RATE: An LLM-Powered Retrieval Augmented Generation Technology-Extraction Pipeline

Alireza Sheikh; Arya Aftab; Karan Mirhosseini

REVIEW 1 major objections 7 minor 41 references

RATE: An LLM-Powered Retrieval Augmented Generation Technology-Extraction Pipeline

T0 review · 1 major / 7 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read A zero-shot LLM pipeline with retrieval and multi-definition validation extracts technology terms at 91.27% F1.

desk verdict A reproducible zero-shot LLM extraction pipeline whose headline F1 is undermined by likely threshold leakage; worth reviewing, not worth quoting the number. read the letter →

arxiv 2507.21125 v1 pith:43JBVA34 submitted 2025-07-19 cs.IR cs.AIcs.LGcs.SYeess.SY

Karan Mirhosseini , Arya Aftab , Alireza Sheikh This is my paper · ORCID

classification cs.IRcs.AIcs.LGcs.SYeess.SY

keywords technologyextractionlargelanguagemodelsretrievalaugmentedgenerationdefinition-drivenvalidationzero-shotnamedentityrecognitionco-occurrencenetworksbrain-computerinterfaceextendedreality

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

This paper tries to establish that technology extraction from scientific literature can be done accurately without any per-domain training data. It introduces RATE, a pipeline that first asks a retrieval-augmented LLM to propose every plausible technology term in a paper, then sends those candidates through heuristic filters and a second LLM that accepts them only if they satisfy four scholarly definitions of technology. On a 70-paper expert-curated gold standard from the BCI-XR literature, the method reaches 94.26% precision, 88.47% recall, and an F1 score of 91.27%, while a supervised BERT baseline reaches 53.73%. If these results hold, automated technology mapping could shed its dependence on expensively annotated corpora and be refreshed on new literature as it appears.

What carries the argument

The load-bearing mechanism is the definition-validation stage. After the first LLM, helped by retrieved technology definitions from a RAG store, proposes candidate terms from the paper text, heuristic rules (phrase presence, partial compound matches, high-confidence retention, and semantic similarity to the source) strip away obvious hallucinations. A second LLM then evaluates each surviving candidate against four published definitions of technology, using explicit 'YES if / NO if' decision criteria and a 1–10 confidence scale, and retains a term only when the verdict is positive and the confidence exceeds 6. The four definitions are the anchor: they turn the contested question of what counts as a technology into a repeatable, contextual scoring rule.

What would settle it

Apply RATE unchanged to a held-out set of 70 papers from a different domain, with a gold standard produced by experts who never saw RATE's four definitions. If the F1 score falls well below 91.27%, or if the gap shows up mostly as false positives, the definition loop itself is doing the work rather than a general technology-extraction ability.

Watch

Extended reading notes

Core claim

The paper's central claim is that a fully automated zero-shot pipeline can pull concrete technology terms out of scientific titles, abstracts, and author keywords nearly as reliably as domain experts. The claim rests on a concrete comparison: on 70 randomly selected papers, RATE's extracted list agrees with the expert gold standard at 91.27% F1, against 53.73% for a fine-tuned BERT named-entity-recognition baseline. The authors argue this shows their definition-driven, two-LLM design—where one LLM generates broadly and a second LLM validates each term against explicit scholarly criteria for what counts as technology—solves the precision problem that earlier automated methods faced from the lack of annotated technology datasets.

Load-bearing premise

The gold standard was produced by three experts using exactly the four technology definitions that RATE's validator is fed, and the pipeline's thresholds were not justified by a separate tuning set, so the reported agreement could partly reflect shared criteria or threshold fitting rather than a truly general extraction capability.

Editorial extensions

If this is right

Technology mapping no longer requires annotated training data per field; a new domain needs only a search query, a corpus, and a curated list of technology definitions for retrieval.
The same definitional-validation architecture can be pointed at other contested concepts, such as 'innovation' or 'emerging technology', by replacing the four definitions and the YES/NO rubric.
The co-occurrence network built from RATE's extractions separates the BCI-XR literature into distinct communities, including a VR-EEG rehabilitation cluster and an AR-SSVEP portable-control cluster, which could guide targeted reviews.
Because the candidate-generation step runs at temperature 0, each extraction is reproducible and every retained term can be traced back to its source text and its validation reasoning.

Reading between the lines

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

The expert gold standard was built using the same four definitions that RATE's validator uses, so the reported F1 may partly measure agreement with a rubric rather than agreement with an independent human notion of 'technology'; a more direct test would have one team of experts label freely and another apply RATE.
The many thresholds in the pipeline (0.7, 0.75, 0.95, 0.70, confidence above 6) are reported without a separate validation set, so a sensitivity analysis on held-out papers is needed to see whether the high F1 is robust or tuned.
The paper never isolates the marginal contribution of retrieval; ablating the RAG context (or feeding unrelated retrieved documents) would show whether the candidates are genuinely grounded in the retrieved definitions or whether the first LLM would perform as well alone.
A cross-domain transfer test—running RATE unchanged on a non-neural field such as clean energy or manufacturing, with an independently created gold standard—would directly test the claimed generalizability beyond BCI-XR.

Share X Bluesky LinkedIn Reddit HN

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

Short version: this is a decent engineering paper with a headline number I wouldn't quote. The RATE pipeline is a sensible combination of RAG, two LLM passes, and rule-based post-processing, and the authors ship code. The multi-definition validation stage is a genuinely new twist for technology extraction. But the 91.27% F1 is not established as a generalizable zero-shot result, because the five thresholds (0.7, 0.75, 0.95, 0.70, confidence >6) appear to be set without a held-out validation set, and the authors admit in Section II-G that outputs are highly sensitive to prompt design. That's textbook threshold leakage. The gold standard also used the same four definitions the LLM validator is given, so part of the agreement is measuring definitional alignment rather than independent extraction.

What the paper does well: the pipeline is described in enough detail to re-implement, there is source code, and the comparison with a fine-tuned BERT NER baseline, though weak, is at least a real baseline. The co-occurrence network analysis is standard bibliometrics but executed carefully enough to be reproducible. For someone building technology maps, this pipeline is a useful starting point.

The soft spots are mostly in the evaluation. There are no confidence intervals or significance tests, the BERT baseline is trained on a tiny private dataset and is not SOTA, and the network mapping includes some odd terms that suggest a few false positives slip through the validation stage. The threshold-leakage problem is the most concrete issue and it's fixable: hold out part of the 70 papers, set thresholds on the rest, then evaluate on the hold-out. A sensitivity analysis over the thresholds would also help. None of this is fatal to the idea, but it means the reported F1 is an optimistic in-sample estimate.

Who should read it: bibliometrics and foresight researchers who want an off-the-shelf zero-shot extractor and are comfortable treating the performance numbers with caution. It deserves a serious referee: the method is new enough, the code is available, and the evaluation can be tightened. I'd send it to review, with a request for a threshold-selection protocol and a stronger baseline.

Referee Report

1 major / 7 minor

Summary. The paper introduces RATE (Retrieval Augmented Technology Extraction), a multi-stage LLM pipeline for extracting technology terms from scientific literature. The pipeline combines a DeepSeek-V3 LLM with RAG for candidate generation, a heuristic validation stage, a second LLM-based definitional validation stage using four scholarly definitions of technology, and post-processing. The authors apply RATE to 678 publications on BCI-XR, evaluate on a gold standard of 70 papers curated by three experts, and report an F1-score of 91.27%, compared with a BERT baseline at 53.73%. The extracted terms are then used to construct a co-occurrence network and community analysis of the BCI-XR technology landscape. The code is publicly available.

Significance. If the reported performance holds, the paper makes a useful contribution by demonstrating that a zero-shot, definition-driven LLM pipeline with RAG can extract technology entities from scientific text at a level approaching human agreement, without task-specific training data. The explicit multi-definition validation step is a sensible response to the ambiguity of the term 'technology' and could be transferable to other entity-extraction tasks. The authors also provide the pipeline code, which supports reproducibility. However, the evaluation as presented does not yet establish the central F1 claim at the level of confidence implied by the paper: the hand-set thresholds lack a validation protocol, the gold standard shares the same definitions fed to the LLM, the BERT baseline is weak, and no uncertainty quantification is reported. These are fixable within the scope of a revision.

major comments (1)

[II-C1 and IV-D] The paper names the pipeline 'Retrieval Augmented Technology Extraction' and claims RAG provides 'domain understanding and term disambiguation,' but it presents no ablation isolating the effect of the RAG component. For a methodological contribution centered on RAG, the reader cannot tell whether RAG materially improves candidate generation or whether the results are driven mainly by the prompt and validation stages. An ablation that removes the retrieval context (or replaces it with static text) would substantiate the role of RAG in the reported performance.

minor comments (7)

[Abstract] There is a capitalization error: 'Consequently, The validated technology terms' should be 'Consequently, the validated technology terms'.
[II-F] The network statistics table reports 'Clustering co-efficient'; correct the spelling to 'coefficient'.
[II-F] The phrase 'Nodes with low edge wight' contains a typo; 'wight' should be 'weight'.
[II-E] The BERT baseline description is vague: 'approximately 2200 rows of tokenized scientific sentences' should specify whether rows are sentences or tokens, and the inter-annotator agreement on the BIO tags is not reported.
[II-C4] The justification for the confidence cutoff 'greater than six' is not given, while the candidate-generation threshold is 0.7 on a different scale; clarify how the 10-point confidence scale is calibrated.
[III-C and Figure 3] There is an unexplained large block of lowercase technology terms immediately before Figure 3; this appears to be either a figure caption or an artifact of the manuscript, and it should be formatted properly.
[II-C2 and II-C4] Both LLM stages use DeepSeek-V3 with temperature 0.0 and max tokens 4096, but the paper does not mention the exact retrieval settings for the RAG stage (e.g., similarity metric, chunk overlap). Adding these details would improve reproducibility.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: RATE's predictions are not constructed from the gold labels; shared definitions measure definitional agreement, and threshold tuning is a validity risk, not a circular step.

full rationale

The derivation chain is self-contained. RATE's output is a function of input text, a fixed prompt, and stated threshold constants; the gold standard was created by three experts independently labeling 70 papers using technology definitions that are also supplied to the LLM validator (Sections II-C4 and III-A). That alignment makes the F1 an agreement measure between the pipeline and human application of the same definition; it does not make the labels an input to the pipeline, and no equation in the paper defines the predicted terms in terms of the gold standard. The five thresholds (0.7, 0.75, 0.95, 0.70, and >6) are presented as fixed pipeline parameters, and the paper contains no description of fitting them to the 70 evaluation papers, so the absence of a separate validation split is a generalizability or overfitting risk rather than circularity, and it cannot be scored as a circular step without speculation about how the thresholds were chosen. There are no author self-citations, no imported uniqueness theorems, and no ansatz smuggled in by citation; the comparison to BERT uses an independently trained baseline. The co-occurrence network and its statistics follow directly from the extracted term lists and are not used to define the extraction targets.

Assumptions & free parameters 6 free parameters · 4 assumptions · 0 invented entities

The pipeline relies on hand-set thresholds and design choices, on the contested notion of 'technology' as encoded in four external definitions, and on the expert gold standard as ground truth. No new physical or conceptual entities are introduced; the invented-entity count is zero. The main concern is that thresholds and evaluation definitions are tightly coupled to the same source, which raises the circularity burden.

free parameters (6)

candidate confidence threshold = 0.7
Terms with LLM confidence below 0.7 are dropped in candidate generation (Section II-C2); no tuning procedure or separate validation is reported.
partial compound match ratio = 0.75
Multi-word candidates accepted if at least 75% of meaningful words appear in source text (Section II-C3); hand-set.
high-confidence retention threshold = 0.95
Candidates with confidence at least 0.95 are kept without textual match (Section II-C3).
semantic similarity threshold = 0.70
spaCy similarity to full text must be at least 0.70 for moderate-confidence candidates (Section II-C3).
definitional validation confidence cutoff = >6
Only candidates with LLM boolean true and confidence greater than 6 are confirmed (Section II-C4).
number of RAG documents = top 20 retrieved, target 7 after diversity filter
Retriever fetches 20 nearest chunks and keeps up to 7 diverse documents (Section II-C1); affects context but not the core F1.

assumptions (4)

domain assumption The four definitions of technology from Puccetti et al. are an adequate ground truth for deciding what is a technology.
Both the gold standard experts and the LLM validation stage use these definitions (Sections II-D and II-C4), so the evaluation concept is defined by the same external source.
domain assumption The three domain experts' gold standard annotations are correct and complete.
F1 is computed against this list; inter-annotator agreement or boundary cases are not reported (Section II-D).
domain assumption The RAG knowledge base built from Wikipedia, CSET, IEA, and O*NET provides sufficient context for disambiguation without leaking candidate terms.
RAG documents are said to serve only for background, but their coverage and any effect on extraction are not measured (Section II-C1).
domain assumption The Web of Science query captures the BCI-XR intersection without major gaps.
Corpus of 678 articles is assumed to represent the field; no validation against other databases (Section II-B).

how reviews work

0 comments

Cite this review

Pith. "Pith review of RATE: An LLM-Powered Retrieval Augmented Generation Technology-Extraction Pipeline." pith.science (2026). https://pith.science/paper/43JBVA34

@misc{pith2026250721125,
  author       = {Pith},
  title        = {Pith review of: RATE: An LLM-Powered Retrieval Augmented Generation Technology-Extraction Pipeline},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/43JBVA34}},
  note         = {Machine review of arXiv:2507.21125}
}

read the original abstract

In an era of radical technology transformations, technology maps play a crucial role in enhancing decision making. These maps heavily rely on automated methods of technology extraction. This paper introduces Retrieval Augmented Technology Extraction (RATE), a Large Language Model (LLM) based pipeline for automated technology extraction from scientific literature. RATE combines Retrieval Augmented Generation (RAG) with multi-definition LLM-based validation. This hybrid method results in high recall in candidate generation alongside with high precision in candidate filtering. While the pipeline is designed to be general and widely applicable, we demonstrate its use on 678 research articles focused on Brain-Computer Interfaces (BCIs) and Extended Reality (XR) as a case study. Consequently, The validated technology terms by RATE were mapped into a co-occurrence network, revealing thematic clusters and structural features of the research landscape. For the purpose of evaluation, a gold standard dataset of technologies in 70 selected random articles had been curated by the experts. In addition, a technology extraction model based on Bidirectional Encoder Representations of Transformers (BERT) was used as a comparative method. RATE achieved F1-score of 91.27%, Significantly outperforming BERT with F1-score of 53.73%. Our findings highlight the promise of definition-driven LLM methods for technology extraction and mapping. They also offer new insights into emerging trends within the BCI-XR field. The source code is available https://github.com/AryaAftab/RATE

Figures

Figures reproduced from arXiv: 2507.21125 by the authors.

**Figure 1.** An overview of technology extraction and validation pipeline The search string for BCI, based on the work of Yin et al. [21], was: TS= (”brain-computer interface*” OR ”brainmachine interface*” OR ”brain machine interface*” OR ”brain computer interface*” OR ”direct neural interface*”) Therefore, the final combined search string used in WoS was: TS= ((”virtual realit*” OR ”augmented realit*” OR ”mixed realit*” OR ”co… view at source ↗

**Figure 2.** Representation of Girvan-Newman communities in the network representation of co-occurring technologies was then created using VOSviewer. This tool was employed for its capabilities in constructing and visualizing bibliometric maps based on co-occurrence data. IV. DISCUSSION This study introduced RATE, a novel multi-stage LLMbased pipeline for technology extraction, and applied it to map the technological landscape… view at source ↗

**Figure 4.** Representation of technologies co-occurrence network with VOSviewer BCIs, heart-rate monitoring, and broader brain wave detection methods. This suggests a research thrust focused on creating richer, multi-sensory immersive experiences where BCI inputs are potentially augmented or correlated with other physiological signals to enhance interaction or user state assessment. C. Broader Network Insights and Emerging Tre… view at source ↗

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

41 extracted references · 31 canonical work pages

[1]

What is xr? towards a framework for augmented and virtual reality,

P. A. Rauschnabel, R. Felix, C. Hinsch, H. Shahab, and F. Alt, “What is xr? towards a framework for augmented and virtual reality,” Computers in human behavior , vol. 133, p. 107289, 2022

work page 2022
[2]

Augmented reality: A comprehensive review,

S. Dargan, S. Bansal, M. Kumar, A. Mittal, and K. Kumar, “Augmented reality: A comprehensive review,” Archives of Computational Methods in Engineering, vol. 30, no. 2, pp. 1057–1080, 2023

work page 2023
[3]

A review on virtual reality and augmented reality use-cases of brain computer interface based applications for smart cities,

V . Kohli, U. Tripathi, V . Chamola, B. K. Rout, and S. S. Kanhere, “A review on virtual reality and augmented reality use-cases of brain computer interface based applications for smart cities,” Microprocessors and Microsystems, vol. 88, p. 104392, 2022

work page 2022
[4]

Virtual and augmented reality applications in medicine: analysis of the scientific literature,

A. W. K. Yeung, A. Tosevska, E. Klager, F. Eibensteiner, D. Laxar, J. Stoyanov, M. Glisic, S. Zeiner, S. T. Kulnik, R. Crutzen et al. , “Virtual and augmented reality applications in medicine: analysis of the scientific literature,” Journal of medical internet research, vol. 23, no. 2, p. e25499, 2021

work page 2021
[5]

Taxonomy of virtual and augmented reality applications in education,

J. Motejlek and E. Alpay, “Taxonomy of virtual and augmented reality applications in education,” IEEE transactions on learning technologies , vol. 14, no. 3, pp. 415–429, 2021

work page 2021
[6]

Augmented reality applications in industry 4.0 environment,

V . Relji ´c, I. Milenkovi ´c, S. Dudi ´c, J. ˇSulc, and B. Baj ˇci, “Augmented reality applications in industry 4.0 environment,” Applied Sciences , vol. 11, no. 12, p. 5592, 2021

work page 2021
[7]

Brain-computer interfaces and aug- mented/virtual reality,

F. Putze, A. V ourvopoulos, A. L ´ecuyer, D. Krusienski, S. Berm ´udez i Badia, T. Mullen, and C. Herff, “Brain-computer interfaces and aug- mented/virtual reality,” p. 144, 2020

work page 2020
[8]

Brain computer interfaces, a review,

L. F. Nicolas-Alonso and J. Gomez-Gil, “Brain computer interfaces, a review,” sensors, vol. 12, no. 2, pp. 1211–1279, 2012

work page 2012

Show all 41 references

[9]

Brain-computer interfaces and augmented reality: A state of the art,

H. Si-Mohammed, F. A. Sanz, G. Casiez, N. Roussel, and A. L ´ecuyer, “Brain-computer interfaces and augmented reality: A state of the art,” in Graz Brain-Computer Interface Conference , 2017

2017
[10]

Impact of virtual reality on brain–computer interface performance in iot control—review of current state of knowledge,

A. Piszcz, I. Rojek, and D. Mikołajewski, “Impact of virtual reality on brain–computer interface performance in iot control—review of current state of knowledge,” Applied Sciences, vol. 14, no. 22, p. 10541, 2024

2024
[11]

Mapping the evolution of scientific fields based on cross-field authors,

X. Sun, K. Ding, and Y . Lin, “Mapping the evolution of scientific fields based on cross-field authors,” Journal of Informetrics, vol. 10, no. 3, pp. 750–761, 2016

2016
[12]

Use of technology mapping in identification of fuel cell sub-technologies,

M. R. Arasti and N. B. Moghaddam, “Use of technology mapping in identification of fuel cell sub-technologies,” international journal of hydrogen energy, vol. 35, no. 17, pp. 9516–9525, 2010

2010
[13]

Technology mapping, business strategy, and market opportunities,

P. E. Castells, M. R. Salvador, and R. M. Bosch, “Technology mapping, business strategy, and market opportunities,” Competitive Intelligence Review: Published in Cooperation with the Society of Competitive Intelligence Professionals, vol. 11, no. 1, pp. 46–57, 2000

2000
[14]

Keyword ex- traction: Issues and methods,

N. Firoozeh, A. Nazarenko, F. Alizon, and B. Daille, “Keyword ex- traction: Issues and methods,” Natural Language Engineering , vol. 26, no. 3, pp. 259–291, 2020

2020
[15]

Keyword and keyphrase extraction tech- niques: a literature review,

S. Siddiqi and A. Sharan, “Keyword and keyphrase extraction tech- niques: a literature review,” International Journal of Computer Applica- tions, vol. 109, no. 2, 2015

2015
[16]

Technology identification from patent texts: A novel named entity recognition method,

G. Puccetti, V . Giordano, I. Spada, F. Chiarello, and G. Fantoni, “Technology identification from patent texts: A novel named entity recognition method,” Technological Forecasting and Social Change, vol. 186, p. 122160, 2023

2023
[17]

Large language models: A survey,

S. Minaee, T. Mikolov, N. Nikzad, M. Chenaghlu, R. Socher, X. Amatriain, and J. Gao, “Large language models: A survey,” 2025. [Online]. Available: https://arxiv.org/abs/2402.06196

2025 arXiv
[18]

A survey of large language models,

W. X. Zhao, K. Zhou, J. Li, T. Tang, X. Wang, Y . Hou, Y . Min, B. Zhang, J. Zhang, Z. Dong et al. , “A survey of large language models,” arXiv preprint arXiv:2303.18223, vol. 1, no. 2, 2023

2023 arXiv
[19]

Large (vision) language models are unsupervised in-context learners,

A. Gadetsky, A. Atanov, Y . Jiang, Z. Gao, G. H. Mighan, A. Zamir, and M. Brbic, “Large (vision) language models are unsupervised in-context learners,” arXiv preprint arXiv:2504.02349 , 2025

2025 arXiv
[20]

Retrieval-augmented generation for large language models: A survey,

Y . Gao, Y . Xiong, X. Gao, K. Jia, J. Pan, Y . Bi, Y . Dai, J. Sun, H. Wang, and H. Wang, “Retrieval-augmented generation for large language models: A survey,” arXiv preprint arXiv:2312.10997 , vol. 2, no. 1, 2023

2023 arXiv
[21]

Bibliometric analysis on brain-computer interfaces in a 30-year period,

Z. Yin, Y . Wan, H. Fang, L. Li, T. Wang, Z. Wang, and D. Tan, “Bibliometric analysis on brain-computer interfaces in a 30-year period,” Applied Intelligence, vol. 53, no. 12, pp. 16 205–16 225, 2023

2023
[22]

Chinese catalogue of technologies prohibited or restricted from export,

Center for Security and Emerging Technology, “Chinese catalogue of technologies prohibited or restricted from export,” https://cset. georgetown.edu/publication/china-export-control-catalog-2023, Dec. 2023, (Center for Security and Emerging Technology, Trans.). Original work pu...

2023
[23]

Etp clean energy tech- nology guide,

International Energy Agency, “Etp clean energy tech- nology guide,” https://www.iea.org/data-and-statistics/data-tools/ etp-clean-energy-technology-guide, Apr. 2025, (Online). [On- line]. Available: https://www.iea.org/data-and-statistics/data-tools/ etp-clean-energy-technology-guide

2025
[24]

O*net online,

National Center for O*NET Development, “O*net online,” https: //www.onetonline.org/, 2024, u.S. Department of Labor, Employment & Training Administration. [Online]. Available: https://www.onetonline. org/

2024
[25]

Langchain (version 0.3.12) [computer software],

H. Chase, “Langchain (version 0.3.12) [computer software],” https://github.com/langchain-ai/langchain, 2024, accessed: 2024-05-31. [Online]. Available: https://github.com/langchain-ai/langchain

2024
[26]

Ollama (version 0.7.0) [computer software],

Ollama, “Ollama (version 0.7.0) [computer software],” https://ollama. com/, 2025, accessed: 2025-05-31. [Online]. Available: https://ollama. com/

2025
[27]

S. Lee, A. Shakir, D. Koenig, and J. Lipp. (2024) Open source strikes bread - new fluffy embedding model. [Online]. Available: https://www.mixedbread.ai/blog/mxbai-embed-large-v1

2024
[28]

Deepseek-v3 technical report,

A. Liu, B. Feng, B. Xue, B. Wang, B. Wu, C. Lu, C. Zhao, C. Deng, C. Zhang, C. Ruan et al., “Deepseek-v3 technical report,” arXiv preprint arXiv:2412.19437, 2024

2024 arXiv
[29]

Bert: Pre-training of deep bidirectional transformers for language understanding,

J. Devlin, M.-W. Chang, K. Lee, and K. Toutanova, “Bert: Pre-training of deep bidirectional transformers for language understanding,” in Pro- ceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technolo...

2019
[30]

Ner-bert: A pre-trained model for low-resource entity tagging,

Z. Liu, F. Jiang, Y . Hu, C. Shi, and P. Fung, “Ner-bert: A pre-trained model for low-resource entity tagging,” arXiv preprint arXiv:2112.00405, 2021

2021 arXiv
[31]

Huggingface’s transformers: State-of-the-art natural language processing,

T. Wolf, L. Debut, V . Sanh, J. Chaumond, C. Delangue, A. Moi, P. Cistac, T. Rault, R. Louf, M. Funtowicz, J. Davison, S. Shleifer, P. von Platen, C. Ma, Y . Jernite, J. Plu, C. Xu, T. L. Scao, S. Gugger, M. Drame, Q. Lhoest, and A. M. Rush, “Huggingface’s transformers: State-...

2020 arXiv
[32]

pandas-dev/pandas: Pandas,

T. pandas development team, “pandas-dev/pandas: Pandas,” Feb. 2020. [Online]. Available: https://doi.org/10.5281/zenodo.3509134

2020 doi
[33]

Exploring network structure, dynamics, and function using networkx,

A. Hagberg, P. J. Swart, and D. A. Schult, “Exploring network structure, dynamics, and function using networkx,” Los Alamos National Labora- tory (LANL), Los Alamos, NM (United States), Tech. Rep., 2008

2008
[34]

Fast unfolding of communities in large networks,

V . D. Blondel, J.-L. Guillaume, R. Lambiotte, and E. Lefebvre, “Fast unfolding of communities in large networks,” Journal of statistical mechanics: theory and experiment , vol. 2008, no. 10, p. P10008, 2008

2008
[35]

Finding and evaluating community structure in networks,

M. E. Newman and M. Girvan, “Finding and evaluating community structure in networks,” Physical review E , vol. 69, no. 2, p. 026113, 2004

2004
[36]

Gephi: an open source software for exploring and manipulating networks,

M. Bastian, S. Heymann, and M. Jacomy, “Gephi: an open source software for exploring and manipulating networks,” in Proceedings of the international AAAI conference on web and social media , vol. 3, no. 1, 2009, pp. 361–362

2009
[37]

Software survey: V osviewer, a computer program for bibliometric mapping,

N. Van Eck and L. Waltman, “Software survey: V osviewer, a computer program for bibliometric mapping,” scientometrics, vol. 84, no. 2, pp. 523–538, 2009

2009
[38]

Understanding and mitigating the bias inheritance in llm-based data augmentation on downstream tasks,

M. Li, H. Chen, Y . Wang, T. Zhu, W. Zhang, K. Zhu, K.-F. Wong, and J. Wang, “Understanding and mitigating the bias inheritance in llm-based data augmentation on downstream tasks,” arXiv preprint arXiv:2502.04419, 2025

2025 arXiv
[39]

Inherent bias in large language models: A random sampling analysis,

N. F. Ayoub, K. Balakrishnan, M. S. Ayoub, T. F. Barrett, A. P. David, and S. T. Gray, “Inherent bias in large language models: A random sampling analysis,” Mayo Clinic Proceedings: Digital Health , vol. 2, no. 2, pp. 186–191, 2024

2024
[40]

Web of science (wos) and scopus: The titans of bibli- ographic information in today’s academic world,

R. Pranckut ˙e, “Web of science (wos) and scopus: The titans of bibli- ographic information in today’s academic world,” Publications, vol. 9, no. 1, p. 12, 2021

2021
[41]

J. Agar, “What is technology? technology: critical history of a concept, by eric schatzberg, chicago and london, university of chicago press, 2018, 352 pp., $27.45 (paperback), isbn: 978-0-226-58383-9,” Annals of Science, vol. 77, no. 3, pp. 377–382, 2020

2018

Pith tools

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

[1] [1]

What is xr? towards a framework for augmented and virtual reality,

P. A. Rauschnabel, R. Felix, C. Hinsch, H. Shahab, and F. Alt, “What is xr? towards a framework for augmented and virtual reality,” Computers in human behavior , vol. 133, p. 107289, 2022

work page 2022

[2] [2]

Augmented reality: A comprehensive review,

S. Dargan, S. Bansal, M. Kumar, A. Mittal, and K. Kumar, “Augmented reality: A comprehensive review,” Archives of Computational Methods in Engineering, vol. 30, no. 2, pp. 1057–1080, 2023

work page 2023

[3] [3]

A review on virtual reality and augmented reality use-cases of brain computer interface based applications for smart cities,

V . Kohli, U. Tripathi, V . Chamola, B. K. Rout, and S. S. Kanhere, “A review on virtual reality and augmented reality use-cases of brain computer interface based applications for smart cities,” Microprocessors and Microsystems, vol. 88, p. 104392, 2022

work page 2022

[4] [4]

Virtual and augmented reality applications in medicine: analysis of the scientific literature,

A. W. K. Yeung, A. Tosevska, E. Klager, F. Eibensteiner, D. Laxar, J. Stoyanov, M. Glisic, S. Zeiner, S. T. Kulnik, R. Crutzen et al. , “Virtual and augmented reality applications in medicine: analysis of the scientific literature,” Journal of medical internet research, vol. 23, no. 2, p. e25499, 2021

work page 2021

[5] [5]

Taxonomy of virtual and augmented reality applications in education,

J. Motejlek and E. Alpay, “Taxonomy of virtual and augmented reality applications in education,” IEEE transactions on learning technologies , vol. 14, no. 3, pp. 415–429, 2021

work page 2021

[6] [6]

Augmented reality applications in industry 4.0 environment,

V . Relji ´c, I. Milenkovi ´c, S. Dudi ´c, J. ˇSulc, and B. Baj ˇci, “Augmented reality applications in industry 4.0 environment,” Applied Sciences , vol. 11, no. 12, p. 5592, 2021

work page 2021

[7] [7]

Brain-computer interfaces and aug- mented/virtual reality,

F. Putze, A. V ourvopoulos, A. L ´ecuyer, D. Krusienski, S. Berm ´udez i Badia, T. Mullen, and C. Herff, “Brain-computer interfaces and aug- mented/virtual reality,” p. 144, 2020

work page 2020

[8] [8]

Brain computer interfaces, a review,

L. F. Nicolas-Alonso and J. Gomez-Gil, “Brain computer interfaces, a review,” sensors, vol. 12, no. 2, pp. 1211–1279, 2012

work page 2012