Pith. sign in

REVIEW 4 major objections 5 minor 1 cited by

From Millions of Tweets to Actionable Insights: Leveraging LLMs for User Profiling

T0 review · 4 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read A two-stage LLM pipeline turns millions of tweets into compact profiles that preserve political stance information better than existing user-profiling methods, reporting a 9.8% improvement over baselines.

desk verdict A solid, well-engineered profiling pipeline whose headline 9.8% claim depends on an incompletely specified size equalization and a mildly circular evaluation; deserves review but needs revision. read the letter →

arxiv 2505.06184 v1 pith:DP3YQZZY submitted 2025-05-09 cs.SI cs.CLcs.IR

Vahid Rahimzadeh , Ali Hamzehpour , Azadeh Shakery , Masoud Asadpour This is my paper · ORCID
classification cs.SIcs.CLcs.IR
keywords userprofilinglargelanguagemodelsstancedetectionsemi-supervisedfilteringknowledgebaseextractivesummarizationabstractivePersianpoliticalTwitter
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 sets out to establish that a user's political stance on Twitter (X) can be distilled into a short, human-readable profile without losing the information that matters. The proposed method first filters a user's timeline with a semi-supervised knowledge-base similarity step, then uses an LLM to generate, for each of 15 expert-selected stance claims, a short factual summary and a set of cited original tweets. The authors argue these profiles are flexible across domains, need no large labeled datasets, and outperform existing summarization, retrieval, and latent-embedding baselines on an open-book stance-detection evaluation. The headline result is a 9.8% improvement over baselines, with the extractive profile variant reaching macro F1 of 0.6668 under a GPT-4o judge.

What carries the argument

The central object is the set of 'domain-defining statements'—15 expert-curated stance claims generated by zero-shot LLM prompting from a held-out split of the data. These statements serve triple duty: they are the aspects around which the profile is organized, the queries for the open-book QA evaluation, and the anchors for human annotation of ground truth. The complementary machinery is the two-stage pipeline: a semi-supervised filter that scores tweets against a Wikidata-derived Persian political knowledge base using BGE-M3 embeddings and a fine-tuned TookaBERT classifier, and the profile generator that produces cited abstractive summaries and extractive tweet selections per statement. The evaluation compares stance-detection accuracy when context is the full pooled tweet history versus the compact profile, treating the gap as a measure of information loss.

What would settle it

Run the same 1,500 user-statement pairs with an evaluator from a different model family—for example an open-weight LLM or a fine-tuned Farsi stance classifier—and check whether the extractive profile's lead over random and BM25 selection persists. If the gap narrows or reverses, the claim that the profiles preserve stance information better than the baselines is not robust to evaluator choice. A second decisive check is to build a fresh set of stance claims by independent experts and see whether the F1 gap survives on claims that were never used to construct the profiles.

Watch

Extended reading notes

Core claim

The paper's central claim is that profiling a user by anchoring on domain-defining statements—here, 15 stance claims about Persian politics—and generating both abstractive summaries and extractive tweet selections for each claim preserves stance-relevant information better than the alternatives tested. On 1,500 human-annotated user-statement pairs, the extractive profiles reach a macro F1 of 0.6668 with GPT-4o as evaluator, significantly ahead of random selection, BM25, semantic retrieval, SemAE, and the Amazon LLM-personalization baselines; the abstractive variant also leads under the Gemini Flash evaluator. The authors attribute the extractive advantage to the fact that it keeps the user's original wording, which avoids cultural and political biases that appear when the model paraphrases in its own words.

Load-bearing premise

The result stands or falls on the assumption that scoring a profile against 15 expert-selected stance claims with an LLM judge is a fair and unbiased test of profile quality; if the judge simply prefers original tweet text or outputs from its own model family, the reported lead could be an artifact of the evaluation rather than a property of the profiles.

Editorial extensions

If this is right

  • A compact profile built from 15 stance-anchored statements can stand in for a user's full tweet history in stance-based analyses, which makes large-scale social-network reasoning computationally feasible.
  • The same pipeline transfers to a new domain by rebuilding the knowledge base and regenerating defining statements; the paper demonstrates the knowledge-base construction using Wikidata for European football.
  • Extractive profiles are the safer default when the target domain is politically or culturally specific, because abstractive summaries inherit the LLM's Western-centric biases.
  • The PersianPol6M dataset and the human-validated LLM evaluation framework give later work a benchmark for comparing profiling methods on stance preservation.

Reading between the lines

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

  • Beyond the paper: if the evaluator-bias hypothesis is correct, the extractive lead should shrink when the judge is an open-weight LLM or a fine-tuned Farsi stance classifier; that is a cheap, direct test of the robustness of the reported gap.
  • Beyond the paper: the 15 claims were selected by experts and used for both generation and evaluation; a second set of claims from independent annotators would reveal whether the method's advantage is specific to the chosen claims or generalizes to other stances.
  • Beyond the paper: because extractive profiles keep the original tweet text with citations, they could double as auditable evidence for downstream tasks such as misinformation detection, where traceability is a requirement.
  • Beyond the paper: the semi-supervised filter's 93% precision on positive examples suggests the same knowledge-base-similarity labeling could bootstrap classifiers for other high-signal domains without manual annotation.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper proposes a two-stage LLM-based user profiling pipeline for Persian political Twitter. In the first stage, a semi-supervised filtering method uses a Wikidata- and Wikipedia-derived Persian political knowledge base with BGE-M3 embeddings to label and filter political tweets, yielding the PersianPol6M dataset of over 1.7 million political tweets. In the second stage, the method generates 15 domain-defining stance claims from a first split of users, then profiles users in a second split by producing, for each claim, an abstractive factual summary and an extractive set of supporting tweets. Profile quality is evaluated by an open-book stance-detection QA task in which an LLM reads the profile and answers the 15 claims; answers are compared against human-annotated ground truth for 100 users. The authors report that the extractive profile achieves macro F1 0.6668 with GPT-4o as evaluator, outperforming baselines including random selection, BM25, semantic retrieval, Amazon summarization/RAG, and SemAE, with statistical significance in most comparisons, and the abstract states a headline improvement of 9.8%.

Significance. If the central comparison is fair, the paper makes a useful contribution: it introduces a new Persian political Twitter dataset, a semi-supervised filtering pipeline that reduces annotation cost, and a natural-language profile representation that is interpretable and usable by LLMs. The evaluation design has genuine strengths: the ground-truth labels are produced by human annotators from pooled tweet histories and are thus independent of profile generation; the authors provide detailed appendices with prompts, annotation guidelines, and implementation parameters; and the use of two LLM evaluators (GPT-4o and Gemini Flash 1.5) is a reasonable attempt at robustness. The strongest claim that extractive, citation-grounded tweet selection preserves stance-relevant information better than abstractive summarization or retrieval baselines is plausible and worth investigating. However, the paper's central quantitative claims are weakened by an uncontrolled profile-size comparison, an unreported derivation of the 9.8% headline number, and an evaluation setup in which the 15 claims are simultaneously the profile-generation dimensions and the evaluation queries.

major comments (4)
  1. [§4.2 and §5, 'Standardized Profile Size'] The comparison is not demonstrably controlled on profile size. Section 4.2 defines the extractive profile as a set of supporting tweets Eu,i for each stance claim si, with no stated cardinality cap on Eu,i. Section 5 states only that 'we limited other selection methods to retrieve the top-1 result for each statement, resulting in 15 tweets per user,' which explicitly caps the baselines but not the proposed extractive profile. If each Eu,i contains more than one tweet, the extractive profile gives the LLM evaluator substantially more context than the 15-tweet baselines, and the reported macro F1 gap could reflect information volume rather than profile quality. The abstractive profile, which generates 'a few sentences for each of the 15 statements,' is likewise not matched in length to the baselines. Please report the mean and maximum number of tweets per statement in the extractive profiles, and add an ablation in which the extractive and abstractive profiles are truncated or padded to match the 15-tweet baseline size.
  2. [Abstract and §6.2] The headline '9.8%' improvement is not derived anywhere in the text. Section 6.2 reports a macro F1 of 0.6668 for the extractive method with GPT-4o as evaluator and states that this 'significantly outperforms all other methods,' but no table or equation gives the baseline F1 values or the pairwise differences that produce 9.8%. Because the profile-size issue in the previous comment affects any quantitative gap, the paper should report the full F1 table for both evaluators and explicitly state which comparison yields the 9.8% figure.
  3. [§4.3, Figure 7, and §4.2] The evaluation is partially circular with respect to the claim of general profile quality. The same 15 stance claims serve both as the dimensions along which profiles are generated (Section 4.2) and as the queries in the open-book QA evaluation (Section 4.3), so the profiles are explicitly constructed to encode the information being tested. A high score on this task shows that the profiling method preserves the target information, but it does not establish that the profiles are generally superior for other stances or downstream tasks. Moreover, the GPT-4o evaluator belongs to the same model family as the GPT-4o-mini profile generator; the observed gap may partly reflect format or family compatibility. Please add an evaluation on held-out stance claims that were not used in profile construction, and report results with an evaluator from a different family or with a non-LLM classifier to test for evaluator bias.
  4. [§4.3, Appendix A.3, Table 3] The statistical significance reporting conflates significance with effect size and does not account for multiple comparisons. Describing the GPT-4o extractive result as '100% statistical significance' because all McNemar p-values are below 0.05 is misleading: with 14 comparisons, the family-wise error rate is not controlled, and several p-values (e.g., 0.0002, 0.0003) are close to the threshold after repeated testing. Please report corrected p-values (e.g., Benjamini-Hochberg) or state that the comparisons are exploratory, and also report effect sizes or confidence intervals for the F1 differences.
minor comments (5)
  1. [Table 1] The caption says 'Percentage changes in the total tweets, users, retweets, likes, and tweet lengths after filtering,' but the table rows show Total Twts, Total Users, Avg Twt/User, Avg Twt Len, and Median Twt Len; retweets and likes are not actually listed. The caption and table content should be aligned.
  2. [Section 2.2] There is a typographical error in the citation '(Kanoje, Girase, and Mukhopadhyay 2015))' with an extra closing parenthesis; also the sentence beginning 'Ongoing research focuses...' is a fragment that should be merged with the preceding sentence.
  3. [Section 5, Baseline descriptions] The random-selection baseline is described as having two variations, one operating on the pooled tweets and one on the complete user tweet history, but the 'complete user history' variation is not clearly defined in terms of how many tweets are used and whether it is subject to the 15-tweet standardization. Please clarify.
  4. [Figure 8] The caption states that results are for 'Gemini Flash 1.5 and GPT 4o mini,' but Section 6.2 refers to 'GPT-4o as evaluator'; the naming should be made consistent (GPT-4o vs. GPT-4o-mini) to avoid confusion about which model produced the reported 0.6668 F1.
  5. [Appendix A.3, Table 3] The table is only referenced indirectly in the main text; please add an explicit pointer in Section 6.2 so readers can find the McNemar results without hunting in the appendix.

Circularity Check

1 steps flagged · score 6.0 of 10

Evaluation queries are the same 15 stance claims used to construct the profiles, so the reported F1 gains are partly by construction.

  1. self definitional [Section 4.2 (User Profiling) and Section 4.3 (LLM-Based Profile Evaluation); Figures 6 and 7]
    "These claims will act as the foundation aspects of our user profiling and evaluation methods. ... we provide the LLM with the complete set of pooled tweets as reference context alongside each stance claim. ... This yields two categories of profiles per user. ... The QA task aligns with stance detection on predefined claims."

    The same 15 stance claims are used twice: as the generation prompt for each profile segment (abstractive summary fi and supporting tweet set Eu,i) and as the query in the evaluation QA task. For user u and claim si, the abstractive profile contains a summary generated specifically about si, and the extractive profile contains tweets cited as supporting si. The evaluator is then asked to decide si from that profile. The reported macro-F1 therefore measures whether the profile-generation prompt succeeded in placing si-relevant content into the profile, not whether a generic profile preserves stance information; the advantage over baselines that receive a single retrieved tweet per claim is forced by construction.

full rationale

The paper contains no load-bearing self-citation chain, no imported uniqueness theorem, and no ansatz smuggled in via citation; the filtering pipeline, human annotations, and evaluator are independent. The main circularity is that the 15 domain-defining statements are explicitly 'the foundation aspects of our user profiling and evaluation methods.' Each proposed profile is a per-claim summary or supporting-tweet set generated with the claim as input (Figure 6), and the evaluation re-asks that same claim (Figure 7). Thus the stance-label prediction is essentially a readout of the claim-specific profile segment, while baselines are limited to one retrieved tweet per claim ('we limited other selection methods to retrieve the top-1 result for each statement'). This makes the headline 9.8% gap partially a construction artifact. The paper's own standardization sentence does not cap the extractive Eu,i, compounding the imbalance. Because the ground-truth labels and evaluator are external and the filtering/annotation pipeline is independent, the circularity is partial rather than total, yielding a score of 6.

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

The claimed performance rests on several modeling choices: threshold and chunk parameters for filtering, the specific set of 15 stance claims, the size of the tweet pool, and the standardization of baseline profile sizes. The paper also assumes that embedding similarity, human annotations with moderate agreement, and LLM evaluator outputs are all faithful measures of profile quality. No new physical or conceptual entities are introduced beyond the methodological constructs of 'domain-defining statements' and the PP6M dataset.

free parameters (5)
  • distance threshold theta = 0.7
    Used in Eq. 1 to label tweets as political or non-political when mean cosine distance to the k nearest knowledge-base chunks is below 1-theta or above theta; tuned empirically for 'high-quality labels' (Section 3.3).
  • nearest chunks k = 10
    Number of closest knowledge-base chunks averaged for labeling in Eq. 1; tuned jointly with theta (Section 3.3).
  • number of stance claims = 15
    Selecting 15 claims from over 500 LLM-generated statements via deduplication and expert selection; the evaluation is defined on these claims, so reported F1 is conditional on this choice (Section 4.2).
  • pool size per user = 80 tweets (20 per each of 4 selection methods)
    Chosen to make human annotation and LLM experiments feasible; profiles and evaluation both draw from this pool (Section 4.2, 'Pooling on Tweets').
  • standardized baseline profile size = 15 tweets (top-1 per statement)
    Baselines are capped at 15 tweets while the proposed abstractive profile includes per-claim summaries and extractive sets may vary, so token budgets are not actually equal (Section 5, 'Standardized Profile Size').
assumptions (5)
  • domain assumption Cosine distance between BGE-M3 embeddings reflects political relevance of a tweet to a Persian political knowledge base
    The semi-supervised filter in Eq. 1 assumes embedding similarity in BGE-M3 captures domain relatedness for Persian political text (Section 3.3).
  • ad hoc to paper The 15 expert-selected LLM-generated stance claims are representative domain-defining statements for Persian political Twitter
    Profiles and evaluation are built around these 15 claims; no evidence is given that they generalize to other claims or domains (Section 4.2, Table 2).
  • domain assumption Human annotations of user-statement pairs from an 80-tweet pool are a reliable ground truth (reported Cohen's kappa 0.63)
    The evaluation compares LLM answers to these labels; moderate agreement means labels are noisy, which the paper acknowledges (Sections 4.3 and 6.3).
  • domain assumption Zero-shot LLM stance detection on the profile context is an unbiased measure of profile information content
    The evaluation assumes the evaluator (GPT-4o or Gemini Flash) answers based only on the profile and treats its accuracy as profile quality (Section 4.3, Figure 7).
  • domain assumption The Wikidata/Wikipedia-based knowledge base covers the Persian political domain well enough for filtering
    Filtering recall depends on the knowledge base's coverage; the paper reports only precision on positives, with no recall estimate (Section 3.2).

how reviews work

0 comments
Cite this review

Pith. "Pith review of From Millions of Tweets to Actionable Insights: Leveraging LLMs for User Profiling." pith.science (2026). https://pith.science/paper/DP3YQZZY

@misc{pith2026250506184,
  author       = {Pith},
  title        = {Pith review of: From Millions of Tweets to Actionable Insights: Leveraging LLMs for User Profiling},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DP3YQZZY}},
  note         = {Machine review of arXiv:2505.06184}
}
read the original abstract

Social media user profiling through content analysis is crucial for tasks like misinformation detection, engagement prediction, hate speech monitoring, and user behavior modeling. However, existing profiling techniques, including tweet summarization, attribute-based profiling, and latent representation learning, face significant limitations: they often lack transferability, produce non-interpretable features, require large labeled datasets, or rely on rigid predefined categories that limit adaptability. We introduce a novel large language model (LLM)-based approach that leverages domain-defining statements, which serve as key characteristics outlining the important pillars of a domain as foundations for profiling. Our two-stage method first employs semi-supervised filtering with a domain-specific knowledge base, then generates both abstractive (synthesized descriptions) and extractive (representative tweet selections) user profiles. By harnessing LLMs' inherent knowledge with minimal human validation, our approach is adaptable across domains while reducing the need for large labeled datasets. Our method generates interpretable natural language user profiles, condensing extensive user data into a scale that unlocks LLMs' reasoning and knowledge capabilities for downstream social network tasks. We contribute a Persian political Twitter (X) dataset and an LLM-based evaluation framework with human validation. Experimental results show our method significantly outperforms state-of-the-art LLM-based and traditional methods by 9.8%, demonstrating its effectiveness in creating flexible, adaptable, and interpretable user profiles.

Figures

Figures reproduced from arXiv: 2505.06184 by the authors.

Figure 1
Figure 1. Dataset curation pipeline for political content related to the 2024 Iranian presidential election, including data collec [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Venn diagram illustrating the overlap of unique [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Confusion matrices for: (a) unsupervised labeling on human annotations, (b) classifier fine-tuning with semi [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Histogram of mean distances between tweets and [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Step-by-step illustration of the user profiling and evaluation process. The pipeline includes (1) user splitting, (2) [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 7
Figure 7. Figure 7: Prompt utilized for evaluation step For our selected political domain, the QA task aligns with stance detection on predefined claims. We created an eval￾uation dataset with 15 political stance claims for 100 users, with human annotators labeling each user-statement pai…
Figure 8
Figure 8. Figure 8: Macro-F1 scores of different user profiling methods evaluated using Gemini Flash 1.5 and GPT 4o mini models. Error [PITH_FULL_IMAGE:figures/full_fig_p008_8.png]
Figure 9
Figure 9. Figure 9: The summarization prompt used in the Amazon [PITH_FULL_IMAGE:figures/full_fig_p010_9.png]
Figure 10
Figure 10. Figure 10: The prompt used during the statement generation [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. PolitiSky24: U.S. Political Bluesky Dataset with User Stance Labels

    cs.CL 2025-06 conditional novelty 5.0 of 10

    PolitiSky24 provides 16,044 AI-labeled user-level stance pairs for Trump and Harris from 8,467 Bluesky users, with the labeling pipeline reporting 81% validation accuracy.

Reference graph

Works this paper leans on

42 extracted references · 25 canonical work pages · cited by 1 Pith paper

  1. [1]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.a...

  2. [2]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...

  3. [3]

    Amati, G. 2009. BM25, 257--260. Boston, MA: Springer US. ISBN 978-0-387-39940-9

    work page 2009

  4. [4]

    W.; Pearson, R.; and Kellermanns, F

    Barnett, T.; Pearson, A. W.; Pearson, R.; and Kellermanns, F. W. 2015. Five-factor model personality traits as predictors of perceived and actual usage of technology. European Journal of Information Systems, 24(4): 374--390

    work page 2015

  5. [5]

    Basu Roy Chowdhury, S.; Zhao, C.; and Chaturvedi, S. 2022. Unsupervised Extractive Opinion Summarization Using Sparse Coding. In Muresan, S.; Nakov, P.; and Villavicencio, A., eds., Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 1209--1225. Dublin, Ireland: Association for Computational Linguistics

    work page 2022

  6. [6]

    D.; Guillaume, J.-L.; Lambiotte, R.; and Lefebvre, E

    Blondel, V. D.; Guillaume, J.-L.; Lambiotte, R.; and Lefebvre, E. 2008. Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, 2008(10): P10008

    2008

  7. [7]

    Chakma, K.; Das, A.; and Debbarma, S. 2021. Summarization of Twitter Events with Deep Neural Network Pre-trained Models. In Lossio-Ventura, J. A.; Valverde-Rebaza, J. C.; D \'i az, E.; and Alatrista-Salas, H., eds., Information Management and Big Data, 45--62. Cham: Springer International Publishing. ISBN 978-3-030-76228-5

    work page 2021

  8. [8]

    Chaudhry, P.; and Lease, M. 2022. You Are What You Tweet: Profiling Users by Past Tweets to Improve Hate Speech Detection. In Smits, M., ed., Information for a Better World: Shaping the Global Future, 195--203. Cham: Springer International Publishing. ISBN 978-3-030-96960-8

    work page 2022

Show all 42 references
  1. [9]

    Chen, J.; Xiao, S.; Zhang, P.; Luo, K.; Lian, D.; and Liu, Z. 2024. BGE M3-Embedding: Multi-Lingual, Multi-Functionality, Multi-Granularity Text Embeddings Through Self-Knowledge Distillation. arXiv:2402.03216

    2024 arXiv

  2. [10]

    Cheng, J.; and Lapata, M. 2016. Neural Summarization by Extracting Sentences and Words. In Erk, K.; and Smith, N. A., eds., Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 484--494. Berlin, Germany: Association f...

    2016

  3. [11]

    Christakopoulou, K.; Lalama, A.; Adams, C.; Qu, I.; Amir, Y.; Chucri, S.; Vollucci, P.; Soldo, F.; Bseiso, D.; Scodel, S.; et al. 2023. Large language models for user interest journeys. arXiv preprint arXiv:2305.15498

    2023 arXiv

  4. [12]

    O.; Goya, D

    de Fran c a, F. O.; Goya, D. H.; and de Camargo Penteado, C. L. 2018. User profiling of the Twitter Social Network during the impeachment of Brazilian President. Social Network Analysis and Mining, 8(1): 5

    2018

  5. [13]

    Dusart, A.; Pinel-Sauvagnat, K.; and Hubert, G. 2021. TSSuBERT: Tweet Stream Summarization Using BERT. arXiv:2106.08770

    2021 arXiv

  6. [14]

    Dvir-Gvirsman, S. 2022. Understanding news engagement on social media: A media repertoire approach. New Media & Society, 24(8): 1791--1812. PMID: 35910420

    2022

  7. [15]

    M.; R, P

    G, B. M.; R, P. K.; V, V. J. D.; N, A.; Faheem, M.; T, S. K. J.; and K, K. 2024. Comparative Evaluation of Large Language Models for Abstractive Summarization. In 2024 14th International Conference on Cloud Computing, Data Science & Engineering (Confluence), 59--64

    2024

  8. [16]

    Gao, C.; Lan, X.; Lu, Z.; Mao, J.; Piao, J.; Wang, H.; Jin, D.; and Li, Y. 2023. S3: Social-network Simulation System with Large Language Model-Empowered Agents. arXiv:2307.14984

    2023 arXiv

  9. [17]

    K.; Chakraborty, R.; and Dandapat, S

    Garg, P. K.; Chakraborty, R.; and Dandapat, S. K. 2024. OntoDSumm: Ontology-Based Tweet Summarization for Disaster Events. IEEE Transactions on Computational Social Systems, 11(2): 2724--2739

    2024

  10. [18]

    J.; and Durrett, G

    Goyal, T.; Li, J. J.; and Durrett, G. 2022. News Summarization and Evaluation in the Era of GPT-3. arXiv:2209.12356

    2022 arXiv

  11. [19]

    Gupta, A.; and Mehta, S. 2022. Automatic Stance Detection for Twitter Data. In 2022 1st International Conference on Informatics (ICI), 223--225

    2022

  12. [20]

    Jiang, J.; and Ferrara, E. 2023. Social-LLM: Modeling User Behavior at Scale using Language Models and Social Network Data. arXiv:2401.00893

    2023 arXiv

  13. [21]

    Jin, X.; and Han, J. 2010. K-Means Clustering, 563--564. Boston, MA: Springer US. ISBN 978-0-387-30164-8

    2010

  14. [22]

    Kanoje, S.; Girase, S.; and Mukhopadhyay, D. 2015. User profiling trends, techniques and applications. arXiv preprint arXiv:1503.07474

    2015 arXiv

  15. [23]

    F.; Lin, K.; Hewitt, J.; Paranjape, A.; Bevilacqua, M.; Petroni, F.; and Liang, P

    Liu, N. F.; Lin, K.; Hewitt, J.; Paranjape, A.; Bevilacqua, M.; Petroni, F.; and Liang, P. 2024. Lost in the Middle: How Language Models Use Long Contexts. Transactions of the Association for Computational Linguistics, 12: 157--173

    2024

  16. [24]

    Mishra, N.; Sahu, G.; Calixto, I.; Abu-Hanna, A.; and Laradji, I. H. 2023. LLM aided semi-supervision for Extractive Dialog Summarization. arXiv:2311.11462

    2023 arXiv

  17. [25]

    Pu, X.; Gao, M.; and Wan, X. 2023. Summarization is (Almost) Dead. arXiv:2309.09558

    2023 arXiv

  18. [26]

    Purificato, E.; Boratto, L.; and De Luca, E. W. 2024. User Modeling and User Profiling: A Comprehensive Survey. arXiv preprint arXiv:2402.09660

    2024 arXiv

  19. [27]

    Z.; and Sethy, A

    Richardson, C.; Zhang, Y.; Gillespie, K.; Kar, S.; Singh, A.; Raeesy, Z.; Khan, O. Z.; and Sethy, A. 2023. Integrating summarization and retrieval for enhanced personalization via large language models. arXiv preprint arXiv:2310.20081

    2023 arXiv

  20. [28]

    Rostami, P.; and Shakery, A. 2023. A deep learning-based expert finding method to retrieve agile software teams from CQAs. Information Processing & Management, 60(2): 103144

    2023

  21. [29]

    Z.; and Salehoof, A

    SadraeiJavaheri, M.; Moghaddaszadeh, A.; Molazadeh, M.; Naeiji, F.; Aghababaloo, F.; Rafiee, H.; Amirmahani, Z.; Abedini, T.; Sheikhi, F. Z.; and Salehoof, A. 2024. TookaBERT: A Step Forward for Persian NLU. arXiv:2407.16382

    2024 arXiv

  22. [30]

    Saleema, A.; and Thampi, S. 2020. User Recognition Using Cognitive Psychology Based Behavior Modeling in Online Social Networks, 130--149. ISBN 978-981-15-4827-7

    2020

  23. [31]

    Salemi, A.; Mysore, S.; Bendersky, M.; and Zamani, H. 2024. L a MP : When Large Language Models Meet Personalization. In Ku, L.-W.; Martins, A.; and Srikumar, V., eds., Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pape...

    2024

  24. [32]

    Schiaffino, S. 2003. User profiling with Case-Based Reasoning and Bayesian Networks

    2003

  25. [33]

    Shu, K.; Wang, S.; and Liu, H. 2018. Understanding User Profiles on Social Media for Fake News Detection

    2018

  26. [34]

    Shu, K.; Zhou, X.; Wang, S.; Zafarani, R.; and Liu, H. 2020. The role of user profiles for fake news detection. In Proceedings of the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining, ASONAM '19, 436–439. New York, NY, USA: Association ...

    2020

  27. [35]

    Stolfo, A. 2024. Groundedness in Retrieval-augmented Long-form Generation: An Empirical Study. arXiv preprint arXiv:2404.07060

    2024 arXiv

  28. [36]

    Tan, Z.; and Jiang, M. 2023. User modeling in the era of large language models: Current research and future directions. arXiv preprint arXiv:2312.11518

    2023 arXiv

  29. [37]

    Vrande c i\' c , D.; and Kr\" o tzsch, M. 2014. Wikidata: a free collaborative knowledgebase. Commun. ACM, 57(10): 78–85

    2014

  30. [38]

    Wu, C.; Wu, F.; Huang, Y.; and Xie, X. 2023. Personalized News Recommendation: Methods and Challenges. ACM Trans. Inf. Syst., 41(1)

    2023

  31. [39]

    Wu, C.; Wu, F.; Qi, T.; and Huang, Y. 2021. Empowering News Recommendation with Pre-trained Language Models. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR '21, 1652–1656. New York, NY, USA: Association...

    2021

  32. [40]

    Wu, C.; Wu, F.; Qi, T.; and Huang, Y. 2022. UserBERT: Pre-training User Model with Contrastive Self-supervision. In Proceedings of the 45th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR '22, 2087–2092. New York, NY, USA: Associa...

    2022

  33. [41]

    T.; and Foster, J

    Zarharan, M.; Hashemi, M.; Behroozrazegh, M.; Eetemadi, S.; Pilehvar, M. T.; and Foster, J. 2025. F ar E x S tance: Explainable Stance Detection for F arsi. In Rambow, O.; Wanner, L.; Apidianaki, M.; Al-Khalifa, H.; Eugenio, B. D.; and Schockaert, S., eds., Proceedings of the ...

    2025

  34. [42]

    Zhang, T.; Ladhak, F.; Durmus, E.; Liang, P.; McKeown, K.; and Hashimoto, T. B. 2024. Benchmarking Large Language Models for News Summarization. Transactions of the Association for Computational Linguistics, 12: 39--57

    2024

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.