Pith. sign in

REVIEW 3 major objections 6 minor 41 references

"This Suits You the Best": Query Focused Comparative Explainable Summarization

T0 review · 3 major / 6 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 The paper claims that LLM prompts can evaluate comparative product summaries with 0.74 average Spearman correlation to human judges, and that a pre-summarization step cuts generation latency by about 40%.

desk verdict A well-scoped, honest empirical paper whose headline 0.74 Spearman is plausible but rests on a small, homogeneous human-judgment set; the task formulation and evaluation infrastructure are the real contributions. read the letter →

arxiv 2507.04733 v1 pith:34LRCEQN submitted 2025-07-07 cs.CL cs.IR

Arnav Attri , Anuj Attri , Pushpak Bhattacharyya , Suman Banerjee , Amey Patil , Muthusamy Chelliah , Nikesh Garera This is my paper
classification cs.CLcs.IR
keywords query-focusedsummarizationcomparativeopinionLLMevaluatione-commercerecommendationsexplainablerecommendationmulti-sourceSpearmancorrelation
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 proposes a new task for e-commerce: given a shopper's query and three recommended products, generate one tabular comparison with a final verdict that tells the shopper which product suits the query and why. To make this feasible, the authors introduce MS-Q2P, a dataset of 7,500 queries mapped to 22,500 products with metadata and reviews, and use LLMs to first condense each product's multiple sources into a Multi-Source Opinion Summary (M-OS), then build the comparative summary from those condensations. The paper's central quantitative claims are that their M-OS intermediate step cuts average inference time by about 40% compared to feeding raw data directly, and that their reference-free LLM evaluation prompts, QF-CES-PROMPT, achieve an average Spearman correlation of 0.74 with human ratings across five quality dimensions. A sympathetic reader would care because this offers a way to automate both the generation and the quality checking of comparative recommendation summaries without retraining or depending on a particular recommender engine.

What carries the argument

Key machinery: M-OS (Multi-Source Opinion Summarization), a condensed text built by an LLM from product title, description, key features, specifications, reviews, and rating, is the intermediate representation that makes QF-CES generation faster and less error-prone than processing raw data. QF-CES-PROMPT is the second load-bearing object: a set of dimension-specific evaluation prompts that ask an LLM to rate a summary on a 1-5 scale with explanation, with the final score computed as the probability-weighted average over roughly 100 samples. CES-EVAL is the benchmark that anchors these numbers: 50 queries, 10 summaries each, 3 expert raters, 5 dimensions, 7,500 ratings in total, with a two-round adjudication process that raises Krippendorff's alpha from 0.50 to 0.80. The summary-level correlation formula used in the paper turns evaluator-human agreement into the reported Spearman values.

What would settle it

Re-annotate the same 50 queries with a new, more diverse rater pool (different ages, genders, shopping backgrounds, and non-researchers), then recompute Spearman correlations between QF-CES-PROMPT and the new human scores; a drop well below 0.74, or a reversal of which model ranks best, would falsify the claim. Alternatively, draw a fresh sample of queries stratified across product categories and show that the evaluator-human agreement and the 40% latency gain do not reproduce.

Watch

Extended reading notes

Core claim

The central discovery claimed is that a prompt-only LLM pipeline can generate query-focused comparative summaries of three recommended products and, separately, that LLM-as-judge evaluation can stand in for human rating of those summaries. Concretely, the paper reports that LLaMA-3.1-70B-Instruct, when guided by the dimension-specific QF-CES-PROMPT prompts, reaches an average Spearman correlation of 0.74 with expert human judgments on clarity, faithfulness, informativeness, format adherence, and query relevance, the best among the five evaluators tested and better than GPT-4o. The authors also report that GPT-4 produces the highest-rated summaries, with Qwen2-7B-Instruct the strongest open-source generator, and that routing generation through M-OS instead of raw product data reduces average latency from 16.55 to 9.99 seconds per summary, about a 40% reduction. The paper positions these results as evidence that the full generate-then-evaluate loop is viable in a privacy-preserving, category-agnostic, recommendation-engine-agnostic setting.

Load-bearing premise

The load-bearing premise is that the CES-EVAL human ratings, collected from 50 sampled queries and three raters from a single research community, are a stable and unbiased ground truth for summary quality; if those ratings do not generalize, the reported 0.74 correlation and model rankings do not either.

Editorial extensions

If this is right

  • If the 0.74 correlation holds, QF-CES-PROMPT can replace expensive human rating for model selection and prompt tuning on comparative summaries.
  • If the 40% latency reduction holds, e-commerce platforms can serve comparative summaries at runtime rather than precomputing them for every query.
  • The approach's recommendation-engine agnosticism means it can be bolted onto any top-k ranking system without retraining.
  • The M-OS intermediate also appears to improve summary quality: Qwen2-7B with M-OS beat its direct-input variant on four of five dimensions, especially informativeness and query relevance.
  • The benchmark and prompts give other researchers a reusable protocol for evaluating comparative summarization.

Reading between the lines

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

  • The paper leaves untested whether 0.74 correlation persists on queries outside the 50 sampled, particularly for long-tail categories or non-electronics products; a natural extension is a stratified evaluation across categories.
  • Because the raters are three male researchers aged 24-32 from one community, the human ground truth itself may carry a narrow notion of clarity and relevance; re-running the annotation with diverse shoppers could shift both the correlation and the model rankings.
  • The M-OS latency gain suggests a broader design principle for LLM pipelines: condense per-input sources once, then generate per-query outputs, which could transfer to other multi-document generation tasks such as legal or medical comparisons.
  • A testable extension would use QF-CES-PROMPT to compare summaries produced by different recommender engines for the same query, checking whether evaluator rankings stay stable when the input product set changes.
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.

Referee Report

3 major / 6 minor

Summary. The paper introduces Query-Focused Comparative Explainable Summarization (QF-CES), a task in which an LLM produces a tabular comparison of top-k recommended products plus a natural-language final verdict that directly addresses a user query. The authors propose Multi-Source Opinion Summarization (M-OS) as a pre-processing step to compress product reviews and metadata before QF-CES generation, introduce the MS-Q2P dataset of 7,752 queries mapped to 23,256 products, and build CES-EVAL, a benchmark of human ratings on 10 model-generated summaries for 50 queries across five dimensions. They propose QF-CES-PROMPT, a set of dimension-specific evaluation prompts, and report that the best evaluator, LLaMA-3.1-70B-Instruct, achieves an average Spearman correlation of 0.74 with human judgments, while M-OS reduces inference latency by about 40% compared to direct raw-data processing. The paper also benchmarks nine open- and closed-source generators and five evaluator models.

Significance. If the reported evaluation correlation is robust, the paper makes a useful contribution: it defines a new summarization task that combines comparative structure with query-focused explanation, and it provides a reference-free evaluation protocol for that task. The empirical framing is non-circular: QF-CES-PROMPT is meta-evaluated against external human judgments rather than fitted to them, and the headline 0.74 is arithmetically consistent with the dimension-wise values in Table 3. The latency arithmetic in footnote 1 is also consistent with the claimed 40% reduction. The paper's main limitations are that the ground-truth benchmark rests on a small, homogeneous rater pool and 50 proprietary queries, that no uncertainty quantification is provided, and that neither the proprietary dataset nor the CES-EVAL annotations are released. These issues affect the generalizability of the central claim and should be addressed before publication.

major comments (3)
  1. [§4.2–4.3, Table 6] The central quantitative claim that QF-CES-PROMPT reaches an average Spearman correlation of 0.74 with human judgments is not yet supported with adequate robustness analysis. The benchmark rests on 50 queries from a proprietary dataset and three raters who are all male, aged 24–32, and from the same opinion-summarization research community. Round-I inter-rater agreement is moderate (Krippendorff alpha = 0.50), with pairwise Spearman correlations as low as 0.32–0.33 for informativeness and query relevance (Table 6), and Round-II alpha = 0.80 is achieved only after a discussion-based adjudication round. No confidence intervals, bootstrap estimates, or leave-one-rater-out analyses are reported, so the 0.74 average in Table 3 could shift substantially under independent re-annotation. Please add bootstrap confidence intervals for the per-query averaged Spearman correlations, a leave-one-rater-out analysis, and ideally an independent re-annotation of a subset by raters outside the authors' research community.
  2. [§4.1–4.3] Reproducibility of the main result is currently impossible: MS-Q2P is proprietary, CES-EVAL is not released, and the 50-query sample is not characterized. The abstract and contributions advertise CES-EVAL as a benchmark dataset, but the paper gives no access mechanism and no evidence that the 50 sampled queries span the category and query distribution of MS-Q2P. Without the annotations and the exact QF-CES-PROMPT definitions, readers cannot verify or extend the claimed 0.74 correlation. At minimum, release CES-EVAL (annotations plus the 50 queries, with appropriate privacy safeguards) and the full evaluation prompts, and report the category distribution and query-length statistics of the sampled subset.
  3. [§6, footnote 1] The 40% inference-latency reduction is presented in the abstract as a headline result, but the measurement is not end-to-end: the M-OS generation time is excluded from the comparison (Section 6, Time Efficiency Results), and the estimate is based on 50 queries with no variance or confidence interval. Footnote 1 does not mention this exclusion. Please either report end-to-end latency including M-OS pre-computation or qualify the abstract claim as a pre-computation scenario, and provide per-query latency distributions with confidence intervals.
minor comments (6)
  1. [Abstract vs. Table 1] The abstract and contributions state 7,500 queries and 22,500 products, while Table 1 reports 7,752 unique queries and 23,256 products; please reconcile these numbers.
  2. [§5.3 vs. Table 4] Section 5.3 lists Mixtral-8x22B-Instruct-v0.1 among the QF-CES generator models, but Table 4 reports results for Mixtral-8x7B-Instruct-v0.1; please correct the model name or clarify which model was actually evaluated.
  3. [§6 vs. Table 4] The QF-CES Results paragraph says GPT-4o excelled, but Table 4 and the rest of the paper report GPT-4 as the generator; please clarify whether the generator was GPT-4 or GPT-4o.
  4. [Appendices C and D] The captions for Figures 8 and 9 appear to be swapped: Figure 8 is described in Section 6 as showing LLM-as-evaluator scores, but its caption says 'rated by human annotators', while Figure 9's caption says 'LLM as Evaluators' but the text describes it as annotator ratings.
  5. [Limitations] In Limitations item 3, the dataset is referred to as 'M2-Q2P'; this should be 'MS-Q2P'.
  6. [§4.2] The CES-EVAL description says '10 model-generated summaries per product for 50 products', but the benchmark is defined over 50 queries, each with 10 summaries; please use 'per query' consistently.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the 0.74 correlation is a meta-evaluation against external human judgments, and the 40% latency figure is a direct wall-clock comparison.

full rationale

The paper's central quantitative claims are (1) that QF-CES-PROMPT evaluators reach an average Spearman correlation of 0.74 with human judgments across five dimensions, and (2) that using M-OS as an intermediate step reduces inference latency by roughly 40%. Both claims are self-contained empirical comparisons rather than derived-from-fitted-parameter predictions. The 0.74 figure is computed by correlating LLM evaluator scores (Table 3) with human annotations on the CES-EVAL benchmark; those human annotations are external ground truth, and no parameter is fitted from them to force the correlation. The latency figure is a measured comparison of generation times for M-OS versus DIA over 50 queries, averaged over 50 iterations, not a quantity implied by construction. The paper does cite prior work by overlapping authors, Siledar et al. (2024), when adapting the OPPROMPT framework to evaluate M-OS and select the M-OS generator, but that cited framework is a published, externally benchmarked method and is not the basis of the main evaluator-agreement or latency results. The Limitations and Ethical Considerations sections candidly acknowledge threats to external validity: only 50 proprietary queries, three raters all male aged 24-32 from the same research community, Round-I Krippendorff alpha of 0.50 rising to 0.80 only after discussion-based adjudication, and the need for a larger query set for the latency claim. Those are generalizability concerns, not circularity. No equation in the paper defines a predicted quantity in terms of the ground-truth labels it is tested against, and no fitted parameter is renamed as a prediction. Therefore no circular step can be exhibited, and the paper's central derivation is self-contained.

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

The paper's claims rest on empirical measurements rather than mathematical derivation. The main implicit assumptions are representativeness of the proprietary dataset, reliability of the human raters, and exclusion of M-OS generation time from the latency comparison. No fitted constants or invented physical entities appear.

assumptions (3)
  • domain assumption The proprietary MS-Q2P dataset is representative of real e-commerce queries and products.
    The dataset was provided by an e-commerce company, details are withheld, and all evaluations of QF-CES and M-OS depend on this data.
  • domain assumption Human raters' consensus scores provide a valid ground truth for summary quality.
    The 0.74 correlation claim rests on the reliability of three expert raters; inter-rater agreement is alpha=0.80 after discussion, but the rater group is homogeneous.
  • domain assumption The 50 queries in CES-EVAL are representative of the full MS-Q2P corpus.
    CES-EVAL samples 50 queries out of 7,752; no sampling strategy is described.

how reviews work

0 comments
Cite this review

Pith. "Pith review of "This Suits You the Best": Query Focused Comparative Explainable Summarization." pith.science (2026). https://pith.science/paper/34LRCEQN

@misc{pith2026250704733,
  author       = {Pith},
  title        = {Pith review of: "This Suits You the Best": Query Focused Comparative Explainable Summarization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/34LRCEQN}},
  note         = {Machine review of arXiv:2507.04733}
}
read the original abstract

Product recommendations inherently involve comparisons, yet traditional opinion summarization often fails to provide holistic comparative insights. We propose the novel task of generating Query-Focused Comparative Explainable Summaries (QF-CES) using Multi-Source Opinion Summarization (M-OS). To address the lack of query-focused recommendation datasets, we introduce MS-Q2P, comprising 7,500 queries mapped to 22,500 recommended products with metadata. We leverage Large Language Models (LLMs) to generate tabular comparative summaries with query-specific explanations. Our approach is personalized, privacy-preserving, recommendation engine-agnostic, and category-agnostic. M-OS as an intermediate step reduces inference latency approximately by 40% compared to the direct input approach (DIA), which processes raw data directly. We evaluate open-source and proprietary LLMs for generating and assessing QF-CES. Extensive evaluations using QF-CES-PROMPT across 5 dimensions (clarity, faithfulness, informativeness, format adherence, and query relevance) showed an average Spearman correlation of 0.74 with human judgments, indicating its potential for QF-CES evaluation.

Figures

Figures reproduced from arXiv: 2507.04733 by the authors.

Figure 1
Figure 1. QF-CES enables quick comparison of top-3 recommended products for confident decisions without tab-switching. Check [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Comparison of approaches: (A) Traditional opinion summaries, (B) M-OS, (C) Single-product views with tab navigation, and (D) textscQF-CES. Unlike traditional methods with isolated summaries, textscQF-CES offers side-by-side comparisons and a final verdict, eliminating tab-switching and enhancing decision-making confidence. eration tailored to user queries; (4) Category￾Agnostic approach applicable across product do￾… view at source ↗
Figure 3
Figure 3. A Multi-phase Pipeline for Generating QF-CES using M-OS and Large Language Models (LLMs). The pipeline involves using LLMs both as summary generators (LLMgen) and summary evaluators (LLMeval) to create and assess QF-CES across various dimensions, incorporating product details from the MS-Q2P dataset. The points 1 through 7 describe the flow of inputs and outputs between the LLMs, from generating M-OS to evaluating Q… view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Ratings Distribution. We plot the average frequency of scores obtained by human raters across 5 dimensions. A score of 4 or 5 is mostly preferred [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 5
Figure 5. Figure 5: QF-CES generated by Qwen2-7B-instruct 5 Experiments We present the generation and evaluation of M￾OS for QF-CES, using LLMs as baseline metrics, followed by implementation details. 5.1 M-OS Models We use a custom prompt for the LLMs to generate M-OS. These models were …
Figure 6
Figure 6. Figure 6: Comparison of inference times for QF-CES generation using M-OS and DIA approach. Each data point represents the average of 50 generations per query. 7 Conclusion & Future Work This paper introduces Query-Focused Compara￾tive Explainable Summarization QF-CES, a novel ta…
Figure 7
Figure 7. Figure 7: represents the model-wise performance across 7 dimensions: fluency (FL), coherence (CO), aspect coverage (AC), faithfulness (FF), relevance (RL), sentiment consistency (SC), and specificity (SP). The scores are given by LLaMA-3.1.70B-Instruct as evaluator, for n=100 ge…
Figure 8
Figure 8. Figure 8: represents model-wise averaged score given by various LLM as evaluators of QF￾CES along 5 dimensions: clarity (CL), faithfulness (FA), informativeness (IF), format adherence (FoA) and query relevance (QR) [PITH_FULL_IMAGE:figures/full_fig_p013_8.png]
Figure 9
Figure 9. Figure 9: represents model-wise averaged annotator ratings of QF-CES along 5 dimensions: clarity (CL), faithfulness (FA), informativeness (IF), format adherence (FoA) and query relevance (QR) [PITH_FULL_IMAGE:figures/full_fig_p013_9.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

41 extracted references · 11 canonical work pages

  1. [1]

    AI@Meta. 2024. https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md Llama 3 model card

    2024

  2. [2]

    Manik Bhandari, Pranav Narayan Gour, Atabak Ashfaq, Pengfei Liu, and Graham Neubig. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.751 Re-evaluating evaluation in text summarization . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 9347--9359, Online. Association for Computational Linguistics

    doi:10.18653/v1/2020.emnlp-main.751 2020

  3. [3]

    Chong Chen, Min Zhang, Yiqun Liu, and Shaoping Ma. 2018. https://doi.org/10.1145/3178876.3186070 Neural attentional rating regression with review-level explanations . In Proceedings of the 2018 World Wide Web Conference, WWW '18, page 1583–1592, Republic and Canton of Geneva, CHE. International World Wide Web Conferences Steering Committee

    arXiv 2018

  4. [5]

    Cheng-Han Chiang and Hung-yi Lee. 2023 b . https://doi.org/10.18653/v1/2023.acl-long.870 Can large language models be an alternative to human evaluations? In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15607--15631, Toronto, Canada. Association for Computational Linguistics

    doi:10.18653/v1/2023.acl-long.870 2023

  5. [6]

    Cheng-Han Chiang and Hung-yi Lee. 2023 c . https://doi.org/10.18653/v1/2023.findings-emnlp.599 A closer look into using large language models for automatic evaluation . In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 8928--8942, Singapore. Association for Computational Linguistics

    doi:10.18653/v1/2023.findings-emnlp.599 2023

  6. [7]

    Gonzalez, Ion Stoica, and Eric P

    Wei-Lin Chiang, Zhuohan Li, Zi Lin, Ying Sheng, Zhanghao Wu, Hao Zhang, Lianmin Zheng, Siyuan Zhuang, Yonghao Zhuang, Joseph E. Gonzalez, Ion Stoica, and Eric P. Xing. 2023. https://lmsys.org/blog/2023-03-30-vicuna/ Vicuna: An open-source chatbot impressing gpt-4 with 90\

    2023

  7. [8]

    Anthony Colas, Jun Araki, Zhengyu Zhou, Bingqing Wang, and Zhe Feng. 2023. https://doi.org/10.18653/v1/2023.blackboxnlp-1.1 Knowledge-grounded natural language recommendation explanation . In Proceedings of the 6th BlackboxNLP Workshop: Analyzing and Interpreting Neural Networks for NLP, pages 1--15, Singapore. Association for Computational Linguistics

    work page doi:10.18653/v1/2023.blackboxnlp-1.1 2023

  8. [9]

    Jessica Maria Echterhoff, An Yan, and Julian McAuley. 2023. https://api.semanticscholar.org/CorpusID:259375808 Comparing apples to apples: Generating aspect-aware comparative sentences from user reviews . ArXiv, abs/2307.03691

    work page Pith review arXiv 2023

Show all 41 references
  1. [10]

    Fabbri, Wojciech Kry \'s ci \'n ski, Bryan McCann, Caiming Xiong, Richard Socher, and Dragomir Radev

    Alexander R. Fabbri, Wojciech Kry \'s ci \'n ski, Bryan McCann, Caiming Xiong, Richard Socher, and Dragomir Radev. 2021. https://doi.org/10.1162/tacl_a_00373 S umm E val: Re-evaluating summarization evaluation . Transactions of the Association for Computational Linguistics, 9:391--409

    2021 doi

  2. [11]

    Jinlan Fu, See-Kiong Ng, Zhengbao Jiang, and Pengfei Liu. 2023. http://arxiv.org/abs/2302.04166 Gptscore: Evaluate as you desire

    2023 arXiv

  3. [12]

    Shen Gao, Yifan Wang, Jiabao Fang, Lisi Chen, Peng Han, and Shuo Shang. 2024. http://arxiv.org/abs/2404.06311 Dre: Generating recommendation explanations by aligning large language models at data-level

    2024 arXiv

  4. [13]

    Dan Gillick and Yang Liu. 2010. https://aclanthology.org/W10-0722 Non-expert evaluation of summarization systems is risky . In Proceedings of the NAACL HLT 2010 Workshop on Creating Speech and Language Data with A mazon ' s Mechanical Turk , pages 148--151, Los Angeles. Associ...

    2010

  5. [14]

    Jinbae Im, Moonki Kim, Hoyeop Lee, Hyunsouk Cho, and Sehee Chung. 2021. https://doi.org/10.18653/v1/2021.acl-long.33 Self-supervised multimodal opinion summarization . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th Inter...

    2021 doi

  6. [15]

    Hayate Iso, Xiaolan Wang, Stefanos Angelidis, and Yoshihiko Suhara. 2022. C omparative O pinion S ummarization via C ollaborative D ecoding. In Findings of the Association for Computational Linguistics (ACL)

    2022

  7. [16]

    Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Florian Bressand, Gianna Lengyel, Guillaume Lample, Lucile Saulnier, Lélio Renard Lavaud, Marie-Anne Lachaux, Pierre Stock, Teven Le Scao, Thibaut Lavril, Thomas ...

    2023 arXiv

  8. [17]

    Albert Q. Jiang, Alexandre Sablayrolles, Antoine Roux, Arthur Mensch, Blanche Savary, Chris Bamford, Devendra Singh Chaplot, Diego de las Casas, Emma Bou Hanna, Florian Bressand, Gianna Lengyel, Guillaume Bour, Guillaume Lample, Lélio Renard Lavaud, Lucile Saulnier, Marie-Anne...

    2024 arXiv

  9. [18]

    Tom Kocmi and Christian Federmann. 2023. https://aclanthology.org/2023.eamt-1.19 Large language models are state-of-the-art evaluators of translation quality . In Proceedings of the 24th Annual Conference of the European Association for Machine Translation, pages 193--203, Tam...

    2023

  10. [19]

    Klaus Krippendorff. 2011. https://api.semanticscholar.org/CorpusID:59901023 Computing krippendorff's alpha-reliability

    2011

  11. [21]

    Haoran Li, Peng Yuan, Song Xu, Youzheng Wu, Xiaodong He, and Bowen Zhou. 2020 a . https://doi.org/10.1609/aaai.v34i05.6332 Aspect-aware multimodal summarization for chinese e-commerce products . Proceedings of the AAAI Conference on Artificial Intelligence, 34(05):8188--8195

    2020 doi

  12. [23]

    Lei Li, Yongfeng Zhang, and Li Chen. 2020 c . https://doi.org/10.1145/3340531.3411992 Generate neural template explanations for recommendation . In Proceedings of the 29th ACM International Conference on Information & Knowledge Management, CIKM '20, page 755–764, New York, NY,...

    2020

  13. [24]

    Lei Li, Yongfeng Zhang, and Li Chen. 2021. https://doi.org/10.18653/v1/2021.acl-long.383 Personalized transformer for explainable recommendation . In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Confer...

    2021 doi

  14. [25]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013 ROUGE : A package for automatic evaluation of summaries . In Text Summarization Branches Out, pages 74--81, Barcelona, Spain. Association for Computational Linguistics

    2004

  15. [26]

    Yang Liu, Dan Iter, Yichong Xu, Shuohang Wang, Ruochen Xu, and Chenguang Zhu. 2023. https://doi.org/10.18653/v1/2023.emnlp-main.153 G -eval: NLG evaluation using gpt-4 with better human alignment . In Proceedings of the 2023 Conference on Empirical Methods in Natural Language ...

    2023 doi

  16. [27]

    Jianmo Ni, Jiacheng Li, and Julian McAuley. 2019. https://doi.org/10.18653/v1/D19-1018 Justifying recommendations using distantly-labeled reviews and fine-grained aspects . In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th In...

    2019 doi

  17. [28]

    OpenAI. 2023. https://arxiv.org/abs/2303.08774 Gpt-4 technical report . ArXiv, abs/2303.08774

    2023 arXiv

  18. [29]

    Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. https://doi.org/10.3115/1073083.1073135 B leu: a method for automatic evaluation of machine translation . In Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pages 311--3...

    2002

  19. [30]

    Yicui Peng, Hao Chen, Chingsheng Lin, Guo Huang, Jinrong Hu, Hui Guo, Bin Kong, Shu Hu, Xi Wu, and Xin Wang. 2024. http://arxiv.org/abs/2402.03366 Uncertainty-aware explainable recommendation with large language models

    2024 arXiv

  20. [31]

    Tejpalsingh Siledar, Jigar Makwana, and Pushpak Bhattacharyya. 2023. https://doi.org/10.1145/3570991.3571035 Aspect-sentiment-based opinion summarization using multiple information sources . In Proceedings of the 6th Joint International Conference on Data Science & Management ...

    2023

  21. [32]

    Tejpalsingh Siledar, Swaroop Nath, Sankara Sri Raghava Ravindra Muddu, Rupasai Rangaraju, Swaprava Nath, Pushpak Bhattacharyya, Suman Banerjee, Amey Patil, Sudhanshu Shekhar Singh, Muthusamy Chelliah, and Nikesh Garera. 2024. http://arxiv.org/abs/2402.11683 One prompt to rule ...

    2024 arXiv

  22. [33]

    Juntao Tan, Shuyuan Xu, Yingqiang Ge, Yunqi Li, Xu Chen, and Yongfeng Zhang. 2021. https://doi.org/10.1145/3459637.3482420 Counterfactual explainable recommendation . In Proceedings of the 30th ACM International Conference on Information & Knowledge Management, CIKM '21, page ...

    2021

  23. [34]

    Wenyi Tay. 2019. https://doi.org/10.18653/v1/P19-2005 Not all reviews are equal: Towards addressing reviewer biases for opinion summarization . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics: Student Research Workshop, pages 34--42, ...

    2019 doi

  24. [35]

    Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivière, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, Léonard Hussenot, Pier Giuseppe Sessa, Aakanksha Chowdhery, Adam Roberts, Aditya Barua, Alex Botev, Alex...

    2024 arXiv

  25. [36]

    Rush, and Thomas Wolf

    Lewis Tunstall, Edward Beeching, Nathan Lambert, Nazneen Rajani, Kashif Rasul, Younes Belkada, Shengyi Huang, Leandro von Werra, Clémentine Fourrier, Nathan Habib, Nathan Sarrazin, Omar Sanseviero, Alexander M. Rush, and Thomas Wolf. 2023. http://arxiv.org/abs/2310.16944 Zephy...

    2023 arXiv

  26. [37]

    Jiaan Wang, Yunlong Liang, Fandong Meng, Zengkui Sun, Haoxiang Shi, Zhixu Li, Jinan Xu, Jianfeng Qu, and Jie Zhou. 2023 a . https://doi.org/10.18653/v1/2023.newsum-1.1 Is C hat GPT a good NLG evaluator? a preliminary study . In Proceedings of the 4th New Frontiers in Summariza...

    2023 doi

  27. [38]

    Nan Wang, Hongning Wang, Yiling Jia, and Yue Yin. 2018. https://doi.org/10.1145/3209978.3210010 Explainable recommendation via multi-task learning in opinionated text data . In The 41st International ACM SIGIR Conference on Research & Development in Information Retrieval, SIGI...

    2018

  28. [39]

    Yequan Wang, Hengran Zhang, Aixin Sun, and Xuying Meng. 2023 b . https://api.semanticscholar.org/CorpusID:257532399 Gcre-gpt: A generative model for comparative relation extraction . ArXiv, abs/2303.08601

    2023 arXiv

  29. [40]

    Jason Wei, Xuezhi Wang, Dale Schuurmans, Maarten Bosma, Brian Ichter, Fei Xia, Ed Chi, Quoc Le, and Denny Zhou. 2023. http://arxiv.org/abs/2201.11903 Chain-of-thought prompting elicits reasoning in large language models

    2023 arXiv

  30. [41]

    Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, Remi Louf, Morgan Funtowicz, Joe Davison, Sam Shleifer, Patrick von Platen, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mari...

    2020 doi

  31. [42]

    An Yang, Baosong Yang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Zhou, Chengpeng Li, Chengyuan Li, Dayiheng Liu, Fei Huang, Guanting Dong, Haoran Wei, Huan Lin, Jialong Tang, Jialin Wang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Ma, Jianxin Yang, Jin Xu, Jingren Zhou, Jinze...

    2024 arXiv

  32. [43]

    Aobo Yang, Nan Wang, Renqin Cai, Hongbo Deng, and Hongning Wang. 2022. https://doi.org/10.1145/3485447.3512031 Comparative explanations of recommendations . In Proceedings of the ACM Web Conference 2022. ACM

    2022

  33. [44]

    Aobo Yang, Nan Wang, Hongbo Deng, and Hongning Wang. 2021. https://doi.org/10.1145/3437963.3441726 Explanation as a defense of recommendation . In Proceedings of the 14th ACM International Conference on Web Search and Data Mining, WSDM '21, page 1029–1037, New York, NY, USA. A...

    2021

Pith tools

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