Pith. sign in

REVIEW 4 major objections 5 minor 28 references

Tripartite-GraphRAG via Plugin Ontologies

T0 review · 4 major / 5 minor · reviewed 2026-08-16 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims that organizing documents, domain concepts, and query objects in a tripartite knowledge graph lets LLM prompts recover more concepts per token than naive RAG.

desk verdict A coherent workshop proposal with a genuinely new tripartite retrieval schema, undermined by an asymmetric three-case evaluation that does not back the headline density claim. read the letter →

arxiv 2504.19667 v3 pith:ISFGWAX6 submitted 2025-04-28 cs.LG cs.AI

Michael Banf , Johannes Kuhn This is my paper · ORCID
classification cs.LGcs.AI
keywords TripartiteknowledgegraphRAGRetrieval-augmentedgenerationPluginontologyPromptoptimizationInformationdensityHealthcarequestionansweringUnsupervisednodeclassification
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proposes a retrieval-augmented generation method in which documents, domain concepts, and complex query objects such as patient anamneses are organized into a tripartite knowledge graph before any question is asked. By summarizing each text chunk with respect to each ontology concept, and each object with respect to each concept, the method creates compact concept-anchored descriptions. Prompt assembly is then treated as an unsupervised node classification problem: a concept-chunk summary is included only if the cosine similarity between its embedding and the object-concept summary embedding exceeds a per-concept statistical threshold. On a healthcare test case, the paper reports that this recovers about eight medical concepts per prompt at roughly 1,500 to 5,500 tokens, whereas a naive RAG baseline recovers four to six concepts using 3,000 to 31,000 tokens. The authors argue this yields cost savings and more consistent, better-arranged answers.

What carries the argument

The load-bearing object is the tripartite knowledge graph with three node types: objects of investigation $o$, ontology concepts $c$, and text chunks $t$, connected by edges $e_{o,c}$ and $e_{c,t}$ whose properties are LLM-generated, concept-specific summaries. From this graph the paper builds a classification graph $G'_o$ whose binary nodes $x\in\{0,1\}$ represent whether the information stored in each $e_{c,t}$ should enter the prompt for object $o$. Inclusion is governed by $w_{o,c,t} = \cos(E_{e_{o,c}}, E_{e_{c,t}})$, the cosine similarity between the embeddings of the two edge summaries; the score is compared with the concept-specific empirical distribution $P_c$ of all such scores, selecting $x=1$ when $P_c(w_{o,c,t}) > \alpha$, with a second, more lenient threshold $\beta$ applied when another concept's summary from the same chunk has already been selected. This transformation is what lets prompt creation be posed as an unsupervised node classification problem and is the mechanism claimed to concentrate relevant information into shorter prompts.

What would settle it

Run the identical healthcare setup but shuffle the observed similarity scores $w_{o,c,t}$ across concept-chunk pairs, preserving the marginal distribution $P_c$ before applying the same $\alpha$ and $\beta$ thresholds; if the number of relevant guideline statements appearing in the final prompts does not fall, then the cosine proxy is not doing the selection work. More directly, build human-annotated relevance labels for all concept-chunk pairs in the six guidelines and compare precision and recall of selected chunks against a random-chunk baseline at matched token budgets.

Watch

Extended reading notes

Core claim

The central claim is that concept-anchored pre-compression and graph-structured prompt selection make LLM question answering more information-dense than embedding-similarity chunk retrieval. Concretely, the paper argues that by first compressing each text chunk into per-concept summaries and by scoring those summaries against an object-concept summary via cosine similarity, the method can discard irrelevant text before prompt construction. The classification step uses each concept's empirical distribution of those similarity scores, with an inclusion threshold $\alpha$ and a co-occurrence threshold $\beta$, so that a chunk discussing several query-relevant concepts is not dropped too early. In the reported healthcare experiment, Tripartite-GraphRAG recovered on average eight concepts at 1,500\,--\,5,500 tokens, compared with naive RAG's four to six concepts at 3,000\,--\,31,000 tokens, and it linked specific anamnesis findings such as blood pressure to two relevant guideline statements that RAG failed to compare.

Load-bearing premise

The method's load-bearing premise is that the cosine similarity between LLM-written summaries of an object-concept relationship and a concept-chunk relationship is a trustworthy measure of relevance, and that a threshold taken from the per-concept distribution of those scores will separate useful from useless context without any human-verified labels.

Editorial extensions

If this is right

  • If the density claim holds, a domain RAG system can answer the same multi-concept queries with roughly one-quarter to one-tenth the tokens, directly lowering inference cost.
  • Because each selected edge summary carries its document reference and because summaries are ordered by the lexical graph's chunk succession, outputs gain source traceability that ordinary chunk retrieval lacks.
  • The per-concept empirical threshold means new documents can be added without re-clustering or entity resolution, since each new chunk is judged only by its edge summaries and the concept's existing score distribution.
  • The two-threshold scheme lets a chunk about two concepts survive even if one concept's score is low, a mechanism the paper argues improves coverage in multi-concept analyses.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The paper does not test whether the density gain transfers outside curated medical ontologies; the gain may be largest where a domain ontology already cleanly partitions the vocabulary, and weaker in heterogeneous general texts.
  • Because $P_c$ is built from observed similarities, the method is sensitive to how many objects and chunks populate each concept; with very few samples, the top-percentile threshold could be unstable, so a bootstrap or Bayesian estimate of $P_c$ is a testable improvement.
  • The blood-pressure example suggests that concept-anchored retrieval may reduce generic answers by arranging evidence by concept before generation, a hypothesis that could be quantified on a larger set of anamneses by measuring factual entailment or citation precision.
  • If the tripartite structure is extended to deeper ontology hierarchies, edge summaries at different levels would likely need separate distributions $P_c$; otherwise the percentile comparison mixes granularities.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes Tripartite-GraphRAG, a retrieval-augmented generation method built on a tripartite knowledge graph whose nodes are domain objects, curated concept nodes, and text chunks. Edges between concepts and chunks and between objects and concepts are created by LLM-generated, concept-specific summaries. Prompt construction for a query about an object is framed as an unsupervised node classification problem: a per-concept empirical distribution of cosine similarities between object-concept and concept-chunk summaries is thresholded (with a secondary interaction threshold) to decide which chunks enter the prompt. The paper reports a healthcare evaluation on three fictitious patient anamneses and six ESC guideline documents, claiming that Tripartite-GraphRAG recovers on average 8 concepts at roughly 1500-5500 tokens versus 4-6 concepts at 3000-31000 tokens for a naive RAG baseline.

Significance. If validated, the tripartite plugin-ontology design is a genuinely useful idea for domain-specific RAG: it avoids entity resolution and deduplication during graph extension, provides a natural provenance path through document-related chunk edges, and recasts prompt selection as a graph-classification problem. The authors are also candid about several limitations. However, the current evidence is not yet sufficient to support the central claim: the quantitative comparison rests on three self-generated cases, the 'concepts recovered' measure is never defined operationally, the two compared pipelines receive different inputs, the cosine-similarity relevance proxy is unvalidated, and no code, data, or significance testing is provided. The method is promising but the paper's headline quantitative claim needs a much more careful evaluation before it can be accepted as a result.

major comments (4)
  1. [Section 3 (first experiment)] The central quantitative claim, 'recovering on average 8 concepts at approx. 1500-5500 tokens, compared to RAG's on average 4-6 concepts at approx. 3000-31000 tokens', is not supported because the paper never defines 'recovered concepts' operationally. It is unclear whether a concept is counted when it appears in the constructed prompt, when it appears in the final LLM answer, or only when it is explicitly supported by a retrieved source statement; it is also unclear who performs the annotation and whether counts are normalized per anamnesis object. With only three objects, no error bars, no confidence intervals, and no significance test, the word 'significantly' has no statistical backing. Please provide a precise annotation protocol, per-object counts, and raw data or confidence intervals.
  2. [Sections 2.2 and 3] The claimed information-density advantage is confounded by an input asymmetry between the two systems. In the Tripartite-GraphRAG prompt, the method first lists the concept-specific summaries e_{o,c} for every concept connected to the object, and these summaries are generated in Section 2.1 by explicitly scanning the anamnesis against the curated 27-concept ontology. A naive RAG prompt, by contrast, contains only retrieved text chunks. If 'concepts recovered' is counted from the final prompt or answer, the tripartite pipeline receives a pre-enumerated concept inventory that RAG does not receive, so the observed gap could be explained without any contribution from graph retrieval. A controlled comparison should either give RAG the same concept inventory as part of the query, or count as recovered only those concepts whose supporting evidence comes from the retrieved source chunks.
  3. [Section 2.2, w_{o,c,t} definition] The relevance proxy is unvalidated. The method assumes that cos(E_{e_{o,c}}, E_{e_{c,t}}) is a faithful measure of whether the information stored in e_{c,t} is relevant to the object's association with concept c, and that the per-concept empirical distribution P_c of these scores provides a meaningful absolute threshold. Because P_c is estimated from the very same cosine similarities that are then thresholded, the selection rule is calibrated to the method's own statistics rather than to an external standard of relevance. Please validate the proxy against human relevance judgments or against a small set of known-relevant chunks, reporting precision and recall of the thresholded selection; a single held-out oracle set of relevant chunks would be enough to make the threshold credible.
  4. [Section 2.1] The paper describes the concept-specific edge summaries as 'information-preserving pre-compression', but no experiment measures whether the LLM extraction retains all information relevant to the concept. If extraction recall is low, the graph-based selection cannot recover what was already lost at construction time. The term 'information-preserving' should be substantiated by a recall measurement, for example by comparing the salient facts or concepts in the original text chunk with those present in the extracted summary e_{c,t} for the same chunk.
minor comments (5)
  1. [Section 2.2] Please clarify whether P_c(w) is a cumulative distribution function or a probability density; as written, the condition 'P_c(w) > alpha' with alpha = 0.9 is ambiguous and the reader cannot tell whether the threshold is a quantile, a tail probability, or a raw density value.
  2. [Section 2.2] The notation 'x ∈ 0, 1' should be written as 'x ∈ {0, 1}' to denote the binary state of the classification node.
  3. [References] References [6] and [11] appear to refer to the same arXiv paper, and references [15] and [19] appear to refer to the same survey; duplicate citations under different numbers are confusing and should be consolidated.
  4. [Section 3] The exact prompts used for the naive RAG baseline and for the Tripartite-GraphRAG final query are not given; 'devising similar queries' is not precise enough to reproduce the comparison, and the prompts should be included in a supplement or appendix.
  5. [General] No data or code availability statement is provided; please state whether the generated anamneses, the guideline-derived chunks, the constructed Neo4j graph, or the evaluation annotations will be released for reproducibility.

Circularity Check

2 steps flagged · score 6.0 of 10

The headline information-density comparison reduces to pre-injected e_{o,c} concept summaries; the P_c threshold is self-referential.

  1. fitted input called prediction [Section 2.1, 2.2, and 3 (prompt construction and density claim)]
    "Again, we task the LLM to scan all objects o regarding any connection to any of the concepts c and, in turn, create relationships e_{o,c}. ... A final prompt regarding the analysis of o is constructed by: i) initially stating a basic query ...; ii) followed by a sequential listing of the concept-specific summaries in e_{o,c}; iii) each separately followed the concept-specific summaries stored in all selected e_{c,t} ... In comparison, we observe a significantly higher information density for Tripartite-GraphRAG, recovering on average 8 concepts at approx."

    The 'concepts recovered' figure in Section 3 is not operationally defined, but the Tripartite prompt is constructed by explicitly listing the concept-specific summaries in e_{o,c}, and e_{o,c} is generated by scanning the anamnesis object against the curated concept ontology. Therefore, any concept count taken from the prompt or from the LLM answer is largely predetermined by the input construction: the concepts are injected before retrieval-based selection happens. The reported 8-versus-4-6 density advantage over naive RAG is thus partly forced by the asymmetry that Tripartite-GraphRAG receives an explicit concept inventory and object-side summaries that naive RAG does not receive, rather than by the graph retrieval or the P_c-based selection mechanism.

  2. self definitional [Section 2.2 (P_c threshold definition)]
    "By collecting all w_{o,c,t} for all o and t connected to a specific concept c, we derive an empirical probability distribution P_c per each c for all observed similarity scores w_{o,c,t} w.r.t. c. Thus, x can be assigned a probability P_c(w_{o,c,t}) w.r.t. the concept c's specific P_c. If P_c(w_{o,c,t})>α then x=1"

    The selection rule defines relevance through the cosine similarity w, then builds P_c from the empirical distribution of exactly those same w values, and finally accepts a node exactly when its own score is in the upper tail of that same distribution. There is no external relevance label, held-out statistic, or independent target against which the threshold is validated. 'Semantic relevance' is therefore defined as being a high quantile of the method's own scoring distribution; the optimization and pre-filtering behavior is a tautology of the construction rather than a derived or empirically grounded property.

full rationale

Walking the derivation chain, the main claimed result is the Section 3 information-density advantage: about 8 concepts recovered at 1500-5500 tokens versus RAG's 4-6 concepts at 3000-31000 tokens. The paper never defines what counts as a 'recovered concept,' who annotates the count, or how RAG outputs are mapped to the 27-concept ontology. More importantly, the Tripartite prompt is built by first listing the concept-specific summaries e_{o,c} for every concept connected to the object o, and those e_{o,c} were created by explicitly scanning the anamnesis against the curated ontology (Section 2.1). Thus the concept inventory is an input to the method and is placed directly into the prompt before any graph-based chunk selection occurs. The 'recovered concepts' claim is therefore, at least in part, a re-statement of the method's own input rather than an independent prediction, which is the core circular step. A second, milder self-reference is the P_c threshold: inclusion is judged by the quantile of the same cosine similarities used to score relevance, so 'optimal' selection is defined relative to the method's own statistic. This is not an invalid algorithm by itself, but it is not an external validation. I found no load-bearing self-citation: the MRF citation [14] is only inspirational, and the lexical-graph citation [10] is not used to justify the central claim. The paper's own Discussion admits the evaluation is 'primarily anecdotal' and limited to naive RAG, which supports the conclusion that the quantitative headline is not yet independently grounded. Overall, the central quantitative claim partially reduces by construction, warranting a score of 6; the underlying graph-construction idea may still have independent merit, but the measured advantage is confounded.

Assumptions & free parameters 3 free parameters · 5 assumptions · 0 invented entities

The pipeline leans on several unvalidated modeling choices: the cosine relevance proxy, the empirical-distribution thresholding, the manual concept ontology, and the claim that LLM pre-compression preserves information. The authors disclose the manual ontology as a limitation, but the relevance proxy and thresholding assumptions are load-bearing and untested.

free parameters (3)
  • alpha = 0.9
    Threshold on the concept-specific empirical distribution of cosine similarities; chosen empirically without systematic optimization, as stated in Section 4.
  • beta = 0.5
    Co-discussion threshold for including a second concept's chunk when a paired concept is selected; chosen empirically without systematic optimization, as stated in Section 4.
  • Medical concept ontology = 27 concepts in 4 classes, manually curated
    The concept vocabulary determines which information is extracted and later counted as recovered. Manual curation is acknowledged as a bottleneck and bias source in Sections 2.1 and 4.
assumptions (5)
  • domain assumption Cosine similarity between embeddings of LLM-generated edge summaries is a valid measure of semantic relevance for prompt inclusion.
    Section 2.2 defines w_{o,c,t}=cos(E_{e_{o,c}},E_{e_{c,t}}) and uses it as the basis for x; no validation against human relevance judgments is provided.
  • domain assumption The empirical distribution P_c across observed similarity scores is a valid probability model for thresholding.
    Section 2.2 assigns x=1 when P_c(w)>alpha; this assumes the observed range of similarities is representative and that a percentile cutoff separates relevant from irrelevant chunks.
  • domain assumption LLM concept-anchored pre-compression preserves all information that later queries may need.
    Section 2.1 calls the summarization 'information-preserving' and Figure 2 shows text being discarded, but no completeness guarantee is provided.
  • ad hoc to paper The manually curated 27-concept ontology is adequate for the use case and is invariant to ontology-creation method.
    Section 2.1 admits the ontology is 'adhoc, i.e. manually curated', and the Discussion identifies manual curation as a bottleneck and a source of bias.
  • standard math Markov Blanket local independence justifies the beta co-selection rule.
    The beta rule is motivated by Markov Random Field principles [13,14], but its validity for this particular e_{c,t} co-occurrence graph is asserted rather than shown.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Tripartite-GraphRAG via Plugin Ontologies." pith.science (2026). https://pith.science/paper/ISFGWAX6

@misc{pith2026250419667,
  author       = {Pith},
  title        = {Pith review of: Tripartite-GraphRAG via Plugin Ontologies},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ISFGWAX6}},
  note         = {Machine review of arXiv:2504.19667}
}
read the original abstract

Large Language Models (LLMs) have shown remarkable capabilities across various domains, yet they struggle with knowledge-intensive tasks in areas that demand factual accuracy, e.g. industrial automation and healthcare. Key limitations include their tendency to hallucinate, lack of source traceability (provenance), and challenges in timely knowledge updates. Combining language models with knowledge graphs (GraphRAG) offers promising avenues for overcoming these deficits. However, a major challenge lies in creating such a knowledge graph in the first place. Here, we propose a novel approach that combines LLMs with a tripartite knowledge graph representation, which is constructed by connecting complex, domain-specific objects via a curated ontology of corresponding, domain-specific concepts to relevant sections within chunks of text through a concept-anchored pre-analysis of source documents starting from an initial lexical graph. Subsequently, we formulate LLM prompt creation as an unsupervised node classification problem allowing for the optimization of information density, coverage, and arrangement of LLM prompts at significantly reduced lengths. An initial experimental evaluation of our approach on a healthcare use case, involving multi-faceted analyses of patient anamneses given a set of medical concepts as well as a series of clinical guideline literature, indicates its potential to optimize information density, coverage, and arrangement of LLM prompts while significantly reducing their lengths, which, in turn, may lead to reduced costs as well as more consistent and reliable LLM outputs.

Figures

Figures reproduced from arXiv: 2504.19667 by the authors.

Figure 1
Figure 1. A tripartite knowledge graph representation connecting complex, domain-specific objects of investi￾gation (yellow) via a curated ontology of corresponding, domain-specific concepts (green) to relevant sections within chunks of text (blue) through a concept-anchored pre-analysis of source documents. However, a major challenge lies in creating such a knowledge graph in the first place - on the basis of suitable data s… view at source ↗
Figure 2
Figure 2. Left: object of investigation 𝑜 (yellow), here fictitious patient anamnesis, and information-preserving pre-compression of 𝑜 w.r.t concept 𝑐 (green, 𝑐 = blood pressure). Right: information-preserving pre-compression of a text chunk 𝑡 (blue) w.r.t. 𝑐, i.e. retaining relevant information (red) while discarding most unrelated text. approach implements a concept-specific, information-preserving pre-compression of textua… view at source ↗
Figure 3
Figure 3. Illustration of graph transformation of 𝐺 related to object 𝑜 to create classification graph 𝐺′ 𝑜 that represents the information in each 𝑒𝑐,𝑡 between concept 𝑐 and text chunk 𝑡 related to 𝑜, measured as 𝑤𝑜,𝑐,𝑡 in relation to 𝑐’s specific distribution of similarity scores 𝑃𝑐. Here 𝐸𝑒𝑜,𝑐 , 𝐸𝑒𝑐,𝑡 denote the vector embeddings of the information stored in 𝑒𝑜,𝑐 and 𝑒𝑐,𝑡 with 𝑐𝑜𝑠(.) denoting the cosine similarity. 𝑤𝑜=Barb… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Illustration of the derivation of 𝑃𝑐 via collection of all combinations 𝑤𝑜,𝑐,𝑡 = 𝑐𝑜𝑠(𝐸𝑒𝑜,𝑐 , 𝐸𝑒𝑐,𝑡 ) for all 𝑜 ∈ (Peter Mueller, Barbara Schmidt, Oskar Lehm) and 𝑡 ∈ (text chunk I, text chunk II), as illustrated in the respective table and exemplary histogram. E.g. ass…
Figure 5
Figure 5. Figure 5: The generated tripartite medical knowledge graph representation in Neo4j. 4. Discussion: current limitations and ongoing research Our Tripartite-GraphRAG approach demonstrates a novel paradigm for combining LLMs with knowl￾edge graphs through concept-anchored pre-analy…
Figure 6
Figure 6. Figure 6: In contrast to a generic response by RAG (left), Tripartite-GraphRAG (right) links finding (yellow) from a patient anamnesis to corresponding concept (green) and identifies two relevant, concept-specific source statements (blue) for examination. References [1] Lewis, P…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

28 extracted references · 23 canonical work pages

  1. [1]

    Retrieval-augmented generation for knowledge-intensive nlp tasks

    Lewis, P., et al. Retrieval-augmented generation for knowledge-intensive nlp tasks. In Advances in Neural Information Processing Systems, volume 33, pages 9459–9474, 2020

    work page 2020

  2. [2]

    Improving language models by retrieving from trillions of tokens

    Borgeaud, S., et al. Improving language models by retrieving from trillions of tokens. In International Conference on Machine Learning, pages 2206–2240. PMLR, 2022

    work page 2022

  3. [3]

    Active retrieval augmented generation

    Jiang, Z., et al. Active retrieval augmented generation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 7969–7992, 2023

    work page 2023

  4. [4]

    Fashion Image Retrieval with Multi-Granular Alignment

    Xu, F.F., Shi, W., and Choi, E. Retrieval-augmented generation for long-form question answering. arXiv preprint arXiv:2302.08902, 2023

    work page Pith review arXiv 2023

  5. [5]

    In-context retrieval-augmented language models

    Ram, O., et al. In-context retrieval-augmented language models. Transactions of the Association for Computational Linguistics, 11:1316–1331, 2023

    work page 2023

  6. [7]

    Deep bidirectional language-knowledge graph pretraining

    Yasunaga, M., et al. Deep bidirectional language-knowledge graph pretraining. In Advances in Neural Information Processing Systems, volume 35, pages 37309–37323, 2022

    work page 2022

  7. [8]

    G-retriever: Retrieval-augmented generation for textual graph understanding and question answering

    He, X., et al. G-retriever: Retrieval-augmented generation for textual graph understanding and question answering. arXiv preprint arXiv:2402.07630, 2024

    arXiv 2024

  8. [9]

    Graphrag: Unlocking llm discovery on narrative private data

    Zhang, J., et al. Graphrag: Unlocking llm discovery on narrative private data. arXiv preprint arXiv:2404.01473, 2024

    arXiv 2024

Show all 28 references
  1. [10]

    Building A Graph and LLM-Powered RAG Application from PDF Documents, https://neo4j.com/blog/developer/ graph-llm-rag-application-pdf-documents/, last accessed 2025/04/16

    Yu, F. Building A Graph and LLM-Powered RAG Application from PDF Documents, https://neo4j.com/blog/developer/ graph-llm-rag-application-pdf-documents/, last accessed 2025/04/16

    2025

  2. [11]

    Edge, D. et al. From Local to Global: A Graph RAG Approach to Query-Focused Summarization. arxiv (2025) https: //arxiv.org/abs/2404.16130

    2025 arXiv

  3. [12]

    Neo4j LLM Knowledge Graph Builder, https://neo4j.com/labs/genai-ecosystem/llm-graph-builder/, last accessed 2025/04/16

    Neo4j labs. Neo4j LLM Knowledge Graph Builder, https://neo4j.com/labs/genai-ecosystem/llm-graph-builder/, last accessed 2025/04/16

    2025

  4. [13]

    Probabilistic Graphical Models

    Koller, D., Friedman, N. Probabilistic Graphical Models. MIT Press. (2009)

    2009

  5. [14]

    Banf, M., Rhee, S. Y. Enhancing gene regulatory network inference through data integration with markov random fields. Scientific Reports 7(1) Nature Publishing (2017)

    2017

  6. [16]

    Human languages with greater information density have higher communication speed but lower conversation breadth

    Aceves, P., Evans, J.A. Human languages with greater information density have higher communication speed but lower conversation breadth. Nat Hum Behav 8, 644–656 (2024)

    2024

  7. [17]

    RAPTOR: Recursive abstractive processing for tree-organized retrieval

    Sarthi, P., et al. RAPTOR: Recursive abstractive processing for tree-organized retrieval. In International Conference on Learning Representations (ICLR), 2024

    2024

  8. [18]

    RAG-Fusion: A new take on retrieval-augmented generation

    Rackauckas, C. RAG-Fusion: A new take on retrieval-augmented generation. arXiv preprint arXiv:2402.03367, 2024

    2024 arXiv

  9. [19]

    Retrieval-augmented generation for large language models: A survey

    Gao, Y., et al. Retrieval-augmented generation for large language models: A survey. arXiv preprint arXiv:2312.10997, 2023

    2023 arXiv

  10. [20]

    SNOMED-CT: The advanced terminology and coding system for eHealth

    Donnelly, K. SNOMED-CT: The advanced terminology and coding system for eHealth. Studies in Health Technology and Informatics, 121:279–290, 2006

    2006

  11. [21]

    The unified medical language system (UMLS): integrating biomedical terminology

    Bodenreider, O. The unified medical language system (UMLS): integrating biomedical terminology. Nucleic Acids Research, 32(suppl_1):D267–D270, 2004

    2004

  12. [22]

    The Human Phenotype Ontology: a tool for annotating and analyzing human hereditary disease

    Robinson, P.N., et al. The Human Phenotype Ontology: a tool for annotating and analyzing human hereditary disease. The American Journal of Human Genetics, 83(5):610–615, 2008

    2008

  13. [23]

    What disease does this patient have? A large-scale open domain question answering dataset from medical exams

    Jin, D., et al. What disease does this patient have? A large-scale open domain question answering dataset from medical exams. Applied Sciences, 11(14):6421, 2021

    2021

  14. [24]

    PubMedQA: A dataset for biomedical research question answering

    Jin, Q., et al. PubMedQA: A dataset for biomedical research question answering. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 2567–2577, 2019

    2019

  15. [25]

    Unifying large language models and knowledge graphs: A roadmap

    Pan, J.Z., et al. Unifying large language models and knowledge graphs: A roadmap. IEEE Transactions on Knowledge and Data Engineering, 36(7):3580–3599, 2024

    2024

  16. [26]

    and Lukasiewicz, T

    Hohenecker, P. and Lukasiewicz, T. Ontology reasoning with deep neural networks. Journal of Artificial Intelligence Research, 68:503–540, 2020

    2020

  17. [27]

    OWL2Vec*: Embedding of OWL ontologies

    Chen, J., et al. OWL2Vec*: Embedding of OWL ontologies. Machine Learning, 112(7):2397–2421, 2023

    2023

  18. [28]

    LLM-guided ontology reasoning for improving knowledge graph quality

    Ye, H., et al. LLM-guided ontology reasoning for improving knowledge graph quality. Knowledge-Based Systems, 289:111518, 2024

    2024

  19. [29]

    Human phenotype ontology-based semantic analysis of large language models for clinical interpretation

    Kommineni, V.K., et al. Human phenotype ontology-based semantic analysis of large language models for clinical interpretation. Bioinformatics, 40(3):btae089, 2024

    2024

  20. [30]

    OntoGLM: Ontology-grounded large language model for enhanced medical reasoning

    Liu, Y., et al. OntoGLM: Ontology-grounded large language model for enhanced medical reasoning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 8234–8249, 2024

    2024

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.