Density, asymmetry and citation dynamics in scientific literature

Nathaniel Imel; Zachary Hafen

REVIEW 3 major objections 5 minor 20 references

Density, asymmetry and citation dynamics in scientific literature

T0 review · 3 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read Adding a measure of semantic-neighborhood density to citation-forecasting models yields consistent, if modest, out-of-sample gains, while an edge-position asymmetry measure does not.

desk verdict A clean framework and a useful null result, but the main predictive claim is compromised by model selection on the test set. read the letter →

arxiv 2506.23366 v1 pith:VOKJ4E35 submitted 2025-06-29 cs.DL cs.CLcs.SI

Nathaniel Imel , Zachary Hafen This is my paper · ORCID

classification cs.DLcs.CLcs.SI

keywords citationpredictionsemanticembeddingsscientificimpactnearestneighborsdensitymetricBayesianhierarchicalregressionscienceofnovelty

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 asks whether a publication's similarity to the literature that precedes it foreshadows how often it will be cited. It defines two geometric measures in a semantic embedding space: density ($\rho$), the number of a paper's nearest previous neighbors per radian of angle enclosing them, and asymmetry ($\alpha$), the average direction from a paper to those neighbors. Testing $\rho$ and $\alpha$ against citation rates for roughly 53,000 papers in nine disciplines with five different embeddings, the authors find that adding density-based predictors to a baseline model (publication age and reference count) consistently improves out-of-sample prediction of log citations per year, even though each individual effect is small and uncertain. Asymmetry does not improve prediction. The result matters because it turns the abstract idea that science balances tradition and innovation into a measurable, scalable signal linking semantic position to scholarly recognition.

What carries the argument

The argument runs on two local-geometry metrics computed from document embeddings, plus a Bayesian hierarchical regression. Density is defined as $\rho(p_i)=k/\arccos(p_i^\top p_k)$: with embeddings normalized to unit length, the inner product is cosine similarity and the arc-cosine is the angular distance to the $k$-th nearest previous neighbor, so $\rho$ counts neighboring papers per radian on the unit hypersphere. Asymmetry, $\alpha(p_i) = \frac{1}{k}\left\|\sum_{j=1}^{k} \frac{p_i-p_j}{\|p_i-p_j\|}\right\|$, is the magnitude of the net unit direction from a paper to its neighbors. These metrics are plugged into hierarchical linear models predicting log citations per year, standardized within each of nine fields, with field-level coefficients drawn from shared normal distributions; 27 model variants and six evaluation metrics are used to test whether adding $\rho$ or $\alpha$ improves held-out prediction.

What would settle it

Re-run the same Bayesian hierarchical models on a sample that does not require the 16 nearest neighbors to stay fixed under further data collection, using the same embeddings and covariates. If the out-of-sample gain from $\rho$ disappears or changes sign once unstable-neighborhood papers are included, the reported predictive signal is an artifact of the stability filter rather than a general property of semantic density.

Watch

Extended reading notes

Core claim

The central claim is that the local density of a publication's semantic neighborhood carries real, if modest, information about its future citation rate. For each paper, density is computed as $\rho(p_i)=k/\arccos(p_i^\top p_k)$, the ratio of the $k=16$ nearest previously published neighbors to the angle enclosing the farthest of them, using cosine distance on unit-normalized embeddings; asymmetry is the average unit direction from the paper to those neighbors. Fitting a Bayesian hierarchical linear regression to log citations per year, with coefficients for each field, the authors report that models including density estimates from the five embeddings together outperform the reference-count-plus-age baseline on every evaluation metric, while models adding $\alpha$ do not. Individual posterior effects of $\rho$ hover near zero; bag-of-words density has the largest magnitude ($\beta\approx -0.05$) and SciBERT density the second largest ($\beta\approx -0.04$), both consistent with zero, so the paper frames the result as weak but systematic structure rather than a strong direct cause.

Load-bearing premise

The analysis assumes that papers whose 16 nearest neighbors stayed fixed while more papers were collected are a fair sample of all science; if unstable neighborhoods mark fast-moving or frontier topics, those papers drop out and the measured density-citation link would be skewed.

Editorial extensions

If this is right

Adding the set of density predictors computed from five embeddings improves out-of-sample prediction when combined with age and reference count, so local semantic density is a usable covariate for citation forecasting.
Because the gain appears even when individual posterior effects are near zero, the signal is aggregate and context-dependent rather than a fixed-size effect.
Asymmetry, the paper's measure of lying at the edge of a neighborhood, shows no predictive benefit and can slightly hurt performance, so prototypicality and edge-position are not interchangeable predictors.
The approach is scalable and released with code and a software library, so density and asymmetry can be recomputed for other corpora, embeddings, or bibliographic sources without bespoke domain feature extraction.
Effect sizes vary by field and embedding, meaning the density-citation relationship is not universal in magnitude and should be estimated hierarchically rather than as a single global coefficient.

Reading between the lines

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

An implication the authors leave implicit is that the sign pattern across embeddings is informative: bag-of-words density is the strongest predictor and negative, whereas contextual embeddings are near null, hinting that surface-level lexical crowding may mildly suppress citations while deeper semantic density neither helps nor hurts.
A plausible mechanism they do not test is that density bundles two opposing channels — larger potential audience (positive) and less perceived novelty (negative) — and the near-zero individual effects are the net of these forces; a future study could separate them with mediating variables such as team size or venue prestige.
A testable extension is to split citations by age window and ask whether density predicts early attention better than late impact; if density tracks community recognition, its predictive gain should be largest in the first years after publication.
Because the stability filter excludes papers whose 16-neighbor set changed when 1,000 new papers were added, frontier and rapidly growing fields are likely underrepresented; replicating on a complete or time-stamped corpus could reveal whether the density signal strengthens precisely where the literature is moving fastest.

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.

Desk Editor's Note

read the letter

The headline is that the paper's main positive claim doesn't survive a close look. The density metric ρ is a neat, simple construction, and the null result for asymmetry is a genuinely useful negative finding. But the evidence that adding ρ improves out-of-sample prediction is compromised by model selection on the test set, so the reported gain (1−RMSE 0.037 vs 0.024) is likely overstated.

What the paper does well: the metrics are clearly defined, the framework is released as code plus a pip-installable library, and the survey across five embeddings and nine fields is thorough. The limitations section is honest about missing validation and controls. That is real work.

The soft spots are concentrated in the statistics. Section 4 says all 27 model variations were evaluated on the same 25% held-out test set, and the 'best-fit' model was chosen from those evaluations. Reporting that model's test-set score as evidence of out-of-sample improvement is a classic selection-on-the-test-set bias. The fact that six metrics pick the same model doesn't help—they all share the same holdout, so they share the bias. You'd need a nested cross-validation, or at least model selection on a separate validation partition, before trusting the gap over the baseline.

There is also a secondary concern with the data filter: requiring each paper's 16 nearest neighbors to remain unchanged while 1,000 papers are added probably selects against frontier, fast-moving areas. That could bias the density-citation relationship, though I'm less certain how strongly.

The individual effect sizes are mostly consistent with zero, so the predictive improvement is the only evidence for the density signal. Once that evidence is tainted, the paper's central claim is not established. The asymmetry null and the release of tools remain useful.

Who would get value? Someone working on embedding-based scientometrics who wants a well-scoped framework and a cautionary example of test-set selection. I'd send it to peer review, because the question and the released code deserve a proper evaluation—but I'd expect heavy revision, and the current central claim should not be taken as is.

Referee Report

3 major / 5 minor

Summary. The manuscript defines two embedding-based metrics for a publication's local semantic neighborhood—density (ρ) and asymmetry (α)—and tests whether these predict citation rate (log citations per year) using Bayesian hierarchical regression on roughly 53,000 papers across nine disciplines and five embedding models. The central claim is that adding density predictors to a baseline with reference count and publication age improves out-of-sample prediction, while asymmetry does not, and the authors release code and data to support replication.

Significance. If the central claim were established, the paper would offer a scalable, domain-general bridge between semantic embedding geometry and scientific impact, with several commendable features: five embedding models, explicit temporal exclusion of future papers when computing neighborhoods, hierarchical modeling across fields, and publicly available code and data. The metrics are not directly circular with the citation outcome, since they are computed from abstracts rather than from citation data. However, the predictive claim is currently not established because model selection and evaluation share the same held-out test set, and the sample construction filters out neighborhoods that change during data collection, which may bias the density–citation relationship.

major comments (3)

[Section 4, Bayesian statistical modeling] The authors state that all 27 model variations were evaluated on the same held-out test set (25% of the data) and that all six metrics identified the same best-fit model. Because the best model was selected using this test set, the reported test-set performance in Table 1 (1−RMSE 0.037 vs. 0.024) is optimistically biased; the consistency of the six metrics does not remove this bias, since the metrics are highly correlated and all reflect the same in-test-set selection. The improvement from adding ρ is therefore not established by the current evidence. Please redo the model comparison with a nested scheme (for example, a separate model-selection set or cross-validation within the training fold) and report the performance of the final model on a truly untouched test set.
[Section 4, Data] The analysis retains only papers whose 16 nearest neighbors are unchanged after adding N=1000 publications, with k=16 selected as a free parameter. This filter is likely to exclude papers in fast-moving or frontier areas where neighborhoods are unstable, and such papers may have systematically different density–citation relationships. The assumption that this yields a representative sample is asserted rather than tested. Please characterize the excluded papers (for example, compare their citation rates, fields, and embedding positions with the retained sample) and provide a sensitivity analysis over k and N, or model the selection step explicitly.
[Section 5, Table 1 and Figure 3] The central quantitative evidence is a single point estimate (1−RMSE) without uncertainty intervals, and Figure 4 displays only within-model scaled scores rather than raw metric values. Given the small differences reported (0.024 vs. 0.037), the fact that all posterior effect sizes for ρ are individually consistent with zero (Figure 3), and the model-selection issue above, the paper should report posterior or bootstrap intervals for performance differences across all six metrics, and should show that the improvement persists when each embedding model or field is removed in turn.

minor comments (5)

[Section 3.2, Equation (2)] Equation (2) is typeset incorrectly in the preprint; the vector norm notation is garbled and should be cleaned up for readability.
[Section 4, Data] The sentence 'This criterion was satisfied by only 10% of papers retrieved from S2' is ambiguous: please clarify whether this is the joint criterion (abstract, publication date, field association) and report how many papers were retrieved at each iterative step.
[Section 4, Document embeddings] For SBERT, treating each abstract as a single sentence should be justified, since abstracts are typically multi-sentence texts and this choice may affect the quality of the resulting embeddings.
[Appendix B, Table 2] The classification accuracies in Table 2 are reported without confidence intervals; this is especially important for the GPT-2 reduced-embedding case (0.25 accuracy), which is near chance.
[Section 6, Limitations] The authors appropriately acknowledge several limitations (no expert validation, no controls for author h-index or journal impact factor, and possibly coarse operationalization of density and asymmetry); these acknowledgments are helpful, but the revision should explain how the proposed reanalysis addresses them or why they do not affect the predictive claim.

Circularity Check

1 steps flagged · score 6.0 of 10

Held-out test set used for model selection and evaluation inflates the reported density improvement.

fitted input called prediction [Section 4 (Bayesian statistical modeling) and Section 5, Table 1]
"The performance of all model variations was evaluated on a held-out test set comprising 25% of our data using six distinct metrics: root mean squared error, mean absolute error, mean absolute percentage error, R2, and the D2 pinball and absolute error scores. All six metrics consistently identified the same best-fit model and yielded the same performance ranking across the other model variations, providing confidence in our model selection and evaluation."

The same held-out test set is used both to select the best among 27 model variations and to report that best model's performance (Table 1: 1-RMSE 0.037 with rho vs 0.024 without). The reported out-of-sample improvement is therefore the maximum test-set score over 27 evaluated variants, which is optimistically biased by construction; the claim that density improves out-of-sample prediction is partly forced by the selection step rather than by an unbiased evaluation of a prespecified model. Coefficients are fit on training data, so this is partial circularity, not full equivalence.

full rationale

The density and asymmetry metrics themselves are not circular: rho (Eq. 1) and alpha (Eq. 2) are defined from abstract embeddings and k previous neighbors, with subsequently published papers explicitly excluded from neighborhood construction, and the outcome (log citations per year) is not an input to these definitions. The self-citation to Imel and Hafen (2023) for the neighborhood-stability filter is not load-bearing for the mathematical derivation, since the filter is described in the text and no uniqueness or validity theorem is imported from that citation. However, the central predictive claim is compromised by an evaluation-protocol circularity: the same 25% held-out test set was used to select the best model among 27 variations and to compute the reported performance gain. The improvement from adding density predictors is thus the maximum of 27 test-set comparisons, not an unbiased out-of-sample estimate, so the headline result is at least partially forced by the selection step. Additional limitations (no human validation of the metric, sample size, missing covariates) are acknowledged by the authors and do not affect the circularity assessment.

Assumptions & free parameters 2 free parameters · 4 assumptions · 2 invented entities

The central claim depends on the validity of embeddings as a proxy for semantic similarity, on the stability filter producing a representative sample, and on the chosen neighborhood size k. These are not externally validated.

free parameters (2)

k (number of nearest neighbors) = 16
Chosen as a trade-off to yield 53,080 converged data points; directly defines the density and asymmetry metrics.
N (number of added publications for convergence) = 1000
Defines the period over which neighborhood stability is checked; a higher or lower value would change which papers are included.

assumptions (4)

domain assumption Text embeddings (SciBERT, SBERT, GPT-2, Word2Vec, BOW) capture meaningful semantic similarity between abstracts.
The entire study relies on this; the paper only sanity-checks field classification, not semantic quality.
domain assumption Cosine distance is a valid measure of semantic dissimilarity.
Standard practice, but not validated against human judgments in this paper.
domain assumption Citation rate is a valid proxy for scientific impact.
Explicitly assumed in Section 4 following prior work.
ad hoc to paper The neighborhood-stability requirement yields a representative sample.
This filter is defined by the authors and is a likely source of selection bias.

invented entities (2)

Density metric ρ
purpose: Quantify the local crowding of a paper's semantic neighborhood based on k nearest previous neighbors.
Defined in Eq. 1; not validated against external density measures or human ratings.
Asymmetry metric α
purpose: Quantify how much a paper lies at the edge of its semantic neighborhood.
Defined in Eq. 2; no external validation; the paper finds it uninformative.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Density, asymmetry and citation dynamics in scientific literature." pith.science (2026). https://pith.science/paper/VOKJ4E35

@misc{pith2026250623366,
  author       = {Pith},
  title        = {Pith review of: Density, asymmetry and citation dynamics in scientific literature},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VOKJ4E35}},
  note         = {Machine review of arXiv:2506.23366}
}

abstract

Scientific behavior is often characterized by a tension between building upon established knowledge and introducing novel ideas. Here, we investigate whether this tension is reflected in the relationship between the similarity of a scientific paper to previous research and its eventual citation rate. To operationalize similarity to previous research, we introduce two complementary metrics to characterize the local geometry of a publication's semantic neighborhood: (1) \emph{density} ($\rho$), defined as the ratio between a fixed number of previously-published papers and the minimum distance enclosing those papers in a semantic embedding space, and (2) asymmetry ($\alpha$), defined as the average directional difference between a paper and its nearest neighbors. We tested the predictive relationship between these two metrics and its subsequent citation rate using a Bayesian hierarchical regression approach, surveying $\sim 53,000$ publications across nine academic disciplines and five different document embeddings. While the individual effects of $\rho$ on citation count are small and variable, incorporating density-based predictors consistently improves out-of-sample prediction when added to baseline models. These results suggest that the density of a paper's surrounding scientific literature may carry modest but informative signals about its eventual impact. Meanwhile, we find no evidence that publication asymmetry improves model predictions of citation rates. Our work provides a scalable framework for linking document embeddings to scientometric outcomes and highlights new questions regarding the role that semantic similarity plays in shaping the dynamics of scientific reward.

Figures

Figures reproduced from arXiv: 2506.23366 by the authors.

**Figure 2.** Visualization of a sample of 9000 scientific publications in semantic space by field using document embeddings obtained from SciBERT [Beltagy et al., 2019]. Each point represents a paper using a document embedding of its abstract that is projected into 2 dimensions via t-SNE dimension reduction. from the embeddings using a simple linear classifier, confirming that the representations captured meaningful semantic str… view at source ↗

**Figure 3.** Estimated effect sizes (βlf ) from our best-performing hierarchical linear regression model, predicting the logarithm of citations per year (scaled within each field) based on reference count, publication age, and our density metric (ρ) calculated for each of five different embedding models, where l denotes the predictor and f denotes the academic field. Including asymmetry (α) estimates as predictors did not impro… view at source ↗

Figures from the paper (4 more)

**Figure 5.** Figure 5: While the asymmetry posterior effect sizes are more consistent across fields (lower variance) [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]

**Figure 4.** Figure 4: Comparison of different Bayesian hierarchical models for predicting citation rates ( [PITH_FULL_IMAGE:figures/full_fig_p012_4.png]

**Figure 5.** Figure 5: Estimated effect sizes (βlf ) from a hierarchical linear regression model that includes both density (ρ) and asymmetry (α) predictors for each of five embedding models, in addition to reference count and publication age. This model extends the main analysis reported in…

**Figure 6.** Figure 6: Visualization of scientific publications in semantic space by field. Each point represents [PITH_FULL_IMAGE:figures/full_fig_p014_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

20 extracted references · 8 canonical work pages

[5]

While the asymmetry posterior effect sizes are more consistent across fields (lower variance) than density effect sizes, we suspect this is because they do not contribute significantly to the model’s performance. 11 base/none (nref,t) per-field (nref,t) per-field, per-model ( ,nref,t) per-field, per-model ( ,nref,t) per-field, per-model ( , ,nref,t) model...

work page 2008
[6]

doi: 10.1038/s42256-023-00735-0

ISSN 2522-5839. doi: 10.1038/s42256-023-00735-0. Yu Zhang, Xiusi Chen, Bowen Jin, Sheng Wang, Shuiwang Ji, Wei Wang, and Jiawei Han. A comprehensive survey of scientific large language models and their applications in scientific discovery,

doi:10.1038/s42256-023-00735-0
[10]

doi: 10.1371/journal.pone.0278389

ISSN 1932-6203. doi: 10.1371/journal.pone.0278389. Daniel Jurafsky and James H. Martin. Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models. 3rd edition,

work page doi:10.1371/journal.pone.0278389 1932
[11]

Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer

Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. Deep Contextualized Word Representations. In Marilyn Walker, Heng Ji, and Amanda Stent, editors, Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol...

work page 2018
[12]

doi: 10.18653/v1/N18-1202

Association for Computational Linguistics. doi: 10.18653/v1/N18-1202. Evelyn Fix and Joseph L. Hodges. Discriminatory analysis. Nonparametric discrimination: Consis- tency properties. Report, USAF School of Aviation Medicine, Randolph Field, Texas,

doi:10.18653/v1/n18-1202
[15]

Citation-Similarity Relationships in Astrophysics Literature

Nathaniel Imel and Zachary Hafen. Citation-Similarity Relationships in Astrophysics Literature. In NeurIPS 2023 AI for Science Workshop,

work page 2023
[16]

SciBERT: A pretrained language model for scientific text

Iz Beltagy, Kyle Lo, and Arman Cohan. SciBERT: A pretrained language model for scientific text. arXiv preprint arXiv:1903.10676,

arXiv 1903
[1962]

I. Lakatos. Falsification and the Methodology of Scientific Research Programmes. In Alan Mus- grave and Imre Lakatos, editors, Criticism and the Growth of Knowledge: Proceedings of the International Colloquium in the Philosophy of Science, London, 1965 , volume 4, pages 91–196. Cambridge University Press, Cambridge,

work page 1965

Show all 20 references

[1967]

doi: 10.1109/TIT.1967.1053964

ISSN 1557-9654. doi: 10.1109/TIT.1967.1053964. George R. Terrell and David W. Scott. Variable Kernel Density Estimation. The Annals of Statistics, 20(3):1236–1265,

1967
[1970]

doi: 10.1017/CBO9781139171434.009

ISBN 978-0-521-09623-2. doi: 10.1017/CBO9781139171434.009. Jacob G. Foster, Andrey Rzhetsky, and James A. Evans. Tradition and Innovation in Scientists’ Research Strategies. American Sociological Review, 80(5):875–908,

doi
[1992]

doi: 10.1214/aos/1176348768

ISSN 0090-5364, 2168-8966. doi: 10.1214/aos/1176348768. Rodney Michael Kinney, Chloe Anastasiades, Russell Authur, Iz Beltagy, Jonathan Bragg, Alexandra Buraczynski, Isabel Cachola, Stefan Candra, Yoganand Chandrasekhar, Arman Cohan, Miles Crawford, Doug Downey, Jason Dunkelbe...
[2008]

A Model comparisons Here we show the metrics across different hierarchical models we tested

doi: 10.1073/pnas.0806977105. A Model comparisons Here we show the metrics across different hierarchical models we tested. It can be seen in Figure 4 that adding asymmetry estimates as predictors did not significantly and consistently improve performance on evaluation metrics....

doi
[2015]

doi: 10.1177/0003122415601618

ISSN 0003-1224. doi: 10.1177/0003122415601618. Brian Uzzi, Satyam Mukherjee, Michael Stringer, and Ben Jones. Atypical combinations and scientific impact. Science (New York, N.Y.), 342(6157):468–472,

doi
[2018]

Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, Anima Anandkumar, Karianne Bergen, Carla P

doi: 10.1126/science.aao0185. Hanchen Wang, Tianfan Fu, Yuanqi Du, Wenhao Gao, Kexin Huang, Ziming Liu, Payal Chandak, Shengchao Liu, Peter Van Katwyk, Andreea Deac, Anima Anandkumar, Karianne Bergen, Carla P. 9 Gomes, Shirley Ho, Pushmeet Kohli, Joan Lasenby, Jure Leskovec, T...

doi
[2020]

doi: 10.18653/v1/2020.acl-main.207

Association for Computational Linguistics. doi: 10.18653/v1/2020.acl-main.207. Laurens van der Maaten and Geoffrey Hinton. Visualizing data using t-sne. Journal of Machine Learning Research, 9(Nov):2579–2605,

2020 doi
[2021]

doi: 10.1007/s11192-020-03759-0

ISSN 1588-2861. doi: 10.1007/s11192-020-03759-0. Filippo Radicchi, Santo Fortunato, and Claudio Castellano. Universality of citation distributions: Toward an objective measure of scientific impact.Proceedings of the National Academy of Sciences, 105(45):17268–17272,

doi
[2022]

doi: 10.1016/j.joi

ISSN 1751-1577. doi: 10.1016/j.joi. 2021.101234. Emmy Liu and Yang Xu. Chaining and the temporal dynamics of scientists’ publishing behaviour. PLOS ONE, 17(12):e0278389,

2021
[2023]

doi: 10.1038/s41586-023-06221-2

ISSN 1476-4687. doi: 10.1038/s41586-023-06221-2. Mario Krenn, Lorenzo Buffoni, Bruno Coutinho, Sagi Eppel, Jacob Gates Foster, Andrew Grit- sevskiy, Harlin Lee, Yichao Lu, João P. Moutinho, Nima Sanjabi, Rishi Sonthalia, Ngoc Mai Tran, Francisco Valente, Yangxinyu Xie, Rose Yu...

doi
[2025]

doi: 10.1038/s42256-025-00994-z

ISSN 2522-5839. doi: 10.1038/s42256-025-00994-z. Microsoft Research AI4Science and Microsoft Azure Quantum. The impact of large language models on scientific discovery: A preliminary study using GPT-4,

doi
[6203]

Yiling Lin, James A

doi: 10.1371/journal.pone.0284567. Yiling Lin, James A. Evans, and Lingfei Wu. New directions in science emerge from disconnection and discord. Journal of Informetrics, 16(1):101234,

doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [5]

While the asymmetry posterior effect sizes are more consistent across fields (lower variance) than density effect sizes, we suspect this is because they do not contribute significantly to the model’s performance. 11 base/none (nref,t) per-field (nref,t) per-field, per-model ( ,nref,t) per-field, per-model ( ,nref,t) per-field, per-model ( , ,nref,t) model...

work page 2008

[2] [6]

doi: 10.1038/s42256-023-00735-0

ISSN 2522-5839. doi: 10.1038/s42256-023-00735-0. Yu Zhang, Xiusi Chen, Bowen Jin, Sheng Wang, Shuiwang Ji, Wei Wang, and Jiawei Han. A comprehensive survey of scientific large language models and their applications in scientific discovery,

doi:10.1038/s42256-023-00735-0

[3] [10]

doi: 10.1371/journal.pone.0278389

ISSN 1932-6203. doi: 10.1371/journal.pone.0278389. Daniel Jurafsky and James H. Martin. Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition with Language Models. 3rd edition,

work page doi:10.1371/journal.pone.0278389 1932

[4] [11]

Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer

Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. Deep Contextualized Word Representations. In Marilyn Walker, Heng Ji, and Amanda Stent, editors, Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Vol...

work page 2018

[5] [12]

doi: 10.18653/v1/N18-1202

Association for Computational Linguistics. doi: 10.18653/v1/N18-1202. Evelyn Fix and Joseph L. Hodges. Discriminatory analysis. Nonparametric discrimination: Consis- tency properties. Report, USAF School of Aviation Medicine, Randolph Field, Texas,

doi:10.18653/v1/n18-1202

[6] [15]

Citation-Similarity Relationships in Astrophysics Literature

Nathaniel Imel and Zachary Hafen. Citation-Similarity Relationships in Astrophysics Literature. In NeurIPS 2023 AI for Science Workshop,

work page 2023

[7] [16]

SciBERT: A pretrained language model for scientific text

Iz Beltagy, Kyle Lo, and Arman Cohan. SciBERT: A pretrained language model for scientific text. arXiv preprint arXiv:1903.10676,

arXiv 1903

[8] [1962]

I. Lakatos. Falsification and the Methodology of Scientific Research Programmes. In Alan Mus- grave and Imre Lakatos, editors, Criticism and the Growth of Knowledge: Proceedings of the International Colloquium in the Philosophy of Science, London, 1965 , volume 4, pages 91–196. Cambridge University Press, Cambridge,

work page 1965