Pith. sign in

REVIEW 4 major objections 38 references

Routing new papers into a personal library works best when an agent reads what each folder already holds, not just its name.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review

2026-07-14 04:43 UTC pith:ET7FRCIC

load-bearing objection Clean formalization of a real personal-library routing problem plus a training-free agent that actually reads folder members; the user-study gains are real but partly softened by post-hoc gold labeling. the 4 major comments →

arxiv 2607.11564 v1 pith:ET7FRCIC submitted 2026-07-13 cs.CL cs.HCcs.IR

PaperRouter-Agent: A Content-Grounded LLM Agent for Personalized Hierarchical Paper Routing

Keshen Zhou , Lintao Wang , Suqin Yuan , Zhuqiang Lu , Yu Luo , Zhiyong Wang This is my paper
classification cs.CL cs.HCcs.IR
keywords personalized hierarchical paper routingfolksonomycontent-grounded inspectionLLM agentsretrieval-augmented generationpersonal knowledge managementhierarchical text classification
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Personal reference libraries are not shared taxonomies: folders may be topical, shorthand, venue-based, or process-oriented, and their meaning is often set by the papers already inside them. The paper formalizes this as personalized hierarchical paper routing (PHPR)—assigning an incoming paper to folders in a user’s own evolving hierarchy without any per-user training. It introduces PaperRouter-Agent, a training-free four-stage LLM agent that first narrows candidates top-down, then chooses evidence by folder type, verifies fit by reading member papers, and injects similar past rejections. On five researchers’ real Zotero libraries the agent lifts overall Recall@1 from 0.39 to 0.61 and Recall@3 from 0.57 to 0.83, with the largest gains on venue and year folders that name-only methods almost never hit. The same content-grounded inspection also raises accuracy on the public LaMP-2 tagging benchmark from 44.5 % to 51.5 % while staying practical in cost.

Core claim

When folder meaning is defined by its members rather than its label, a training-free LLM agent that inspects those members (and routes metadata folders by metadata) substantially outperforms name-only single-shot classification on real personal libraries and on a personalized tagging proxy, recovering the organizational folders that name matching collapses on.

What carries the argument

PaperRouter-Agent: a four-stage, training-free pipeline (Planner for top-down candidate narrowing, Retriever for type-aware evidence selection, Inspector for content-grounded Fit/No-Fit/Abstain verification plus cross-comparison, Reflector for similarity-gated reuse of past user rejections).

Load-bearing premise

That results from a two-round study with only five research students and twenty papers each, with gold labels collected after the system has already proposed folders, generalize to the wider range of personal libraries and folder styles.

What would settle it

A larger multi-user study in which content-grounded inspection fails to improve Recall@1 over a strong name-based baseline on venue/year or shorthand folders, or where steady-state latency and call counts exceed the single-shot baseline without recovering those folder types.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Reference managers can automate folder placement by reading existing members instead of relying on folder names.
  • Organizational folders defined by venue, year, or status become recoverable once the agent routes them by metadata rather than content alone.
  • Online accept/reject memory can refine later decisions without any per-user fine-tuning.
  • The same content-grounded inspection transfers to other personalized tagging settings that lack a shared taxonomy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Any personal-organization task whose labels are folksonomic (bookmarks, notes, media libraries) is likely to benefit from the same member-first inspection pattern.
  • Process/status folders remain hard because membership encodes user intent rather than content or metadata; solving them may require an explicit intent model.
  • Building an open multi-user PHPR benchmark would let long-term Reflector gains be measured properly, which the single-session study cannot yet do.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 0 minor

Summary. The paper formalizes personalized hierarchical paper routing (PHPR) as the task of assigning an incoming paper to folders in a user-specific, evolving folksonomy (rather than a fixed shared taxonomy), without per-user training. It proposes PaperRouter-Agent, a training-free four-stage LLM agent (Planner, Retriever, Inspector, Reflector) that narrows candidates top-down, types folders for type-aware evidence, verifies fit by inspecting sampled member papers (with abstention), and injects similarity-gated past rejections. On a two-round within-subjects study of five researchers’ real Zotero libraries (100 papers total), it reports Recall@1 rising from 0.39 to 0.61 and Recall@3 from 0.57 to 0.83 versus a single-shot name-based baseline, with largest gains on venue/year folders (0.09→0.50). On LaMP-2 it lifts accuracy from 44.5% to 51.5% (+9.0 macro-F1) by content-grounded inspection alone, with a cost/latency analysis showing practical steady-state overhead.

Significance. If the gains hold, the work usefully carves out PHPR as distinct from classical hierarchical text classification and from flat personalized tagging, and shows that content-grounded inspection of folder members (plus metadata routing for organizational folders) recovers failures that pure name matching cannot. Strengths include a clear problem formalization (Table 1), a deployed Zotero integration used in the study, explicit per-type breakdowns, cost/latency measurements against single-shot and inspect-all, and a public-benchmark result that matches a retrieval-augmented LLM without a separate retriever or fine-tuning. The training-free, online-feedback design is practically relevant for personal knowledge management tools.

major comments (4)
  1. Section 5.2 and Appendix B (Table 5): the gold set R⋆ is constructed after participants see the system’s ranked candidates (accept correct returned folders and name omitted correct ones). Because the same papers are run first under the single-shot baseline then under the agent, and because the agent surfaces metadata folders the baseline rarely proposes, ratification bias and order/learning effects can inflate the reported Recall@1/3 lifts (overall 0.39→0.61, venue/year 0.09→0.50). An independent pre-system gold, counterbalanced condition order, or a blinded second labeling pass is needed to secure the magnitude of the central claim.
  2. Section 5.2: n=5 participants × 20 papers is too small and homogeneous (research students) to underwrite generalization across library sizes, depths, and folder-type mixtures in the broader population. Per-type cells are especially thin (process/status n=12, venue/year n=18). Either enlarge the study or substantially strengthen the limitations discussion and confidence intervals; the current formative framing understates how load-bearing these numbers are for the abstract and introduction claims.
  3. Section 5.3 and Discussion: LaMP-2 evaluates only the Inspector on a flat, fixed-tag proxy and does not exercise Planner, Retriever, or Reflector, nor hierarchical multi-type routing. The Reflector’s online feedback mechanism is described but never longitudinally measured (explicitly left to future work). The paper’s strongest claim is hierarchical multi-type gains on real libraries; the public result cannot substitute for that evaluation. Either add a longitudinal or leave-one-out feedback experiment, or clearly demote the Reflector from a core contribution to a designed-but-untested component.
  4. Section 4.3–4.4 and Table 3: process/status folders remain a residual hard case (Recall@1 0.12→0.44) and the Retriever’s binary topical-vs-organizational typing leaves them without a principled evidence route. Given that the paper advertises four folder types, the agent’s handling of process/status needs either a concrete mechanism (e.g., user-intent cues or explicit abstention policy) or a clearer statement that these folders are out of scope for content grounding.

Circularity Check

0 steps flagged

Empirical systems paper with no derivation chain that reduces to its inputs; claims are measured against user gold and a public benchmark.

full rationale

PaperRouter-Agent is a training-free multi-stage LLM agent evaluated with standard Recall@K on five users' Zotero libraries and accuracy/macro-F1 on the public LaMP-2 benchmark. There is no mathematical derivation, fitted parameter renamed as a prediction, uniqueness theorem imported from the authors, or ansatz smuggled via self-citation. The agent does not train on the evaluation libraries; folder typing and member sampling are operational design choices, not circular definitions of the reported metrics. Self-citations in Related Work are ordinary positioning, not load-bearing premises that force the empirical gains. Methodological concerns about post-hoc gold labeling (participants naming correct folders after seeing candidates) affect evaluation validity, not circularity of a derivation. Score 0 is therefore appropriate: the paper is self-contained against external labels and contains no circular reduction.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 2 invented entities

The work is an empirical LLM-agent systems paper. It inherits standard assumptions about LLM zero-shot reasoning and retrieval, introduces the PHPR task definition and the four-stage agent as its main constructs, and fits no continuous free parameters to data. The only tunable knobs are discrete hyperparameters (sample size N=5, temperature 0, etc.).

free parameters (2)
  • members sampled per folder (N) = 5
    Fixed at 5 (first-N deterministic); controls how much content the Inspector sees.
  • Planner top-k branches kept = 5
    Default 5; discrete design choice that trades recall of candidates against cost.
axioms (3)
  • domain assumption A modern LLM (gpt-4o-mini) can judge topical fit of a new paper against a small sample of member abstracts with usable accuracy.
    Core premise of the Inspector stage (Section 4.4); never proved, only measured.
  • domain assumption Folder type (topical vs organizational) can be reliably inferred from name plus a quick look at members.
    Retriever typing step (Section 4.3); residual process/status folders remain hard.
  • domain assumption User accept/reject feedback is a sufficient online signal for the Reflector without offline training.
    Stated in Section 4.5; longitudinal evaluation left to future work.
invented entities (2)
  • Personalized Hierarchical Paper Routing (PHPR) no independent evidence
    purpose: Task formalization that distinguishes user-specific evolving folksonomies from shared HTC taxonomies.
    Defined in Section 3 and Table 1; the paper's primary conceptual contribution.
  • PaperRouter-Agent four-stage pipeline (Planner/Retriever/Inspector/Reflector) no independent evidence
    purpose: Concrete training-free architecture that realizes content-grounded routing.
    Introduced in Section 4; evaluated end-to-end on real libraries and LaMP-2.

reviewed 2026-07-14 · how reviews work

0 comments
Cite this review

Pith. "Pith review of PaperRouter-Agent: A Content-Grounded LLM Agent for Personalized Hierarchical Paper Routing." pith.science (2026). https://pith.science/paper/ET7FRCIC

@misc{pith2026260711564,
  author       = {Pith},
  title        = {Pith review of: PaperRouter-Agent: A Content-Grounded LLM Agent for Personalized Hierarchical Paper Routing},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ET7FRCIC}},
  note         = {Machine review of arXiv:2607.11564}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Researchers organize the papers they collect into personal folder hierarchies in reference managers, and route each new paper into the folder where it belongs. This task differs from standard hierarchical text classification. A user's folder hierarchy is not a fixed, shared taxonomy but a private and evolving folksonomy whose folder meanings may be topical, shorthand, venue-based, or process-oriented, and are often defined by the papers already stored inside them. We formalize this setting as personalized hierarchical paper routing (PHPR): assigning an incoming paper to folders in a user-specific hierarchy without per-user training. We propose PaperRouter-Agent, a training-free LLM agent that grounds routing decisions in folder members rather than folder names alone. The agent first narrows the candidate hierarchy, retrieves folder-specific evidence, verifies fit by inspecting member papers, and incorporates similarity-gated feedback from past user rejections. A formative study on real personal libraries shows that PaperRouter-Agent raises overall Recall@1 from 0.39 to 0.61 and Recall@3 from 0.57 to 0.83, with the largest gains on organizational folders defined by metadata such as venue or year, where single-shot methods collapses (Recall@1 0.09 to 0.50). On the public LaMP-2 benchmark, the same approach improves accuracy from 44.5% to 51.5% (+9.0 macro-F1) over a single-shot baseline, while remaining low-cost for practical use.

Figures

Figures reproduced from arXiv: 2607.11564 by Keshen Zhou, Lintao Wang, Suqin Yuan, Yu Luo, Zhiyong Wang, Zhuqiang Lu.

Figure 1
Figure 1. Figure 1: Example: Comparison of a name-based baseline and the proposed content￾grounded PaperRouter-Agent. While the baseline misclassifies a multimodal paper based solely on its title, the proposed agent accurately routes it to appropriate topical and metadata-defined folders. containing hundreds of papers. New papers keep arriving and the hierarchy itself keeps changing, so routing each paper into the right folde… view at source ↗
Figure 2
Figure 2. Figure 2: PaperRouter-Agent: a training-free four-stage agent that routes an incoming paper p into the user’s hierarchy Hu. The Planner short-lists candidate folders by top-down traversal. The Retriever types each folder and routes it, sampling member papers for topical folders while matching venue/year folders directly by metadata. The Inspector then reads the sampled evidence to decide Fit/No-Fit/Abstain and cross… view at source ↗
Figure 3
Figure 3. Figure 3: Per-class F1 on LaMP-2 (five most frequent tags, n = count in dev), single￾shot baseline vs. PaperRouter-Agent. Content-grounding helps most on subtle, content￾determined tags such as violence (+0.32) and twist ending. 5.4 Cost and Latency Analysis PaperRouter-Agent normally calls the LLM more than once per paper, so it can add latency over a single call. We report latency and call counts on gpt-4o-mini, a… view at source ↗
Figure 4
Figure 4. Figure 4: Cost, latency, and quality (gpt-4o-mini). Columns are (a) LLM calls, (b) latency, and (c) quality. The two rows are LaMP-2 (per paper) and real libraries (per query). The lighter bar is the one-time per-folder summary, and the gray dashed line marks the baseline. routing, and in particular the long-term gains the Reflector is designed for, be measured properly. We also plan to keep refining the system on t… view at source ↗
Figure 5
Figure 5. Figure 5: shows the configuration panel. Two APIs are set independently: an embedding API for the vector-similarity steps (candidate narrowing and the similarity-gated retrieval of past rejections in the Reflector), and an LLM API for zero-shot classification and the content-grounded verification of the Inspector. Decoupling them lets the embedding and LLM models be chosen independently, and cached embeddings are re… view at source ↗
Figure 6
Figure 6. Figure 6: Routing output with per-folder confidence: (a) SAM 3: Segment Anything with Concepts and (b) Unifying Scene Representation and Hand-Eye Calibration with 3D Foundation Models [PITH_FULL_IMAGE:figures/full_fig_p018_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Routing output with per-folder confidence, continued: (c) 3D-LLM: Injecting the 3D World into Large Language Models. C Inspector Pseudocode Algorithm 1 gives the pseudocode for the Inspector’s two steps: per-folder veri￾fication, then cross-comparison of the surviving Fit candidates (Section 4.4). Algorithm 1 Inspector: per-folder verification, then cross-comparison. Require: incoming paper p; topical cand… view at source ↗
Figure 8
Figure 8. Figure 8: Content-grounded inspection for 3D-LLaVA (scroll positions 1–2 of three): each candidate folder shows a Fit / No-Fit / Abstain verdict, a confidence, and the scope inferred from its members [PITH_FULL_IMAGE:figures/full_fig_p020_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: 3D-LLaVA, continued (scroll position 3 of three). D.1 Hyperparameters [PITH_FULL_IMAGE:figures/full_fig_p021_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Content-grounded inspection for EmbodiedSAM (two scroll positions). The Mode toggle selects the single-shot Baseline or the Inspector [PITH_FULL_IMAGE:figures/full_fig_p022_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: A failure case: Few-shot Scene-adaptive Anomaly Detection is incorrectly ac￾cepted into TMM (Fit, 0.85). You are deciding whether a new paper belongs in one specific folder of a researcher’s personal library. You are given the new paper, the folder’s NAME, and a sample of the papers ALREADY in that folder (this sample MAY BE EMPTY if the folder has no papers yet). Judge membership PRIMARILY from what the … view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

38 extracted references · 6 canonical work pages · 1 internal anchor

  1. [1]

    Cornell University Library (2026), https://arxiv.org/stats/monthly_submissions, accessed 2026-05-19; monthly submissions 24,226 (Oct 2024), 27,692 (Oct 2025), 30,045 (Mar 2026)

    arXiv: arXiv Monthly Submission Statistics. Cornell University Library (2026), https://arxiv.org/stats/monthly_submissions, accessed 2026-05-19; monthly submissions 24,226 (Oct 2024), 27,692 (Oct 2025), 30,045 (Mar 2026)

    2026

  2. [2]

    In: Proceedings of the International Conference on Learning Representations (ICLR) (2024)

    Asai, A., Wu, Z., Wang, Y., Sil, A., Hajishirzi, H.: Self-RAG: Learning to retrieve, generate, and critique through self-reflection. In: Proceedings of the International Conference on Learning Representations (ICLR) (2024)

    2024

  3. [3]

    BYOC: Personalized Few-Shot Classification with Co-Authored Class Descriptions

    Bohra, A., Verkes, G., Harutyunyan, A., Weinberger, P., Campagna, G.: BYOC: Personalized few-shot classification with co-authored class descriptions. In: Find- ings of the Association for Computational Linguistics: EMNLP (2023).https: //doi.org/10.48550/arXiv.2310.06111

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2310.06111 2023

  4. [4]

    In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2023).https://doi.org/10.1145/3544548.3580847

    Chang, J.C., Zhang, A.X., Bragg, J., Head, A., Lo, K., Downey, D., Weld, D.S.: CiteSee: Augmenting citations in scientific papers with persistent and personalized historical context. In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2023).https://doi.org/10.1145/3544548.3580847

    doi:10.1145/3544548.3580847 2023

  5. [5]

    Journal of Machine Learning Research (JMLR) (2024)

    Chung, H.W., Hou, L., Longpre, S., Zoph, B., Tay, Y., Fedus, W., Li, Y., Wang, X., Dehghani, M., Brahma, S., Webson, A., Gu, S.S., Dai, Z., Suzgun, M., Chen, X., Chowdhery, A., Castro-Ros, A., Pellat, M., Robinson, K., Valter, D., Narang, S., Mishra, G., Yu, A., Zhao, V., Huang, Y., Dai, A., Yu, H., Petrov, S., Chi, E.H., Dean, J., Devlin, J., Roberts, A....

    2024

  6. [6]

    In: Proceedings of NAACL: System Demonstrations (2022).https: //doi.org/10.18653/v1/2022.naacl-demo.10

    Ding, H., Yang, J., Deng, Y., Zhang, H., Roth, D.: Towards open-domain topic classification. In: Proceedings of NAACL: System Demonstrations (2022).https: //doi.org/10.18653/v1/2022.naacl-demo.10

    work page doi:10.18653/v1/2022.naacl-demo.10 2022

  7. [7]

    In: Proceedings of the Conference of the European Chap- ter of the Association for Computational Linguistics (EACL) (2026)

    Golde, J., Jedema, N.P., Krishnan, R., Le, P.: Hierarchical text classification with LLM-refined taxonomies. In: Proceedings of the Conference of the European Chap- ter of the Association for Computational Linguistics (EACL) (2026)

    2026

  8. [8]

    Journal of Information Science32(2), 198–208 (2006)

    Golder, S.A., Huberman, B.A.: Usage patterns of collaborative tagging systems. Journal of Information Science32(2), 198–208 (2006)

    2006

  9. [9]

    arXiv preprint (2025).https://doi.org/10.48550/arXiv.2510.19897

    Hassell, J., Zhang, D., Kim, H., Mitchell, T., Hruschka, E.: Learning from supervi- sionwithsemanticandepisodicmemory:Areflectiveapproachtoagentadaptation. arXiv preprint (2025).https://doi.org/10.48550/arXiv.2510.19897

    Pith/arXiv arXiv doi:10.48550/arxiv.2510.19897 2025

  10. [10]

    Transac- tions on Machine Learning Research (TMLR) (2022)

    Izacard, G., Caron, M., Hosseini, L., Riedel, S., Bojanowski, P., Joulin, A., Grave, E.: Unsupervised dense information retrieval with contrastive learning. Transac- tions on Machine Learning Research (TMLR) (2022)

    2022

  11. [11]

    In: Proceedings of the Annual Meeting of the Association for Computa- tional Linguistics (ACL) (2025).https://doi.org/10.18653/v1/2025.acl-long

    Kargupta, P., Zhang, N., Zhang, Y., Zhang, R., Mitra, P., Han, J.: TaxoAdapt: Aligning LLM-based multidimensional taxonomy construction to evolving research corpora. In: Proceedings of the Annual Meeting of the Association for Computa- tional Linguistics (ACL) (2025).https://doi.org/10.18653/v1/2025.acl-long. 1442

    doi:10.18653/v1/2025.acl-long 2025

  12. [12]

    In: Pro- ceedings of the 16th IEEE International Conference on Machine Learning and Applications (ICMLA)

    Kowsari, K., Brown, D.E., Heidarysafa, M., Jafari Meimandi, K., Gerber, M.S., Barnes, L.E.: HDLTex: Hierarchical deep learning for text classification. In: Pro- ceedings of the 16th IEEE International Conference on Machine Learning and Applications (ICMLA). pp. 364–371 (2017)

    2017

  13. [13]

    In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2024).https://doi.org/10.1145/ 3613904.3642196 14 K

    Lee, Y., Kang, H.B., Latzke, M., Kim, J., Bragg, J., Chang, J.C., Siangliulue, P.: PaperWeaver: Enriching topical paper alerts by contextualizing recommended papers with user-collected papers. In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2024).https://doi.org/10.1145/ 3613904.3642196 14 K. Zhou et al

    arXiv 2024

  14. [14]

    Semantic Web6(2), 167–195 (2015)

    Lehmann, J., Isele, R., Jakob, M., Jentzsch, A., Kontokostas, D., Mendes, P.N., Hellmann, S., Morsey, M., van Kleef, P., Auer, S., Bizer, C.: DBpedia – a large- scale, multilingual knowledge base extracted from Wikipedia. Semantic Web6(2), 167–195 (2015)

    2015

  15. [15]

    arXiv preprint (2026).https://doi.org/10.48550/arXiv.2603.19711

    Li, Y., Ma, T., Ye, Y.: EvoTaxo: Building and evolving taxonomy from social media streams. arXiv preprint (2026).https://doi.org/10.48550/arXiv.2603.19711

    work page doi:10.48550/arxiv.2603.19711 2026

  16. [16]

    arXiv preprint (2024).https: //doi.org/10.48550/arXiv.2401.05459

    Li, Y., Wen, H., Wang, W., Li, X., Yuan, Y.: Personal LLM agents: Insights and survey about the capability, efficiency and security. arXiv preprint (2024).https: //doi.org/10.48550/arXiv.2401.05459

    Pith/arXiv arXiv doi:10.48550/arxiv.2401.05459 2024

  17. [17]

    Bulletin of the Medical Library Association88(3), 265–266 (2000)

    Lipscomb, C.E.: Medical subject headings (MeSH). Bulletin of the Medical Library Association88(3), 265–266 (2000)

    2000

  18. [18]

    GitHub repository (2026),https://github.com/ lostwildland/zotero-llm-tagger, accessed 2026-05-19

    lostwildland: AI Tagger for Zotero (zotero-llm-tagger): LLM-based automatic tag suggestion for zotero 7. GitHub repository (2026),https://github.com/ lostwildland/zotero-llm-tagger, accessed 2026-05-19

    2026

  19. [19]

    arXiv preprint (2025).https://doi.org/10.48550/arXiv.2510.01609

    Ma, B., Li, H., Hu, Z., Gui, X., Liu, L.: AgentRec: Next-generation LLM- powered multi-agent collaborative recommendation with adaptive intelligence. arXiv preprint (2025).https://doi.org/10.48550/arXiv.2510.01609

    work page doi:10.48550/arxiv.2510.01609 2025

  20. [20]

    In: Advances in Neural Information Processing Systems (NeurIPS) (2022)

    Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C.L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P., Leike, J., Lowe, R.: Train- ing language models to follow instructions with human feedback. In: Advances in Neural Information Process...

    2022

  21. [21]

    GitHub repository (2024),https://github.com/papersgpt/ papersgpt-for-zotero, accessed 2026-05-19

    papersgpt contributors: PapersGPT: Chat with pdfs and autopilot summaries for zotero (MCP server). GitHub repository (2024),https://github.com/papersgpt/ papersgpt-for-zotero, accessed 2026-05-19

    2024

  22. [22]

    Radford, A., Kim, J.W., Hallacy, C., Ramesh, A., Goh, G., Agarwal, S., Sastry, G., Askell, A., Mishkin, P., Clark, J., Krueger, G., Sutskever, I.: Learning transferable visual models from natural language supervision (2021),https://arxiv.org/abs/ 2103.00020

    Pith/arXiv arXiv 2021

  23. [23]

    In: Proceedings of the Annual Meeting of the Asso- ciation for Computational Linguistics (ACL) (2024)

    Salemi, A., Mysore, S., Bendersky, M., Zamani, H.: LaMP: When large language models meet personalization. In: Proceedings of the Annual Meeting of the Asso- ciation for Computational Linguistics (ACL) (2024)

    2024

  24. [25]

    In: Proceedings of IEEE International Conference on Big Data (BigData) (2024).https://doi.org/ 10.1109/BigData62323.2024.10825412

    Schmidt, F., Hammerfald, K., Jahren, H.H., Payberah, A.H., Vlassov, V.: Single- pass hierarchical text classification with large language models. In: Proceedings of IEEE International Conference on Big Data (BigData) (2024).https://doi.org/ 10.1109/BigData62323.2024.10825412

    doi:10.1109/bigdata62323.2024.10825412 2024

  25. [26]

    In: Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL-HLT)

    Shen, J., Qiu, W., Meng, Y., Shang, J., Ren, X., Han, J.: TaxoClass: Hierarchical multi-label text classification using only class names. In: Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics (NAACL-HLT). pp. 4239–4249 (2021)

    2021

  26. [27]

    In: Advances in Neural Information Processing Systems (NeurIPS) (2023)

    Shinn, N., Cassano, F., Berman, E., Gopinath, A., Narasimhan, K., Yao, S.: Reflex- ion: Language agents with verbal reinforcement learning. In: Advances in Neural Information Processing Systems (NeurIPS) (2023)

    2023

  27. [28]

    In: Proceedings of the ACM Interna- tional Conference on Information and Knowledge Management (CIKM) (2023)

    Singh, M., Cambronero, J., Gulwani, S., Le, V., Verbruggen, G.: EmFore: Online learning of email folder classification rules. In: Proceedings of the ACM Interna- tional Conference on Information and Knowledge Management (CIKM) (2023). https://doi.org/10.1145/3583780.3614863 PaperRouter-Agent 15

    doi:10.1145/3583780.3614863 2023

  28. [29]

    Publi- cations7(1), 18 (2019).https://doi.org/10.3390/publications7010018

    Tenopir, C., Christian, L., Kaufman, J.: Seeking, reading, and use of scholarly articles: An international study of perceptions and behavior of researchers. Publi- cations7(1), 18 (2019).https://doi.org/10.3390/publications7010018

    work page doi:10.3390/publications7010018 2019

  29. [30]

    Vander Wal, T.: Folksonomy coinage and definition.https://vanderwal.net/ folksonomy.html(2007), accessed 2026-06-14

    2007

  30. [31]

    In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2025).https://doi.org/10.1145/3706598.3713691

    Wang, L., Yurechko, K., Dani, P., Chen, Q.Z., Zhang, A.X.: End user authoring of personalized content classifiers: Comparing example labeling, rule writing, and LLM prompting. In: Proceedings of the ACM CHI Conference on Human Factors in Computing Systems (2025).https://doi.org/10.1145/3706598.3713691

    doi:10.1145/3706598.3713691 2025

  31. [32]

    In: Proceedings of the Inter- national Conference on Learning Representations (ICLR) (2023)

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., Cao, Y.: ReAct: Synergizing reasoning and acting in language models. In: Proceedings of the Inter- national Conference on Learning Representations (ICLR) (2023)

    2023

  32. [33]

    Yoshimura, K., Kashima, H.: Hierarchical text classification using black box large languagemodels.arXivpreprint(2025).https://doi.org/10.48550/arXiv.2508. 04219

    doi:10.48550/arxiv.2508 2025

  33. [34]

    Electronics13(7), 1199(2024).https://doi.org/10.3390/electronics13071199

    Zangari, A., Bertolucci, M., Frasca, M.: Hierarchical text classification: A survey. Electronics13(7), 1199(2024).https://doi.org/10.3390/electronics13071199

    work page doi:10.3390/electronics13071199 2024

  34. [35]

    In: Findings of the Association for Computational Linguistics: EMNLP (2025).https://doi.org/10.18653/v1/ 2025.findings-emnlp.207

    Zhang, Q., Su, Q., Zhu, W., Pang, Y.: HierPrompt: Zero-shot hierarchical text classification with LLM-enhanced prototypes. In: Findings of the Association for Computational Linguistics: EMNLP (2025).https://doi.org/10.18653/v1/ 2025.findings-emnlp.207

    doi:10.18653/v1/ 2025

  35. [36]

    In: Proceedings of the ACM Web Conference (WWW) (2024).https://doi.org/10

    Zhang, Y., Yang, R., Xu, X., Li, R., Xiao, J.: TELEClass: Taxonomy enrichment and LLM-enhanced hierarchical text classification with minimal supervision. In: Proceedings of the ACM Web Conference (WWW) (2024).https://doi.org/10. 1145/3696410.3714940

    arXiv 2024

  36. [37]

    arXiv preprint (2024).https://doi.org/10.48550/ arXiv.2411.00027

    Zhang, Z., Rossi, R.A., Kveton, B., Shao, Y., Yang, D.: Personalization of large language models: A survey. arXiv preprint (2024).https://doi.org/10.48550/ arXiv.2411.00027

    Pith/arXiv arXiv 2024

  37. [38]

    arXiv preprint (2025).https://doi.org/10.48550/arXiv

    Zhou, H., Chen, Y., Guo, S., et al.: Memento: Fine-tuning LLM agents with- out fine-tuning LLMs. arXiv preprint (2025).https://doi.org/10.48550/arXiv. 2508.16153

    Pith/arXiv arXiv doi:10.48550/arxiv 2025

  38. [39]

    Randomly pick 20 papers you usually read and file each into your own hierarchy (collections)

    Zhou, J., Ma, C., Long, D., Xu, G., Ding, N., Zhang, H., Xie, P., Liu, G.: Hierarchy- aware global model for hierarchical text classification. In: Proceedings of the 58th AnnualMeetingoftheAssociationforComputationalLinguistics(ACL).pp.1106– 1117 (2020) 16 K. Zhou et al. Appendix This appendix collects material referenced from the main paper: qualitative ...

    2020

This paper was first reviewed by grok-4.5 on July 14, 2026.