---
title: "Knowledge graphs & ontologies — papers"
description: "Knowledge graphs & ontologies: 78 papers of the Elamur Research library, the most cited first."
canonical: "https://research.elamur.ai/free/en/topics/knowledge-graphs/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/topics/knowledge-graphs/)

[home](https://md.research.elamur.ai/free/en/) / [topics](https://md.research.elamur.ai/free/en/topics/) / Knowledge graphs & ontologies

# 🕸️ Knowledge graphs & ontologies

78 papers · page 1 of 2 · 1–25 · the most cited first · [all topics](https://md.research.elamur.ai/free/en/topics/)

Facts as linked graphs: knowledge bases, ontologies and the semantic web.

💡 Explained: [Knowledge representation and reasoning](https://md.research.elamur.ai/free/en/concepts/knowledge-representation/) · [Ontology engineering](https://md.research.elamur.ai/free/en/concepts/ontology-engineering/)

A survey · about 17 minutes to read · every paper named here has its own page in this library

**In this article**

1. [What it is](#a-what-it-is)
2. [History and milestones](#a-history-and-milestones)
3. [Main problems and tasks](#a-main-problems-and-tasks)
4. [Methods and schools](#a-methods-and-schools)
5. [Datasets, resources and benchmarks](#a-datasets-resources-and-benchmarks)
6. [Landmark papers in this library](#a-landmark-papers-in-this-library)
7. [People](#a-people)
8. [Institutions and places](#a-institutions-and-places)
9. [Open questions](#a-open-questions)
10. [Neighbouring topics](#a-neighbouring-topics)
11. [Key works in this library](#a-key-works-in-this-library)

A knowledge graph stores facts as a network: things in the world become nodes, and the relations between them become labelled edges, so that "Kyiv — capital of — Ukraine" is a single edge between two nodes. An ontology is the schema that gives such a network its meaning: it names the classes of things (city, country, person), arranges them in hierarchies, and states what may be said about each. Together, knowledge graphs, knowledge bases and ontologies form the symbolic memory of language technology, the place where facts are kept explicitly, can be inspected and corrected, and can be queried with precision.

The topic sits where computational linguistics meets databases, logic and the Semantic Web. The papers here range from early linguistic ontologies built to support translation and generation, through the great Wikipedia-derived knowledge bases and the decade of knowledge graph embeddings, to the current work that asks how large language models and explicit graphs can help each other. A distinct strand comes from linguistic linked data, where dictionaries, morphologies and treebanks are themselves published as graphs.

## What it is

The core data model is simple. A fact is a triple of subject, predicate and object, and a set of triples is a directed graph with labelled edges. The W3C standard for this model is RDF, queried with SPARQL; ontologies on top of it are written in RDFS or OWL, whose description-logic semantics lets a reasoner derive facts that were never stated. Property graphs, popular in industry, add key–value attributes to nodes and edges. The survey [Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hogan-2020-knowledge-graphs/) treats both models side by side and adopts a deliberately broad definition: any graph of data meant to accumulate and convey knowledge about the real world.

Several neighbouring notions overlap with this one:

- A **knowledge base** is the older and wider term, which may also hold rules.
- An **ontology** in the computer-science sense is a formal, shared specification of the concepts of a domain. Lexical resources such as WordNet are close relatives: they organise word senses by hypernymy, but by linguistic rather than ontological criteria, a difference several papers here take seriously.
- A **commonsense knowledge graph** such as ConceptNet or ATOMIC stores loosely structured, often free-text knowledge about everyday situations rather than encyclopedic facts about named entities.
- **Linguistic linked data** applies the same RDF machinery to language resources themselves: lexicons, morpheme inventories, annotated corpora.

What separates the topic from its neighbours is the emphasis on explicit structure. [Information extraction](https://md.research.elamur.ai/free/en/topics/information-extraction/) produces facts from text; this topic is about where those facts go and what can be done with them once stored. [Semantic parsing](https://md.research.elamur.ai/free/en/topics/semantic-parsing/) translates questions into queries; the knowledge graph is what those queries run against.

## History and milestones

Graph-shaped knowledge is as old as artificial intelligence: semantic networks, frames, description logics and the Cyc project (from 1984) all tried to write down what a machine should know. WordNet, begun at Princeton in the mid-1980s, became the most widely used lexical network in natural language processing. In machine translation, interlingua systems needed an ontology of meanings shared across languages, and [Approximating an Interlingua in a Principled Way](https://md.research.elamur.ai/free/en/papers/hovy-1992-approximating-interlingua-principled-way/) (1992) proposed building one language at a time, adding a concept only when analysis or generation actually needs the distinction. Two years later, [The Merged Upper Model: A Linguistic Ontology for German and English](https://md.research.elamur.ai/free/en/papers/henschel-1994-merged-upper-model-linguistic-ontology/) applied that recipe to fuse the English and German upper models used for sentence generation.

Landmark papers on this topic: the 12 most cited, in time The 12 most cited papers on this topic span 2013–2023; the dot grows with the citations.

**The numbers**

| Paper | Citations |
| --- | --- |
| [2013 — Semantic Parsing on Freebase from Question-Answer Pairs](https://md.research.elamur.ai/free/en/papers/berant-2013-semantic-parsing-freebase-question-answer/) | 785 |
| [2014 — Semantic Parsing via Paraphrasing](https://md.research.elamur.ai/free/en/papers/berant-2014-semantic-parsing-paraphrasing/) | 681 |
| [2015 — Knowledge Graph Embedding via Dynamic Mapping Matrix](https://md.research.elamur.ai/free/en/papers/ji-2015-knowledge-graph-embedding-dynamic-mapping/) | 1,936 |
| [2015 — Representing Text for Joint Embedding of Text and Knowledge Bases](https://md.research.elamur.ai/free/en/papers/toutanova-2015-representing-text-joint-embedding-text/) | 904 |
| [2015 — Modeling Relation Paths for Representation Learning of Knowledge Bases](https://md.research.elamur.ai/free/en/papers/lin-2015-modeling-relation-paths-representation-learning/) | 695 |
| [2018 — A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network](https://md.research.elamur.ai/free/en/papers/nguyen-2018-novel-embedding-model-knowledge-base/) | 944 |
| [2019 — Language Models as Knowledge Bases?](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/) | 2,977 |
| [2019 — COMET: Commonsense Transformers for Automatic Knowledge Graph Construction](https://md.research.elamur.ai/free/en/papers/bosselut-2019-comet-commonsense-transformers-automatic-knowledge/) | 1,282 |
| [2019 — K-BERT: Enabling Language Representation with Knowledge Graph](https://md.research.elamur.ai/free/en/papers/liu-2019-k-bert-enabling-language-representation/) | 1,006 |
| [2019 — TuckER: Tensor Factorization for Knowledge Graph Completion](https://md.research.elamur.ai/free/en/papers/balazevic-2019-tucker-tensor-factorization-knowledge-graph/) | 792 |
| [2020 — Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hogan-2020-knowledge-graphs/) | 2,011 |
| [2023 — Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://md.research.elamur.ai/free/en/papers/pan-2023-unifying-large-language-models-knowledge/) | 1,545 |

The second era began with the Semantic Web vision around 2001 and the Linked Data principles of 2006. Wikipedia turned out to be the richest source of structured facts: DBpedia and YAGO both appeared in 2007, Freebase the same year, and Wikidata in 2012. That year Google announced its Knowledge Graph, which gave the field its current name. The papers here describe this generation of resources in detail: [DBpedia: A Multilingual Cross-domain Knowledge Base](https://md.research.elamur.ai/free/en/papers/mendes-2012-dbpedia-multilingual-cross-domain-knowledge/), its later journal account [DBpedia – A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia](https://md.research.elamur.ai/free/en/papers/lehmann-2015-dbpedia-large-scale-multilingual-knowledge/), and [YAGO3: A Knowledge Base from Multilingual Wikipedias](https://md.research.elamur.ai/free/en/papers/mahdisoltani-2015-yago3-knowledge-base-multilingual-wikipedias/).

Papers on this topic by year 78 dated papers from 1992 to 2026; the busiest year is 2026 with 12, and 63% of them appeared in 2018 or later.

**The numbers**

| Year | Papers |
| --- | --- |
| 2026 | 12 |
| 2025 | 4 |
| 2024 | 1 |
| 2023 | 6 |
| 2022 | 3 |
| 2021 | 3 |
| 2020 | 7 |
| 2019 | 8 |
| 2018 | 5 |
| 2017 | 2 |
| 2016 | 3 |
| 2015 | 9 |
| 2014 | 5 |
| 2013 | 4 |
| 2012 | 2 |
| 2010 | 2 |
| 1994 | 1 |
| 1992 | 1 |

The third era, roughly 2013 to 2019, was dominated by representation learning. With Freebase and WordNet available as benchmarks, researchers learned vectors for entities and relations and used them to predict missing links. The neural tensor network, DistMult, TransD, PTransE, ConvKB and TuckER all belong here, as does the parallel line of question answering over Freebase. This period produced the most cited work in this library.

The fourth era started when pretrained language models arrived. [Language Models as Knowledge Bases?](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/) asked in 2019 whether a model like BERT already contains a knowledge base, and since then much of the field has been about the relation between implicit knowledge in model parameters and explicit knowledge in graphs. The newest papers use language models to build, align and query ontologies.

## Main problems and tasks

**Construction.** Graphs are built by hand (WordNet, Cyc), from semi-structured sources (DBpedia and YAGO from Wikipedia), from free text, and most recently by generation with language models (COMET). Each route trades precision against coverage.

**Completion, or link prediction.** Every real graph is incomplete. Given a query such as (Marie Curie, nationality, ?), a model must rank candidate answers. This is the task behind most of the embedding papers, evaluated on benchmarks derived from Freebase and WordNet.

**Entity linking.** Before text can be connected to a graph, mentions must be mapped to nodes, or recognised as absent. [Entity Disambiguation for Knowledge Base Population](https://md.research.elamur.ai/free/en/papers/dredze-2010-entity-disambiguation-knowledge-base-population/) is the classic treatment here.

**Relation extraction.** Finding new triples in text, often with distant supervision from an existing knowledge base.

**Question answering over knowledge bases (KBQA).** Mapping a natural-language question to a query whose execution yields the answer, or scoring candidate answers directly in an embedding space. See also [Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/).

**Text generation from data.** Verbalising a set of triples as fluent text, the inverse of extraction, studied in the WebNLG challenge and related work; see [Text generation](https://md.research.elamur.ai/free/en/topics/text-generation/).

**Knowledge injection.** Supplying a language model or classifier with facts from a graph.

## Methods and schools

### Symbolic and ontological engineering

The oldest school treats an ontology as a carefully engineered artefact. Its concerns are top-level categories, the correctness of is-a links, and alignment with foundational ontologies such as DOLCE or BFO. [The Arabic Ontology -- An Arabic Wordnet with Ontologically Clean Content](https://md.research.elamur.ai/free/en/papers/jarrar-2022-arabic-ontology-arabic-wordnet-ontologically/) is a clear modern statement of this position: it argues that hierarchies resting on native-speaker intuition contain errors that ontological analysis and expert classification would catch. [Al –-Khalil : The Arabic Linguistic Ontology Project](https://md.research.elamur.ai/free/en/papers/aliane-2010-al-khalil-arabic-linguistic-ontology/) shows another typical move, hanging the categories of a national grammatical tradition under the shared GOLD linguistic ontology so they stay interoperable.

### Extraction from semi-structured sources

DBpedia and YAGO represent the school that harvests what Wikipedia editors have already structured. Their main difficulty is heterogeneity: infobox templates differ between languages and articles. DBpedia's answer was a community-maintained mapping to one shared ontology; YAGO3 instead mapped foreign infobox attributes by checking them statistically against facts already known from the English edition, and anchored every entity in the WordNet taxonomy.

### Latent feature models: knowledge graph embeddings

From about 2013 the dominant approach to completion was to embed entities and relations in a vector space and define a scoring function over triples. Three families stand out:

- **Translational models** treat a relation as a vector offset from head to tail (TransE, 2013). [Knowledge Graph Embedding via Dynamic Mapping Matrix](https://md.research.elamur.ai/free/en/papers/ji-2015-knowledge-graph-embedding-dynamic-mapping/) (TransD) refined this with entity-specific projections, and [Modeling Relation Paths for Representation Learning of Knowledge Bases](https://md.research.elamur.ai/free/en/papers/lin-2015-modeling-relation-paths-representation-learning/) (PTransE) extended it from single edges to multi-step paths.
- **Bilinear and tensor-factorisation models** score a triple through a multiplicative interaction. [Reasoning With Neural Tensor Networks for Knowledge Base Completion](https://md.research.elamur.ai/free/en/papers/socher-2013-reasoning-neural-tensor-networks-knowledge/), [Embedding Entities and Relations for Learning and Inference in Knowledge Bases](https://md.research.elamur.ai/free/en/papers/yang-2014-embedding-entities-relations-learning-inference/) (DistMult) and [TuckER: Tensor Factorization for Knowledge Graph Completion](https://md.research.elamur.ai/free/en/papers/balazevic-2019-tucker-tensor-factorization-knowledge-graph/) belong here, with TuckER showing that several earlier bilinear models are special cases of a single Tucker decomposition.
- **Neural models** use convolution or graph networks over embeddings, as in [A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network](https://md.research.elamur.ai/free/en/papers/nguyen-2018-novel-embedding-model-knowledge-base/) (ConvKB).

A recurring lesson of this period was methodological. [Observed versus latent features for knowledge base and text inference](https://md.research.elamur.ai/free/en/papers/toutanova-2015-observed-versus-latent-features-knowledge/) showed that a trivial model exploiting inverse and duplicate relations could beat sophisticated embeddings on the standard FB15k and WN18 benchmarks, which led to cleaner test sets such as FB15k-237. The ConvKB paper similarly showed that a well-tuned TransE remains a strong baseline.

### Path-based and compositional reasoning

A second line looks at chains of relations rather than single edges. The Path Ranking Algorithm used paths as features; [Compositional Vector Space Models for Knowledge Base Completion](https://md.research.elamur.ai/free/en/papers/neelakantan-2015-compositional-vector-space-models-knowledge/) composed relation vectors along a path with a recurrent network, so that unseen paths could still be scored. DistMult's authors also mined Horn rules by multiplying relation matrices, an early bridge between embeddings and logical rules.

### Joining text and graph

[Representing Text for Joint Embedding of Text and Knowledge Bases](https://md.research.elamur.ai/free/en/papers/toutanova-2015-representing-text-joint-embedding-text/) put textual dependency paths and graph relations into one space, building each textual relation from its words rather than treating it as an opaque symbol; it also introduced FB15k-237 paired with ClueWeb text. [Distant Supervision for Relation Extraction with an Incomplete Knowledge Base](https://md.research.elamur.ai/free/en/papers/min-2013-distant-supervision-relation-extraction-incomplete/) attacked the reverse direction, showing that a missing fact in Freebase is not evidence that the fact is false.

### Language models and graphs

Since 2019 the central debate has been whether explicit graphs are still needed once language models absorb so many facts from text. The case for "the model is the knowledge base" rests on probing results such as LAMA. The case against rests on hallucination, the opacity of parametric knowledge, its staleness (examined in [Time-Aware Language Models as Temporal Knowledge Bases](https://md.research.elamur.ai/free/en/papers/dhingra-2022-time-aware-language-models-as/)), and evidence that carefully curated graphs still add knowledge that models lack, as argued in [COMET-ATOMIC 2020: On Symbolic and Neural Commonsense Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hwang-2020-comet-atomic-2020-symbolic-neural/). The survey [Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://md.research.elamur.ai/free/en/papers/pan-2023-unifying-large-language-models-knowledge/) organises the current landscape into graphs that improve language models, language models that improve graphs, and systems in which the two reason together.

Integration takes several forms: changing the architecture (K-BERT), converting the graph into text for pretraining ([Knowledge Graph Based Synthetic Corpus Generation for Knowledge-Enhanced Language Model Pre-training](https://md.research.elamur.ai/free/en/papers/agarwal-2021-knowledge-graph-based-synthetic-corpus/)), retrieving facts into a prompt ([Knowledge-Augmented Language Model Prompting for Zero-Shot Knowledge Graph Question Answering](https://md.research.elamur.ai/free/en/papers/baek-2023-knowledge-augmented-language-model-prompting/)), or using an ontology to steer retrieval ([OG-RAG: Ontology-grounded retrieval-augmented generation for large language models](https://md.research.elamur.ai/free/en/papers/sharma-2025-og-rag-ontology-grounded-retrieval/)). In the other direction, language models are used to propose ontology concepts, as in [Towards Ontology Construction with Language Models](https://md.research.elamur.ai/free/en/papers/funk-2023-towards-ontology-construction-language-models/), or to judge candidate correspondences, as in [Large Language Models as Oracles for Ontology Alignment](https://md.research.elamur.ai/free/en/papers/lushnei-2026-large-language-models-as-oracles/).

## Datasets, resources and benchmarks

The field rests on a handful of large public resources:

| Resource | Kind | Origin |
| --- | --- | --- |
| WordNet | lexical network of English word senses | Princeton, from the mid-1980s |
| Freebase | collaborative encyclopedic graph, later shut down and migrated to Wikidata | 2007 |
| DBpedia | RDF extracted from Wikipedia infoboxes, categories and links | 2007 |
| YAGO | Wikipedia facts anchored in the WordNet taxonomy | 2007; YAGO3 multilingual |
| Wikidata | crowdsourced, multilingual, with qualifiers and references | 2012 |
| ConceptNet | commonsense relations between free-text concepts | grew out of Open Mind Common Sense |
| ATOMIC | if-then knowledge about events and social situations | 2019; ATOMIC 2020 extended |

[Linked Data Quality of DBpedia, Freebase, OpenCyc, Wikidata, and YAGO](https://md.research.elamur.ai/free/en/papers/farber-2018-linked-data-quality-dbpedia-freebase/) compares five of these against a common set of quality criteria. How to put Wikidata's qualified statements into RDF is the question of [Reifying RDF: What Works Well With Wikidata?](https://md.research.elamur.ai/free/en/papers/hernandez-2015-reifying-rdf-what-works-well/). [The DARPA Wikidata Overlay: Wikidata as an ontology for natural language processing](https://md.research.elamur.ai/free/en/papers/spaulding-2023-darpa-wikidata-overlay-wikidata-as/) adds PropBank participant roles to Wikidata's event classes so that it can serve event extraction.

The main benchmarks for completion are FB15k and WN18, and their harder successors FB15k-237 and WN18RR, from which inverse-relation shortcuts were removed. For question answering, WebQuestions (introduced with SEMPRE), WebQuestionsSP with its SPARQL annotations, ComplexWebQuestions, Free917 and MetaQA appear repeatedly. For the alignment of text and triples, [T-REx: A Large Scale Alignment of Natural Language with Knowledge Base Triples](https://md.research.elamur.ai/free/en/papers/elsahar-2018-t-rex-large-scale-alignment/) links Wikipedia abstracts to Wikidata at large scale and later supplied the facts for the LAMA probe. For generation, the WebNLG data maps DBpedia triples to text. For dialogue, OpenDialKG pairs conversations with graph paths.

Lexical linked data has its own infrastructure: the OntoLex-Lemon model for lexicons, [Dbnary: Wiktionary as a LMF based Multilingual RDF network](https://md.research.elamur.ai/free/en/papers/serasset-2012-dbnary-wiktionary-as-lmf-based/) for Wiktionary, the MMoOn morpheme ontology, and CoNLL-RDF for annotated corpora.

## Landmark papers in this library

### Language Models as Knowledge Bases? (2019)

[Language Models as Knowledge Bases?](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/) introduced the LAMA probe, which rewrites facts from Wikidata, ConceptNet and SQuAD as fill-in-the-blank sentences and checks whether a pretrained model ranks the right object highly. BERT-large did surprisingly well, approaching a knowledge base built by a conventional extraction pipeline, though relations with many possible objects defeated every model. The paper started an entire literature on probing and on the limits of parametric knowledge, and it is the most cited work in this library.

### Knowledge Graphs (2020)

[Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hogan-2020-knowledge-graphs/) is a long tutorial survey by a large international team, built around a running example of a tourism graph for Chile. It covers data models and query languages, schema and identity, deductive reasoning with ontologies and rules, inductive methods from analytics to graph neural networks, and the whole life cycle of creation, refinement and publication. It has become the standard entry point to the field.

### TransD (2015)

[Knowledge Graph Embedding via Dynamic Mapping Matrix](https://md.research.elamur.ai/free/en/papers/ji-2015-knowledge-graph-embedding-dynamic-mapping/) gave each entity and relation a second, projection vector and built the projection matrix for every entity–relation pair from those vectors. This captured the diversity of entities attached to one relation while avoiding the cost of TransR's full matrices, and it was adopted widely as a translational baseline.

### DistMult and the neural tensor network (2013–2014)

[Reasoning With Neural Tensor Networks for Knowledge Base Completion](https://md.research.elamur.ai/free/en/papers/socher-2013-reasoning-neural-tensor-networks-knowledge/) modelled each relation with a bilinear tensor layer and represented entities by averaging the word vectors of their names, so that related entities could share evidence. A year later [Embedding Entities and Relations for Learning and Inference in Knowledge Bases](https://md.research.elamur.ai/free/en/papers/yang-2014-embedding-entities-relations-learning-inference/) placed such models in a common framework and found that the simplest variant, a diagonal bilinear form, outperformed TransE on Freebase. DistMult became a default baseline for the rest of the decade.

### TuckER (2019)

[TuckER: Tensor Factorization for Knowledge Graph Completion](https://md.research.elamur.ai/free/en/papers/balazevic-2019-tucker-tensor-factorization-knowledge-graph/) treated the graph as a binary three-way tensor and factorised it with a shared core tensor. The authors proved the model fully expressive and showed that RESCAL, DistMult, ComplEx and SimplE are its special cases, giving a tidy theoretical frame for the linear models while beating more complex nonlinear ones.

### Semantic parsing over Freebase (2013–2016)

[Semantic Parsing on Freebase from Question-Answer Pairs](https://md.research.elamur.ai/free/en/papers/berant-2013-semantic-parsing-freebase-question-answer/) trained a parser for a huge open-domain knowledge base using only question–answer pairs, with logical forms as latent variables, and released the WebQuestions dataset. [Semantic Parsing via Paraphrasing](https://md.research.elamur.ai/free/en/papers/berant-2014-semantic-parsing-paraphrasing/) inverted the problem by generating canonical sentences for candidate queries and choosing the one that best paraphrases the question. [Semantic Parsing via Staged Query Graph Generation: Question Answering with Knowledge Base](https://md.research.elamur.ai/free/en/papers/yih-2015-semantic-parsing-staged-query-graph/) built query graphs step by step, consulting the knowledge base from the first step, and [The Value of Semantic Parse Labeling for Knowledge Base Question Answering](https://md.research.elamur.ai/free/en/papers/yih-2016-value-semantic-parse-labeling-knowledge/) showed that full parse annotation (WebQuestionsSP) was cheaper than expected and improved accuracy.

### COMET (2019)

[COMET: Commonsense Transformers for Automatic Knowledge Graph Construction](https://md.research.elamur.ai/free/en/papers/bosselut-2019-comet-commonsense-transformers-automatic-knowledge/) recast knowledge base completion as generation: a pretrained transformer fine-tuned on ATOMIC or ConceptNet tuples writes new object phrases for a given subject and relation. Many of its outputs were judged plausible yet were new to the graph, making generation a serious alternative to extraction for commonsense resources, and COMET-ATOMIC 2020 extended both the graph and the argument.

### K-BERT (2019)

[K-BERT: Enabling Language Representation with Knowledge Graph](https://md.research.elamur.ai/free/en/papers/liu-2019-k-bert-enabling-language-representation/) inserted relevant triples into the input sentence as branches of a small tree, using soft positions and an attention mask so the added facts would not distort the original sentence. It needed no extra pretraining and helped mostly on domain tasks such as medical entity recognition, a pattern repeated by later knowledge-injection work.

### EmbedKGQA (2020)

[Improving Multi-hop Question Answering over Knowledge Graphs using Knowledge Base Embeddings](https://md.research.elamur.ai/free/en/papers/saxena-2020-improving-multi-hop-question-answering/) answered multi-hop questions by scoring every entity in the ComplEx embedding space against the question vector. Because embeddings can infer missing links, the method still found answers when the graph lacked the direct path, which made it robust on sparse graphs.

### The WebNLG challenge (2017)

[The WebNLG Challenge: Generating Text from RDF Data](https://md.research.elamur.ai/free/en/papers/gardent-2017-webnlg-challenge-generating-text-rdf/) asked systems to turn sets of DBpedia triples into text, treating lexical choice, aggregation, referring expressions and sentence planning as one problem. Neural systems led on familiar categories but collapsed on unseen ones, where grammar-based systems held up; the dataset became the standard benchmark for data-to-text generation.

Citations of the papers on this topic, by when they were published Papers published in 2015–2019 gather 61% of the 23,432 citations of the papers on this topic.

**The numbers**

| Published | Papers and their citations |
| --- | --- |
| 2025–2029 | 16 papers, 59 citations |
| 2020–2024 | 20 papers, 5,760 citations |
| 2015–2019 | 27 papers, 14,290 citations |
| 2010–2014 | 13 papers, 3,276 citations |
| 1990–1994 | 2 papers, 47 citations |

## People

[Jonathan Berant](https://md.research.elamur.ai/free/en/authors/jonathan-berant/), now at Google and Tel Aviv University, is the most prolific author here. His work with [Percy Liang](https://md.research.elamur.ai/free/en/authors/percy-liang/) at Stanford on Freebase semantic parsing and paraphrasing, and with [Alon Talmor](https://md.research.elamur.ai/free/en/authors/alon-talmor/) on [The Web as a Knowledge-Base for Answering Complex Questions](https://md.research.elamur.ai/free/en/papers/talmor-2018-web-as-knowledge-base-answering/), spans the move from curated graphs to the web as a source of answers.

[Wen-tau Yih](https://md.research.elamur.ai/free/en/authors/wen-tau-yih/), now at the University of Washington, co-wrote DistMult and the staged query graph parser and led the WebQuestionsSP annotation, with [Ming-Wei Chang](https://md.research.elamur.ai/free/en/authors/ming-wei-chang/), [Jianfeng Gao](https://md.research.elamur.ai/free/en/authors/jianfeng-gao/) of Microsoft and [Xiaodong He](https://md.research.elamur.ai/free/en/authors/xiaodong-he/).

[Danqi Chen](https://md.research.elamur.ai/free/en/authors/danqi-chen/), now at Princeton, appears on the neural tensor network paper with [Richard Socher](https://md.research.elamur.ai/free/en/authors/richard-socher/), [Christopher D. Manning](https://md.research.elamur.ai/free/en/authors/christopher-d-manning/) and [Andrew Y. Ng](https://md.research.elamur.ai/free/en/authors/andrew-y-ng/), and on both 2015 papers with [Kristina Toutanova](https://md.research.elamur.ai/free/en/authors/kristina-toutanova/), whose work on observed and latent features and on joint text–graph embeddings reshaped how completion is evaluated.

[Roberto Navigli](https://md.research.elamur.ai/free/en/authors/roberto-navigli/) of Sapienza University of Rome, known for BabelNet, co-wrote the Knowledge Graphs survey and, with [Michele Bevilacqua](https://md.research.elamur.ai/free/en/authors/michele-bevilacqua/), [Breaking Through the 80% Glass Ceiling: Raising the State of the Art in Word Sense Disambiguation by Incorporating Knowledge Graph Information](https://md.research.elamur.ai/free/en/papers/bevilacqua-2020-breaking-through-80-glass-ceiling/), which spread classifier scores along WordNet edges.

[Aidan Hogan](https://md.research.elamur.ai/free/en/authors/aidan-hogan/) of the University of Chile led the Knowledge Graphs survey and co-wrote the study of reifying Wikidata in RDF.

[Antoine Bosselut](https://md.research.elamur.ai/free/en/authors/antoine-bosselut/), now at EPFL, and [Yejin Choi](https://md.research.elamur.ai/free/en/authors/yejin-choi/), now at Stanford, are behind COMET and COMET-ATOMIC 2020, the central papers on commonsense knowledge graphs here.

[Shirui Pan](https://md.research.elamur.ai/free/en/authors/shirui-pan/), with [Yufei Wang](https://md.research.elamur.ai/free/en/authors/yufei-wang/) and [Chen Chen](https://md.research.elamur.ai/free/en/authors/chen-chen/), wrote the language model and knowledge graph roadmap; Chen and Wang also co-wrote [Dipping PLMs Sauce: Bridging Structure and Text for Effective Knowledge Graph Completion via Conditional Soft Prompting](https://md.research.elamur.ai/free/en/papers/chen-2023-dipping-plms-sauce-bridging-structure/).

[Zhiyuan Liu](https://md.research.elamur.ai/free/en/authors/zhiyuan-liu/) of Tsinghua University co-wrote PTransE and [Grounded Conversation Generation as Guided Traverses in Commonsense Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/zhang-2020-grounded-conversation-generation-as-guided/), which uses ConceptNet to steer dialogue toward related concepts.

[Christian Bizer](https://md.research.elamur.ai/free/en/authors/christian-bizer/) of the University of Mannheim, a founder of DBpedia and of the Linked Data movement, co-wrote both DBpedia papers here with [Max Jakob](https://md.research.elamur.ai/free/en/authors/max-jakob/) and [Jens Lehmann](https://md.research.elamur.ai/free/en/authors/jens-lehmann/).

[Fabian M. Suchanek](https://md.research.elamur.ai/free/en/authors/fabian-suchanek/), a creator of YAGO, is the senior author of the YAGO3 paper.

[Fabio Petroni](https://md.research.elamur.ai/free/en/authors/fabio-petroni/), [Sebastian Riedel](https://md.research.elamur.ai/free/en/authors/sebastian-riedel/), [Tim Rocktäschel](https://md.research.elamur.ai/free/en/authors/tim-rocktaschel/) and [Patrick Lewis](https://md.research.elamur.ai/free/en/authors/patrick-lewis/) wrote the LAMA paper.

[Eduard Hovy](https://md.research.elamur.ai/free/en/authors/eduard-hovy/), now at the University of Melbourne, links the oldest and a more recent strand: the 1992 interlingua ontology paper with [Sergei Nirenburg](https://md.research.elamur.ai/free/en/authors/sergei-nirenburg/), and [Ontology-Aware Token Embeddings for Prepositional Phrase Attachment](https://md.research.elamur.ai/free/en/papers/dasigi-2017-ontology-aware-token-embeddings-prepositional/), which builds token vectors from WordNet senses and hypernyms.

[Martha Palmer](https://md.research.elamur.ai/free/en/authors/martha-palmer/) connects ontologies to predicate–argument resources, with [SemLink+: FrameNet, VerbNet and Event Ontologies](https://md.research.elamur.ai/free/en/papers/palmer-2014-semlink-framenet-verbnet-event-ontologies/) and the DARPA Wikidata Overlay.

In linguistic linked data, [Christian Chiarcos](https://md.research.elamur.ai/free/en/authors/christian-chiarcos/) of the University of Augsburg, [John P. McCrae](https://md.research.elamur.ai/free/en/authors/john-p-mccrae/) of the University of Galway and [Bettina Klimek](https://md.research.elamur.ai/free/en/authors/bettina-klimek/) are the most visible names, through CoNLL-RDF, OntoLex-Lemon and the MMoOn morpheme ontology.

Authors with the most papers on this topic Jonathan Berant leads with 3 papers; the 12 authors shown wrote 27 of the papers on this topic between them, counting each paper once per author.

**The numbers**

| Author | Papers here · their citations |
| --- | --- |
| [Jonathan Berant](https://md.research.elamur.ai/free/en/authors/jonathan-berant/) | 3 papers · 1,782 citations |
| [Danqi Chen](https://md.research.elamur.ai/free/en/authors/danqi-chen/) | 3 papers · 1,368 citations |
| [Wen-tau Yih](https://md.research.elamur.ai/free/en/authors/wen-tau-yih/) | 3 papers · 915 citations |
| [Roberto Navigli](https://md.research.elamur.ai/free/en/authors/roberto-navigli/) | 2 papers · 2,243 citations |
| [Aidan Hogan](https://md.research.elamur.ai/free/en/authors/aidan-hogan/) | 2 papers · 2,018 citations |
| [Chen Chen](https://md.research.elamur.ai/free/en/authors/chen-chen/) | 2 papers · 1,620 citations |
| [Yufei Wang](https://md.research.elamur.ai/free/en/authors/yufei-wang/) | 2 papers · 1,620 citations |
| [Antoine Bosselut](https://md.research.elamur.ai/free/en/authors/antoine-bosselut/) | 2 papers · 1,517 citations |
| [Yejin Choi](https://md.research.elamur.ai/free/en/authors/yejin-choi/) | 2 papers · 1,517 citations |
| [Percy Liang](https://md.research.elamur.ai/free/en/authors/percy-liang/) | 2 papers · 1,466 citations |
| [Kristina Toutanova](https://md.research.elamur.ai/free/en/authors/kristina-toutanova/) | 2 papers · 1,114 citations |
| [Zhiyuan Liu](https://md.research.elamur.ai/free/en/authors/zhiyuan-liu/) | 2 papers · 876 citations |

## Institutions and places

[Google](https://md.research.elamur.ai/free/en/companies/google/) and [Google DeepMind](https://md.research.elamur.ai/free/en/companies/google-deepmind/) are the current homes of several authors here; [Microsoft](https://md.research.elamur.ai/free/en/companies/microsoft/) is where the staged query graph and WebQuestionsSP work was done. Among universities, [Stanford University](https://md.research.elamur.ai/free/en/universities/stanford-university/), [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/), the [University of Washington](https://md.research.elamur.ai/free/en/universities/university-of-washington/), [Princeton University](https://md.research.elamur.ai/free/en/universities/princeton-university/), [Cornell University](https://md.research.elamur.ai/free/en/universities/cornell-university/) and the [University of Pennsylvania](https://md.research.elamur.ai/free/en/universities/university-of-pennsylvania/) stand out in the United States, alongside the [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/), home of much commonsense-graph work. [Tel Aviv University](https://md.research.elamur.ai/free/en/universities/tel-aviv-university/) is the Israeli centre for question answering over knowledge bases.

China is the second country by authorship, with the [Chinese Academy of Sciences](https://md.research.elamur.ai/free/en/government/chinese-academy-of-sciences/), the [University of Chinese Academy of Sciences](https://md.research.elamur.ai/free/en/universities/university-of-chinese-academy-of-sciences/), [Wuhan University](https://md.research.elamur.ai/free/en/universities/wuhan-university/), [Beihang University](https://md.research.elamur.ai/free/en/universities/beihang-university/) and the [Beijing University of Posts and Telecommunications](https://md.research.elamur.ai/free/en/universities/beijing-university-of-posts-and-telecommunications/). In Europe, German work centres on linked data and DBpedia, with the [Center for Scalable Data Analytics and Artificial Intelligence](https://md.research.elamur.ai/free/en/universities/center-for-scalable-data-analytics-and-artificial-intelligence/) among the institutions, and French work includes the [Centre National de la Recherche Scientifique](https://md.research.elamur.ai/free/en/labs/centre-national-de-la-recherche-scientifique/).

Where the authors on this topic work now Google is the current home of authors of 9 papers on this topic; counted are the authors with a page here, by their latest institution.

**The numbers**

| Institution | Papers |
| --- | --- |
| [Google](https://md.research.elamur.ai/free/en/companies/google/) | 9 papers |
| [Chinese Academy of Sciences](https://md.research.elamur.ai/free/en/government/chinese-academy-of-sciences/) | 8 papers |
| [Google DeepMind](https://md.research.elamur.ai/free/en/companies/google-deepmind/) | 8 papers |
| [University of Chinese Academy of Sciences](https://md.research.elamur.ai/free/en/universities/university-of-chinese-academy-of-sciences/) | 7 papers |
| [Stanford University](https://md.research.elamur.ai/free/en/universities/stanford-university/) | 6 papers |
| [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/) | 5 papers |
| [University of Washington](https://md.research.elamur.ai/free/en/universities/university-of-washington/) | 5 papers |
| [Allen Institute for Artificial Intelligence](https://md.research.elamur.ai/free/en/labs/allen-institute-for-artificial-intelligence/) | 4 papers |
| [Microsoft](https://md.research.elamur.ai/free/en/companies/microsoft/) | 4 papers |
| [Princeton University](https://md.research.elamur.ai/free/en/universities/princeton-university/) | 4 papers |

Countries of the authors on this topic United States comes first: authors working there now wrote 33 papers on this topic.

**The numbers**

| Country | Papers |
| --- | --- |
| United States | 33 papers |
| China | 18 papers |
| United Kingdom | 16 papers |
| Germany | 11 papers |
| France | 8 papers |
| Australia | 5 papers |
| Israel | 5 papers |
| Italy | 4 papers |
| Hong Kong | 3 papers |
| Switzerland | 3 papers |

The venues show two cultures: ACL and EMNLP carry the embedding and question-answering work, while LREC hosts the resource papers, from DBpedia and Dbnary to T-REx and the linked-data treebanks.

Where the papers on this topic were published Proceedings of the Fifteenth Language Resources and Evaluation Conference carries the most, 4 of 78.

**The numbers**

| Venue | Papers |
| --- | --- |
| Proceedings of the Fifteenth Language Resources and Evaluation Conference | 4 papers |
| Proceedings of the 53rd Annual Meeting of the Association for Computational Linguistics and the 7th International Joint Conference on Natural Language Processing (Volume 1: Long Papers) | 3 papers |
| Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics | 3 papers |
| Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics | 3 papers |
| Proceedings of the 2013 Conference on Empirical Methods in Natural Language Processing | 2 papers |
| Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing | 2 papers |
| Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12) | 2 papers |
| Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018) | 2 papers |
| Proceedings of the Twelfth Language Resources and Evaluation Conference | 2 papers |
| CIDR 2015 | 1 paper |

## Open questions

**Explicit versus parametric knowledge.** How much of a knowledge graph's role can a language model absorb, and for which kinds of facts? Probing shows good recall of simple, frequent facts and poor recall of many-valued, rare or changing ones. Whether graphs become sources for retrieval, training data, or verification tools is unsettled.

**Evaluation.** Link prediction benchmarks have repeatedly turned out to contain shortcuts, and the closed-world assumption treats every missing fact as false. Text-to-SPARQL and KGQA benchmarks face similar questions of coverage and annotation quality.

**Concepts, events and time.** Most graphs are entity-centric. Events and their participants, as the DARPA Overlay and SemLink+ show, are poorly represented; and facts change, which static graphs and static models both handle badly. Answering questions about concepts rather than named entities is the problem behind [Ontology-Guided Reverse Thinking Makes Large Language Models Stronger on Knowledge Graph Question Answering](https://md.research.elamur.ai/free/en/papers/liu-2025-ontology-guided-reverse-thinking-makes/).

**Multilinguality and low-resource languages.** The large graphs are strongest in English. YAGO3, DBpedia's mapping communities and the Arabic resources show the effort needed to serve other languages, and [Developing an Arabic Infectious Disease Ontology to Include Non-Standard Terminology](https://md.research.elamur.ai/free/en/papers/alsudias-2020-developing-arabic-infectious-disease-ontology/) shows how dialect and lay vocabulary fall outside ontologies built from formal sources.

**Automatic construction with quality guarantees.** Language models can propose taxonomies and align ontologies, but their output follows everyday rather than expert classification and carries cultural bias. Combining cheap generation with reliable human or formal checking remains open.

**Representing linguistic knowledge itself.** Modelling paradigms, morphemes and trees in RDF is still under construction, as [Challenges for the Representation of Morphology in Ontology Lexicons](https://md.research.elamur.ai/free/en/papers/klimek-2019-ontolex-morphology/) and [Bridging the Gap between Ontologies and Dictionaries: Requirements and Implementation of a New Core for OntoLex-Lemon](https://md.research.elamur.ai/free/en/papers/mccrae-2026-ontolex-new-core/) make clear.

## Neighbouring topics

The closest neighbour by shared papers is [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/), through probing, knowledge injection and the roadmap literature. [Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) shares the work on multi-hop inference, rules and commonsense. [Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/) and [Semantic parsing & meaning representation](https://md.research.elamur.ai/free/en/topics/semantic-parsing/) share the KBQA papers. [Information extraction](https://md.research.elamur.ai/free/en/topics/information-extraction/) and [Named entity recognition](https://md.research.elamur.ai/free/en/topics/named-entity-recognition/) cover the steps of mining and linking that populate a graph, with [Matching the Blanks: Distributional Similarity for Relation Learning](https://md.research.elamur.ai/free/en/papers/soares-2019-matching-blanks-distributional-similarity-relation/) a shared landmark.

The Arabic and Hebrew ontology and lexicon papers connect the topic to [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/), [Hebrew & Aramaic](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/) and [Morphology](https://md.research.elamur.ai/free/en/topics/morphology/). The linked-data treebanks tie it to [Universal Dependencies](https://md.research.elamur.ai/free/en/topics/universal-dependencies/) and [Latin](https://md.research.elamur.ai/free/en/topics/latin/), through [The Index Thomisticus Treebank as Linked Data in the LiLa Knowledge Base](https://md.research.elamur.ai/free/en/papers/mambrini-2022-index-thomisticus-treebank-as-linked/). Knowledge-enhanced stance detection links it to [Sentiment & emotion](https://md.research.elamur.ai/free/en/topics/sentiment-emotion/); graph-grounded conversation to [Dialogue systems](https://md.research.elamur.ai/free/en/topics/dialogue-systems/); event ontologies to [Semantic role labeling](https://md.research.elamur.ai/free/en/topics/semantic-role-labeling/); the idiom graph to [Idioms & multiword expressions](https://md.research.elamur.ai/free/en/topics/multiword-expressions/); and WordNet-based disambiguation to [Word sense disambiguation](https://md.research.elamur.ai/free/en/topics/word-sense-disambiguation/) and [Lexical semantics & word embeddings](https://md.research.elamur.ai/free/en/topics/lexical-semantics/). Interlingua ontologies recall the topic's roots in [Machine translation](https://md.research.elamur.ai/free/en/topics/machine-translation/).

Neighbouring topics: papers shared Language models & pretraining shares the most: 10 of the 78 papers here (13%) stand on both.

**The numbers**

| Topic | Papers shared |
| --- | --- |
| [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) | 10 shared · 13% |
| [Reasoning](https://md.research.elamur.ai/free/en/topics/reasoning/) | 10 shared · 13% |
| [Question answering](https://md.research.elamur.ai/free/en/topics/question-answering/) | 7 shared · 9% |
| [Multilingual & cross-lingual NLP](https://md.research.elamur.ai/free/en/topics/multilingual-nlp/) | 6 shared · 8% |
| [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) | 6 shared · 8% |
| [Semantic parsing & meaning representation](https://md.research.elamur.ai/free/en/topics/semantic-parsing/) | 4 shared · 5% |
| [Information extraction](https://md.research.elamur.ai/free/en/topics/information-extraction/) | 3 shared · 4% |
| [Hebrew & Aramaic](https://md.research.elamur.ai/free/en/topics/hebrew-aramaic/) | 2 shared · 3% |
| [Morphology](https://md.research.elamur.ai/free/en/topics/morphology/) | 2 shared · 3% |
| [Named entity recognition](https://md.research.elamur.ai/free/en/topics/named-entity-recognition/) | 2 shared · 3% |

By shelf, the topic stands mostly on [AI & machine learning](https://md.research.elamur.ai/free/en/shelves/ai/), then on [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/) and [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/), with a smaller presence on the [Machine translation](https://md.research.elamur.ai/free/en/shelves/machine-translation/) shelf.

The shelves this topic stands on AI & machine learning: 33 of the 78 papers here.

**The numbers**

| Shelf | Papers |
| --- | --- |
| [AI & machine learning](https://md.research.elamur.ai/free/en/shelves/ai/) | 33 papers |
| [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/) | 27 papers |
| [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/) | 20 papers |
| [Machine translation](https://md.research.elamur.ai/free/en/shelves/machine-translation/) | 6 papers |
| [Other](https://md.research.elamur.ai/free/en/shelves/other/) | 1 paper |

## Key works in this library

- [Language Models as Knowledge Bases?](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/) — the LAMA probe of factual knowledge in pretrained models.
- [Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hogan-2020-knowledge-graphs/) — the standard tutorial survey of the field.
- [Knowledge Graph Embedding via Dynamic Mapping Matrix](https://md.research.elamur.ai/free/en/papers/ji-2015-knowledge-graph-embedding-dynamic-mapping/) — TransD, translational embeddings with per-pair projections.
- [Unifying Large Language Models and Knowledge Graphs: A Roadmap](https://md.research.elamur.ai/free/en/papers/pan-2023-unifying-large-language-models-knowledge/) — three ways to combine graphs and large models.
- [COMET: Commonsense Transformers for Automatic Knowledge Graph Construction](https://md.research.elamur.ai/free/en/papers/bosselut-2019-comet-commonsense-transformers-automatic-knowledge/) — generating commonsense triples with a transformer.
- [K-BERT: Enabling Language Representation with Knowledge Graph](https://md.research.elamur.ai/free/en/papers/liu-2019-k-bert-enabling-language-representation/) — injecting triples into BERT's input.
- [A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network](https://md.research.elamur.ai/free/en/papers/nguyen-2018-novel-embedding-model-knowledge-base/) — ConvKB, convolution over whole triples.
- [Representing Text for Joint Embedding of Text and Knowledge Bases](https://md.research.elamur.ai/free/en/papers/toutanova-2015-representing-text-joint-embedding-text/) — compositional textual relations; FB15k-237 with text.
- [TuckER: Tensor Factorization for Knowledge Graph Completion](https://md.research.elamur.ai/free/en/papers/balazevic-2019-tucker-tensor-factorization-knowledge-graph/) — a unifying linear factorisation model.
- [Semantic Parsing on Freebase from Question-Answer Pairs](https://md.research.elamur.ai/free/en/papers/berant-2013-semantic-parsing-freebase-question-answer/) — weakly supervised KBQA and WebQuestions.
- [The WebNLG Challenge: Generating Text from RDF Data](https://md.research.elamur.ai/free/en/papers/gardent-2017-webnlg-challenge-generating-text-rdf/) — the data-to-text benchmark over DBpedia.
- [DBpedia – A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia](https://md.research.elamur.ai/free/en/papers/lehmann-2015-dbpedia-large-scale-multilingual-knowledge/) — the main account of DBpedia.
- [Distant Supervision for Relation Extraction with an Incomplete Knowledge Base](https://md.research.elamur.ai/free/en/papers/min-2013-distant-supervision-relation-extraction-incomplete/) — treating missing facts as unlabelled, not negative.
- [COMET-ATOMIC 2020: On Symbolic and Neural Commonsense Knowledge Graphs](https://md.research.elamur.ai/free/en/papers/hwang-2020-comet-atomic-2020-symbolic-neural/) — why curated commonsense graphs still matter.
- [The Arabic Ontology -- An Arabic Wordnet with Ontologically Clean Content](https://md.research.elamur.ai/free/en/papers/jarrar-2022-arabic-ontology-arabic-wordnet-ontologically/) — an expert-grounded Arabic wordnet.

## All 78 papers on this topic

**1** [2](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/page/2/) [next →](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/page/2/)

1. [**Language Models as Knowledge Bases?**Fabio Petroni, Tim Rocktäschel, Sebastian Riedel, et al.*2019 · 2 977 citations*](https://md.research.elamur.ai/free/en/papers/petroni-2019-language-models-as-knowledge-bases/)
2. [**Knowledge Graphs**Aidan Hogan, Eva Blomqvist, Michael Cochez, et al.*2020 · 2 011 citations*](https://md.research.elamur.ai/free/en/papers/hogan-2020-knowledge-graphs/)
3. [**Knowledge Graph Embedding via Dynamic Mapping Matrix**Guoliang Ji, Shizhu He, Liheng Xu, et al.*2015 · 1 936 citations*](https://md.research.elamur.ai/free/en/papers/ji-2015-knowledge-graph-embedding-dynamic-mapping/)
4. [**Unifying Large Language Models and Knowledge Graphs: A Roadmap**Shirui Pan, Linhao Luo, Yufei Wang, et al.*2023 · 1 545 citations*](https://md.research.elamur.ai/free/en/papers/pan-2023-unifying-large-language-models-knowledge/)
5. [**COMET: Commonsense Transformers for Automatic Knowledge Graph Construction**Antoine Bosselut, Hannah Rashkin, Maarten Sap, et al.*2019 · 1 282 citations*](https://md.research.elamur.ai/free/en/papers/bosselut-2019-comet-commonsense-transformers-automatic-knowledge/)
6. [**K-BERT: Enabling Language Representation with Knowledge Graph**Weijie Liu, Peng Zhou, Zhe Zhao, et al.*2019 · 1 006 citations*](https://md.research.elamur.ai/free/en/papers/liu-2019-k-bert-enabling-language-representation/)
7. [**A Novel Embedding Model for Knowledge Base Completion Based on Convolutional Neural Network**Dai Quoc Nguyen, Tu Dinh Nguyen, Dat Quoc Nguyen, et al.*2018 · 944 citations*](https://md.research.elamur.ai/free/en/papers/nguyen-2018-novel-embedding-model-knowledge-base/)
8. [**Representing Text for Joint Embedding of Text and Knowledge Bases**Kristina Toutanova, Danqi Chen, Patrick Pantel, et al.*2015 · 904 citations*](https://md.research.elamur.ai/free/en/papers/toutanova-2015-representing-text-joint-embedding-text/)
9. [**TuckER: Tensor Factorization for Knowledge Graph Completion**Ivana Balažević, Carl Allen, Timothy M. Hospedales*2019 · 792 citations*](https://md.research.elamur.ai/free/en/papers/balazevic-2019-tucker-tensor-factorization-knowledge-graph/)
10. [**Semantic Parsing on Freebase from Question-Answer Pairs**Jonathan Berant, Andrew Chou, Roy Frostig, et al.*2013 · 785 citations*](https://md.research.elamur.ai/free/en/papers/berant-2013-semantic-parsing-freebase-question-answer/)
11. [**Modeling Relation Paths for Representation Learning of Knowledge Bases**Yankai Lin, Zhiyuan Liu, Huanbo Luan, et al.*2015 · 695 citations*](https://md.research.elamur.ai/free/en/papers/lin-2015-modeling-relation-paths-representation-learning/)
12. [**Semantic Parsing via Paraphrasing**Jonathan Berant, Percy Liang*2014 · 681 citations*](https://md.research.elamur.ai/free/en/papers/berant-2014-semantic-parsing-paraphrasing/)
13. [**Improving Multi-hop Question Answering over Knowledge Graphs using Knowledge Base Embeddings**Apoorv Saxena, Aditay Tripathi, Partha Talukdar*2020 · 670 citations*](https://md.research.elamur.ai/free/en/papers/saxena-2020-improving-multi-hop-question-answering/)
14. [**The WebNLG Challenge: Generating Text from RDF Data**Claire Gardent, Anastasia Shimorina, Shashi Narayan, et al.*2017 · 579 citations*](https://md.research.elamur.ai/free/en/papers/gardent-2017-webnlg-challenge-generating-text-rdf/)
15. [**Embedding Entities and Relations for Learning and Inference in Knowledge Bases**Bishan Yang, Wen-tau Yih, Xiaodong He, et al.*2014 · 458 citations*](https://md.research.elamur.ai/free/en/papers/yang-2014-embedding-entities-relations-learning-inference/)
16. [**OpenDialKG: Explainable Conversational Reasoning with Attention-based Walks over Knowledge Graphs**Seungwhan Moon, Pararth Shah, Anuj Kumar, et al.*2019 · 444 citations*](https://md.research.elamur.ai/free/en/papers/moon-2019-opendialkg-explainable-conversational-reasoning-attention/)
17. [**DBpedia – A Large-scale, Multilingual Knowledge Base Extracted from Wikipedia**Jens Lehmann, Robert Isele, Max Jakob, et al.*2015 · 394 citations*](https://md.research.elamur.ai/free/en/papers/lehmann-2015-dbpedia-large-scale-multilingual-knowledge/)
18. [**Entity Disambiguation for Knowledge Base Population**Mark Dredze, Paul McNamee, Delip Rao, et al.*2010 · 394 citations*](https://md.research.elamur.ai/free/en/papers/dredze-2010-entity-disambiguation-knowledge-base-population/)
19. [**The Web as a Knowledge-Base for Answering Complex Questions**Alon Talmor, Jonathan Berant*2018 · 316 citations*](https://md.research.elamur.ai/free/en/papers/talmor-2018-web-as-knowledge-base-answering/)
20. [**Distant Supervision for Relation Extraction with an Incomplete Knowledge Base**Bonan Min, Ralph Grishman, Li Wan, et al.*2013 · 299 citations*](https://md.research.elamur.ai/free/en/papers/min-2013-distant-supervision-relation-extraction-incomplete/)
21. [**Matching the Blanks: Distributional Similarity for Relation Learning**Livio Baldini Soares, Nicholas FitzGerald, Jeffrey Ling, et al.*2019 · 294 citations*](https://md.research.elamur.ai/free/en/papers/soares-2019-matching-blanks-distributional-similarity-relation/)
22. [**Text Generation from Knowledge Graphs with Graph Transformers**Rik Koncel-Kedziorski, Dhanush Bekal, Yi Luan, et al.*2019 · 284 citations*](https://md.research.elamur.ai/free/en/papers/kedziorski-2019-text-generation-knowledge-graphs-graph/)
23. [**The Value of Semantic Parse Labeling for Knowledge Base Question Answering**Wen-tau Yih, Matthew Richardson, Chris Meek, et al.*2016 · 254 citations*](https://md.research.elamur.ai/free/en/papers/yih-2016-value-semantic-parse-labeling-knowledge/)
24. [**Reasoning With Neural Tensor Networks for Knowledge Base Completion**Richard Socher, Danqi Chen, Christopher D. Manning, et al.*2013 · 254 citations*](https://md.research.elamur.ai/free/en/papers/socher-2013-reasoning-neural-tensor-networks-knowledge/)
25. [**Knowledge Graph Based Synthetic Corpus Generation for Knowledge-Enhanced Language Model Pre-training**Oshin Agarwal, Heming Ge, Siamak Shakeri, et al.*2021 · 238 citations*](https://md.research.elamur.ai/free/en/papers/agarwal-2021-knowledge-graph-based-synthetic-corpus/)

**1** [2](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/page/2/) [next →](https://md.research.elamur.ai/free/en/topics/knowledge-graphs/page/2/)

- [← previous topic Interpretability](https://md.research.elamur.ai/free/en/topics/interpretability/)
- [next topic → Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/)
- [↑ back to topics](https://md.research.elamur.ai/free/en/topics/)