or KG Completion, please refer to the above detailed sections.
I have listed more tools/benchmarks/APIs/demos in each sub-page of different topics in this repo. Feel free to check!
Knowledge Graph Learning Resources A curated collection of knowledge graph learning resources, including papers, tutorials, datasets, implementations, tools, and communities. Resources are available in both English and Chinese. The language of each resource is indicated when it may not be clear from the title. Link status: All links were checked on August 26, 2026. Contributions that report broken links or improve the organization of the list are welcome. Date convention: The year shown for each resource is its latest meaningful update: recent repository or collection activity for maintained projects, and the publication or revision year for static papers, datasets, tutorials, and slides. Entries in each section are ordered from newest to oldest. Contents Papers and Notes Tutorials, Surveys, and Slides Courses and Lectures Datasets Projects and Implementations Libraries and Tools Communities Contributing Papers and Notes Paper notes are organized with GitHub issue labels. Most existing notes are written in Chinese, but contributions in any language are welcome. Knowledge graph tasks 2023 Information Extraction and Open IE 2023 Named Entity Recognition 2023 Relation Extraction 2020 Named Entity Linking 2020 Knowledge Representation Learning and Knowledge Graph Embeddings 2020 Knowledge Graph Population and Construction — constructing knowledge graphs from different sources 2019 Knowledge-based Recommendation Systems 2019 Knowledge Base Completion and Knowledge Graph Reasoning — entity prediction and link prediction 2019 Ontology-based Information Extraction 2019 Semantic Role Labeling Related NLP and machine learning tasks Tasks 2024 Recommendation 2023 Question Answering and Machine Comprehension 2023 Relational Reasoning 2021 Annotation 2021 Domain Adaptation and Domain-specific Learning 2020 Coreference Resolution 2019 Data Augmentation 2019 Dependency Parsing 2019 Natural Language Understanding 2019 Neural Machine Translation 2019 Slot Filling 2019 Summarization 2019 Text Classification Models and methods 2023 BERT 2023 Embeddings and Pre-trained Models 2023 Multi-task and Joint Learning 2019 End-to-end Models 2019 Graph Neural Networks 2019 Transformer-based Models Tutorials, Surveys, and Slides General tutorials and articles 2026 Awesome GraphRAG — actively maintained collection of GraphRAG surveys, papers, benchmarks, and open-source projects 2025 Ontotext Blog 2021 Recent Trends in Entity Linking 2018 Knowledge Extraction and Inference from Text — KDD tutorial: Part 1 , Part 2 2018 Knowledge Graphs: The Power of Graph-based Search — Neo4j slides 2018 Mining Knowledge Graphs from Text — WSDM tutorial 2018 知识图谱论文合集 (Chinese) 2018 知识图谱入门（三） (Chinese) 2017 Enterprise Knowledge Graphs for Large Scale Analytics — IBM tutorial 2017 Getting Started with Knowledge Graphs — metaphacts slides 2016 Knowledge Integration in Practice — Yahoo slides Surveys 2020 A Survey on Knowledge Graphs: Representation, Acquisition and Applications — Shaoxiong Ji et al. 2019 Deep Learning in Knowledge Graph — notes 2018 Summary of Translation Models for Knowledge Graph Embeddings 2017 知识图谱研究进展 — 漆桂林等 (Chinese) 2016 知识图谱构建技术综述 — 刘峤等 (Chinese) 2016 知识图谱技术综述 — 徐增林等 (Chinese) 2014 垂直知识图谱构造工具与行业应用 — 阮彤 (Chinese) 2014 知识图谱：大数据语义链接的基石 — 李涓子 (Chinese) Relation extraction 2007 A Survey on Relation Extraction — CMU slides Event extraction 2018 事件抽取与金融事件图谱构建 (Chinese) Courses and Lectures 2018 从零开始构建知识图谱 — Zhihu column (Chinese) Datasets Relation extraction and semantic relationships 2021 Annotated Semantic Relationships Datasets (English) 2019 SemEval-2010 Task 8 — alternative implementation 2017 TACRED Open and domain-specific knowledge graphs 2026 KG-Hub and KG-OBO — biomedical knowledge graph resources 2026 OpenKG.CN Datasets (Chinese) 2024 PheKnowLator — heterogeneous biomedical knowledge graphs and benchmarks GraphRAG benchmarks 2026 GraphRAG-Bench — benchmark, datasets, and evaluation code for graph-based retrieval-augmented generation Projects and Implementations GraphRAG and LLM–KG systems 2026 HippoRAG — knowledge graph and Personalized PageRank-based long-term memory framework for LLMs 2026 KAG — logical form-guided reasoning and retrieval framework for domain knowledge bases 2026 Knowledge Graph Builder — application for transforming documents and web content into a knowledge graph with LLMs 2026 LightRAG — graph-based retrieval-augmented generation framework with multiple storage backends and local deployment support Knowledge graph applications 2020 KG-demo-for-movie — movie knowledge graph and KBQA; article (Chinese) 2020 Z_knowledge_graph — movie knowledge graph tutorial (Chinese) 2019 KGQA_HLM — Dream of the Red Chamber character graph and question answering system (Chinese) Libraries and Tools Knowledge graph construction and representation 2026 BioCypher — framework for creating and maintaining biomedical knowledge graphs 2026 PyKEEN — Python library for learning and evaluating knowledge graph embeddings 2024 GRAPE — Rust/Python library for graph representation learning and evaluation 2024 OpenKE — knowledge graph embedding framework 2022 cnSchema — open Chinese knowledge graph schema 2020 KnowledgeGraph — Chinese knowledge graph API 2017 DeepDive — system for extracting structured data from unstructured sources; homepage , papers Semantic Web, RDF, and SPARQL 2026 Apache Jena — Java framework for Semantic Web and Linked Data applications 2026 Oxigraph — RDF and SPARQL toolkit and graph database with Rust, Python, and JavaScript interfaces 2026 RDFLib — Python library for working with RDF and SPARQL Information extraction and annotation 2026 DeepKE — toolkit for entity, relation, attribute, and event extraction in knowledge graph construction 2026 doccano — text annotation tool 2025 HanLP — multilingual natural language processing library 2024 OpenNRE — neural relation extraction toolkit; related papers 2016 fact_triple_extraction — dependency parsing-based fact triple extraction (Chinese) Visualization and APIs 2025 InteractiveGraph — graph visualization tool Additional collections 2025 Useful Tools and Lectures Related to Data Science (Chinese) Communities 2026 OpenKG.CN — Chinese Open Knowledge Graph community 2016 Beijing Knowledge Graph Study Group (Chinese) Contributing Contributions are welcome. Before opening a pull request, please make sure that: The resource is directly relevant to knowledge graph learning or engineering. The resource is publicly accessible and provides substantial educational or technical value. The link is stable and not duplicated elsewhere in the list. The entry includes a short, factual description and a language label when appropriate. The contribution does not primarily promote a paid product, commercial service, organization, or the contributor's own content. A pull request contains focused changes that are easy to review. Commercial promotion, affiliate links, link farms, and low-content self-promotion will not be accepted. Maintainers may decline submissions at their discretion to keep the collection neutral and useful. If you find a broken or outdated link, please open an issue or submit a pull request with a replacement.
A curated collection of knowledge graph learning resources, including papers, tutorials, datasets, implementations, tools, and communities.
Resources are available in both English and Chinese. The language of each resource is indicated when it may not be clear from the title.
Link status: All links were checked on August 26, 2026. Contributions that report broken links or improve the organization of the list are welcome.
5. Training Protocol, Convergence Behavior & Experimental Evaluation
Date convention: The year shown for each resource is its latest meaningful update: recent repository or collection activity for maintained projects, and the publication or revision year for static papers, datasets, tutorials, and slides. Entries in each section are ordered from newest to oldest.
Paper notes are organized with GitHub issue labels. Most existing notes are written in Chinese, but contributions in any language are welcome.
Commercial promotion, affiliate links, link farms, and low-content self-promotion will not be accepted. Maintainers may decline submissions at their discretion to keep the collection neutral and useful.
If you find a broken or outdated link, please open an issue or submit a pull request with a replacement.
Access to this page requires authorization. You can try signing in or changing directories .
The hardest part of working with graphs is building the graph in the first place . Manually curating entities and relationships from thousands of documents is prohibitively expensive. AI Functions in Azure HorizonDB solve this issue by bringing LLM-powered intelligence directly into SQL, so you can extract, structure, and query knowledge graphs without leaving the database.
azure_ai.extract() discovers hidden relationships and entities from unstructured text, right inside a SQL query. Feed it contracts, support tickets, research papers, or any text-heavy data, and it pulls out the structured relationships you need to populate your knowledge graph.
Before running this tutorial, you need an Azure HorizonDB instance, an AI model configured via the model registry, and the required PostgreSQL extensions.
Model Figures & Evaluation Curves (4 Images)
Deep Learning Execution Pipeline
(Preprocessing)
(Feature Extraction)
(Self-Attention)
(Loss Optimization)