Solved Status: 850 Problems

GPT-5.5-pro analyzed every identified research problem in the KG field. Not a single one is fully solved.

761

Active (89.5%)

Superseded (6.5%)

Partial (4%)

Fully Solved

0.75

Avg Confidence

Zero problems are fully solved or abandoned. The entire KG research field is in active evolution — problems get partially addressed or reframed, but never fully closed. LLM-related problems (hallucination, knowledge limitations) are the fastest-growing, appearing only since 2021.

Top Active Problems (by paper count)

Problem	Papers	Years	Status	Conf
kg incompleteness KG incompleteness remains a fundamental open-world problem: embedding-based knowledge graph completion, open-world KGC benchmarks, representation lear...	1,726	1995–2026	active	0.93
kg question answering KG question answering has seen substantial progress through semantic parsing, entity-linking pipelines, neural retrieval/reasoning models, KGQA system...	348	2015–2026	active	0.92
automated knowledge graph construction Automated knowledge graph construction has progressed through extraction/linking pipelines, domain-specific construction methods such as course knowle...	303	2009–2026	active	0.90
knowledge graph embedding learning Knowledge graph embedding learning has become a mature and widely used approach for KG completion and link prediction, with standard families such as ...	233	2015–2026	active	0.88
llm hallucination LLM hallucination in KG contexts is being actively mitigated through KG-enhanced LLMs, KG-enhanced RAG, graph reasoning, QA systems, and agentic retri...	197	2023–2026	active	0.92
temporal kg reasoning Temporal KG reasoning has seen strong progress through temporal KG embeddings, attention/GNN-based temporal encoders, subgraph extraction, adaptive te...	184	2016–2026	active	0.90
high-dimensional kge cost High-dimensional KGE cost has been attacked through tensor/matrix factorization, more efficient training frameworks, convolutional or dynamic embeddin...	181	2014–2026	active	0.84
kg recommendation data sparsity Knowledge graphs have clearly helped alleviate recommender-system data sparsity by injecting entity relations, attributes, paths, and neighborhoods in...	160	2016–2026	active	0.86
llm knowledge limitations LLM knowledge limitations remain an active problem: KG-enhanced LLMs, KG-enhanced RAG, graph retrieval, prompting, and KG reasoning have improved fact...	141	2021–2026	active	0.88
kg link prediction KG link prediction has made substantial progress through knowledge graph embeddings, tensor/matrix factorization, GNN-based models, temporal regulariz...	130	2015–2026	active	0.88
data sparsity and cold start KG-based recommendation has become the dominant mitigation strategy for sparsity and cold start, using entity relations, embeddings, graph attention/p...	130	2016–2026	active	0.89
heterogeneous data integration Heterogeneous data integration remains an open KG problem: ontology/semantic integration, KG construction pipelines, embeddings/representation learnin...	127	2005–2026	active	0.87
domain-specific kg construction challenges The literature has moved from conventional domain-specific construction pipelines, graph databases, and manually engineered schemas toward LLM-assiste...	112	2016–2026	active	0.82
kg entity alignment KG entity alignment has seen substantial methodological progress, moving from embedding and GCN/attention-based approaches toward hierarchical alignme...	94	2018–2026	active	0.88
rag knowledge conflicts RAG knowledge conflicts are a very recent and rapidly growing problem, especially where biomedical answers must reconcile retrieved evidence, knowledg...	94	2024–2026	active	0.78
multi-hop kg reasoning Multi-hop KG reasoning remains an active open problem: early RL/path-based methods and later GNN, subgraph, noise-aware, KGQA, and LLM-augmented appro...	89	2017–2026	active	0.91
static kg temporal limitation The community has mainly addressed static KG temporal limitations by moving toward temporal knowledge graphs and time-aware embedding or tensor models...	76	2015–2026	active	0.84
llm hallucination in kgc LLM hallucination in knowledge graph completion has become a sharply active problem as LLMs are increasingly used to predict missing KG facts, generat...	75	2020–2026	active	0.88
joint entity-relation extraction Joint entity-relation extraction has seen substantial progress from neural joint models, especially transformer-based span, table-filling, sequence-to...	75	2008–2026	active	0.88
large-scale knowledge graph management Large-scale knowledge graph management has seen incremental techniques such as scalable reasoning, distributed KG use, embedding variants, corpus augm...	65	2015–2026	active	0.78

Partially Solved (34)

Problem	Papers	Key Solution	Year	Conf
cf sparsity and cold start	26	knowledge-graph collaborative filtering with knowledge-enhanced GNNs and relation-aware attention propagation	2021	0.78
multi-modal kg resource need	24	Multimodal KG construction and enrichment using DBpedia/Wikidata extensions, domain-specific MMKG construction, and multimodal knowledge fusion/representation learning	2022	0.67
up-to-date biomedical knowledge access	17	automated biomedical knowledge graph construction using biomedical NER/relation extraction, disease-specific KG generators, and KG-enhanced retrieval interfaces	2021	0.72
biodiversity data integration	14	domain-specific biodiversity knowledge graphs with FAIR/global identifier reconciliation	2018	0.74
author name disambiguation	13	metadata-rich, graph-based author identity resolution using multimodal literal features, bibliometric/semantic KG context, and curated evaluation datasets	2021	0.68
research hotspot identification	13	CiteSpace-based bibliometric knowledge graph analysis with co-citation, keyword co-occurrence, burst detection, and visualization	2006	0.76
traditional recommender limitations	12	KG-enhanced recommendation using knowledge graph embeddings and GNN-based reasoning	2019	0.72
heterogeneous vulnerability data management	12	ontology- and RDF/SPARQL-based domain knowledge graphs, often reusing Wikidata or domain ontologies	2020	0.68
human disease mechanism representation	10	PheKnowLator-style biomedical knowledge graph construction with ontology/data-source integration and OWL reasoning closure	2021	0.72
collaborative filtering limitations	9	knowledge graph-enhanced collaborative filtering / knowledge graph-based recommendation	2020	0.74
relation pattern modeling	8	RotatE-style relational rotation embeddings with self-adversarial negative sampling	2019	0.78
heterogeneous climate data integration	7	linked-data/RDF climate knowledge graph platforms with SPARQL endpoints, graph databases, and emerging virtual knowledge graph integration	2022	0.72
scalable reasoning integration	7	restricted-fragment scalable reasoning, especially OWL 2 RL/Datalog-style distributed materialization, with probabilistic soft logic for soft or uncertain inference	2016	0.66
seed-dependent entity alignment	7	semi-supervised seed expansion and one-to-one matching, later complemented by LLM/text-based entity alignment	2021	0.72
distant supervision label noise	7	bag-level multi-instance learning with attention/instance selection, often combined with relation embeddings	2016	0.78
medical named entity recognition	7	BERT-based neural sequence labeling, especially BERT-BiLSTM-CRF and KG-infused BERT variants	2020	0.74
systematic review need	6	systematic literature reviews and paradigm-comparison/resource-brief frameworks	2023	0.68
drug discovery data integration	6	biomedical knowledge graph construction with RDF/SPARQL access, exemplified by OREGANO drug knowledge graph resources	2023	0.68
incomplete type constraints	6	local closed-world type approximation, later complemented by soft type-constrained latent and embedding models	2015	0.68
knowledge noise in kg-enhanced text	6	K-BERT-style knowledge injection with soft-position embeddings and visible-matrix attention masking	2019	0.68
missing semantic hierarchy modeling	6	HAKE-style polar-coordinate / radial-angular hierarchy-aware embeddings	2019	0.74
semantic scholarly querying	6	Open Research Knowledge Graph (ORKG) infrastructure with semantic scholarly APIs and structured contribution modeling	2018	0.72
translation kge misses relation patterns	6	relation-pattern-aware KGE models, especially RotatE/HRotatE-style rotation embeddings and rule-augmented embedding methods	2019	0.72
using side information for recommendation	5	KG-enhanced recommendation using multi-task feature learning, especially MKR-style cross-compress feature sharing between recommendation and KG representation learning	2019	0.70
power equipment information retrieval	5	domain-specific power knowledge graph construction with a semantic retrieval layer	2019	0.62
knowledge graph querying	5	standardized graph query languages and optimized graph/RDF query engines, especially SPARQL/SPARQL 1.1, triplestores, graph-database APIs, and abstract query-interface layers	2013	0.68
loss of kg structure in linearized encoders	5	structure-preserving graph encoders, especially bidirectional Graph2Seq/subgraph-GCN encoders with node-level copy decoders	2019	0.64
scalable rdf knowledge graph creation from complex data	5	RML-family declarative mappings with scalable RDFizer engines such as SDM-RDFizer and optimized logical-operator execution	2020	0.70
multi-domain dialogue state tracking	5	DST-as-question-answering with turn-level domain-slot questions, augmented by dynamic or conversational knowledge-graph state representations	2019	0.62
user-item graph insufficiency	5	knowledge graph-based recommendation with KG-enhanced user/item representations, propagation/attention, and contrastive-learning variants	2022	0.68
no datasets for video kg extraction	5	video KG extraction task formulations with heterogeneous video knowledge graph datasets	2020	0.68
traditional database limitations	5	native graph databases and RDF/graph stores, especially Neo4j graph storage, combined with graph-specific indexing, pruning, and query optimization	2015	0.64
ned-based retrieval limitation	5	CLOCQ-style top-k KB item retrieval with KB-aware pruning	2021	0.68
persistent dataset metadata and citation	5	JSON-LD knowledge-graph dataset descriptions paired with persistent DOI metadata records	2026	0.68

Superseded (55)

Problem	Papers	What Happened	Conf
scholarly kg knowledge discovery	34	The standalone problem of doing knowledge discovery directly over scholarly KGs has not converged on a standard solution: most work is proof-of-concept, with small clusters around semantic graph parti	0.72
incomplete knowledge bases	21	The underlying issue of incomplete knowledge bases has not been solved, but the problem label has largely been absorbed into knowledge graph completion, link prediction, entity linking, cross-lingual	0.74
triple-independent kg embeddings	16	The limitation of treating KG triples as independent has been mitigated by coupled tensor-matrix factorization, latent imputation, neighborhood attention, multi-hop reasoning, and rule/logic-enhanced	0.74
external knowledge underuse	14	Work on external knowledge underuse mainly tried KG/GNN-style augmentation, including syntax-knowledge GCNs, KG representations and embeddings, heterogeneous document graphs, entity-comparison network	0.72
knowledge base completion	13	Knowledge base completion was not truly solved; embedding, classification, co-occurrence, multilingual, hyperbolic, and GNN-based methods improved benchmark link prediction but did not eliminate open-	0.78
natural language question understanding	11	As a stand-alone KGQA problem, natural language question understanding has not been resolved by a single broadly adopted solution. The listed work is mostly complete KGQA systems, domain-specific know	0.72
target-specific kg exploitation	11	Target-specific KG exploitation saw a small burst of work around 2020–2022 using target-specific subgraph distillation, knowledge-aware attention, KG embeddings, and entity-description augmentation, b	0.68
pipeline error propagation	10	Pipeline error propagation has not been cleanly solved as a standalone entity-linking problem; instead, it has been mitigated through joint/end-to-end EL and KGQA models, zero-shot evaluation splits,	0.72
covid-19 information overload	10	COVID-19 information overload was addressed mainly through COVID-specific knowledge graph applications, Neo4j-backed graph stores, biomedical entity enrichment, and discovery frameworks, which helped	0.72
covid-19 drug repurposing	9	COVID-19 drug repurposing with knowledge graphs was mainly an urgent early-pandemic application: groups built or merged biomedical KGs, selected semantic predications, and applied probabilistic reason	0.82
document-based scholarly communication	8	The original framing—moving scholarly communication from static documents to machine-actionable KG-based artifacts—had a small burst of work around 2019, mainly around ORKG-style scholarly KG infrastr	0.74
entity-pair relation prediction	8	Entity-pair relation prediction had a small burst of work around 2018-2019 using latent path modeling, structural supervised methods, variational architectures such as DiVa, attention embeddings such	0.74
scalable semantic web kr	8	The exact framing of "scalable semantic web KR" shows little sustained activity: after one 2008 paper, it reappeared sparsely from 2018 to 2022, mostly as representation, benchmark, survey, visualizat	0.74
specific disease kg reasoning gap	8	Early work tried to close disease-specific KG reasoning gaps through NER/relation-extraction benchmarks, SDKG construction pipelines/resources, curated schema frameworks, and embedding or tensor-based	0.66
language-specific relation extraction limitation	8	As a standalone KG extraction problem, language-specific relation extraction limitation has not been convincingly solved; the proposed fixes are one-off supervised resources, transfer experiments, fea	0.72