Ranking sentences from product description & bullets for better search

Aliasgar Kutiyanawala; Ke Shen; Prateek Verma

arxiv: 1907.06330 · v1 · pith:TZYMN6XAnew · submitted 2019-07-15 · 💻 cs.IR · cs.CL· cs.LG

Ranking sentences from product description & bullets for better search

Prateek Verma , Aliasgar Kutiyanawala , Ke Shen This is my paper

Pith reviewed 2026-05-24 21:38 UTC · model grok-4.3

classification 💻 cs.IR cs.CLcs.LG

keywords sentence rankingextractive summarizationreinforcement learningproduct descriptionse-commerce searchclick through logsinformation retrieval

0 comments

The pith

Reinforcement learning models rank sentences in product descriptions and bullets by using titles and click logs to improve search relevance.

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

Verbose product descriptions and bullet points often include information irrelevant to search, which hurts full-text matching and attribute extraction. The paper develops two extractive summarization methods that use reinforcement learning to rank these sentences. These methods draw on product titles and search click-through logs as signals for what matters from a search viewpoint. If effective, this selective ranking would allow systems to focus on pertinent content rather than treating every sentence equally. The authors compare the accuracy of the two approaches.

Core claim

The paper establishes that two reinforcement learning approaches to extractive summarization, leveraging product titles and search click-through logs, can rank sentences from product descriptions and bullets according to their relevance from a search perspective, thereby mitigating the problems caused by verbose fields in full text matching and entity extraction.

What carries the argument

Extractive summarization with reinforcement learning that scores sentences using title and click log information.

If this is right

Better full text search matches by excluding irrelevant sentences.
Improved accuracy in NER-based attribute extraction from product fields.
Facilitates better ontology development and semantic search in e-commerce.
The two models can be compared directly on accuracy metrics.

Where Pith is reading between the lines

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

This ranking could extend to other product fields like customer reviews to enhance search.
Dependence on click logs may favor popular items and overlook niche products.
Integration with additional signals beyond titles and clicks might further refine the rankings.

Load-bearing premise

Product titles and search click-through logs supply reliable, unbiased signals for determining which sentences are relevant from a search perspective.

What would settle it

Running the ranked sentences through a search system and measuring no gain in full-text match quality or attribute extraction precision compared to using unranked verbose fields.

Figures

Figures reproduced from arXiv: 1907.06330 by Aliasgar Kutiyanawala, Ke Shen, Prateek Verma.

**Figure 1.** Figure 1: which shows a sample item and some information associated in bullet form from an ecommerce website. e second bullet contains the terms "soups", "casseroles" and "meat" because of which, the item (mushroom) will be present in the recall set for the search queries containing tokens like "soups" and "casseroles" due to full text match, leading to poor search relevancy. Relevant features for this SKU can be … view at source ↗

**Figure 2.** Figure 2: Sample SKU Description aribute extraction from the catalog data is oen done in order to enrich SKUs (documents) with relevant aributes. In this paper we describe a method to rank sentences based on if they are relevant from search perspective, and select top K sentences for search from these elds. Top K ranked set of sentences can lead to beer full text match and can also help in extracting aributes… view at source ↗

**Figure 3.** Figure 3: Network architecture [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Retrieval by matching query understanding with [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗

**Figure 5.** Figure 5: Precision @ k for the baseline common practice to adjust the denominator to 1 + |{d ∈ D : t ∈ d}|. For baseline, we use tdf based model. Our baseline consists of three aproaches that utilizes tdf to score the sentences to select top K. For the rst approach, we sum up (unweighted) tdf score of the words to measure importance of a sentence and then select top K as the summary. Here, tf is computed at the… view at source ↗

**Figure 7.** Figure 7: Input to the model: product description [PITH_FULL_IMAGE:figures/full_fig_p006_7.png] view at source ↗

**Figure 8.** Figure 8: Model 1’s output (title) [PITH_FULL_IMAGE:figures/full_fig_p006_8.png] view at source ↗

read the original abstract

Products in an ecommerce catalog contain information-rich fields like description and bullets that can be useful to extract entities (attributes) using NER based systems. However, these fields are often verbose and contain lot of information that is not relevant from a search perspective. Treating each sentence within these fields equally can lead to poor full text match and introduce problems in extracting attributes to develop ontologies, semantic search etc. To address this issue, we describe two methods based on extractive summarization with reinforcement learning by leveraging information in product titles and search click through logs to rank sentences from bullets, description, etc. Finally, we compare the accuracy of these two models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper sketches two RL extractive methods to rank product sentences using titles and click logs, but the abstract gives no results, baselines, or bias handling, leaving the claims untestable.

read the letter

The main takeaway is that this paper takes established reinforcement learning for extractive summarization and points it at ranking sentences inside e-commerce product descriptions and bullets. It uses product titles and raw search click logs as the supervision signal, with the goal of dropping sentences that do not help full-text match or attribute extraction. That specific application to catalog text is the only real novelty; the underlying RL setup is routine extension of prior work on summarization rewards. The problem statement itself is clear and practical—verbose fields do dilute search signals and complicate NER-based attribute work. The paper earns credit for keeping the approach extractive and for naming the two external signals it plans to leverage. Beyond that framing, there is little else shown. The abstract states that accuracy is compared between the two models, yet it supplies no metrics, no datasets, no baselines, and no description of the reward function or training procedure. This absence makes any performance claim impossible to evaluate. The stress-test concern about click logs holds up on the given text: position bias and popularity bias are standard in raw CTR data, and nothing in the abstract indicates inverse-propensity scoring, examination models, or any other correction. Without that step the policy would optimize for observed clicks rather than search relevance, which undercuts the central promise. The methods appear to avoid self-referential fitting and instead rely on external logs, which is a positive design choice, but it does not compensate for the missing evidence. This work would interest only practitioners at e-commerce shops who already have click data and want a starting recipe to try. A reader seeking validated improvements or careful handling of supervision bias would find nothing to use. Because the claims rest on unshown experiments and an unaddressed bias issue, the paper does not merit peer review in its current state.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes two methods for ranking sentences within verbose product description and bullet fields using extractive summarization trained via reinforcement learning. The methods leverage signals from product titles and raw search click-through logs as supervision to prioritize search-relevant content, with the stated goal of improving full-text match quality and downstream attribute extraction for ontologies and semantic search. The abstract concludes that the accuracy of the two models is compared.

Significance. If the experimental validation were present and the click-log supervision were shown to be reliable, the approach could offer a practical technique for filtering irrelevant sentences in e-commerce catalogs, thereby improving search precision and entity extraction pipelines. The core idea of using RL for extractive ranking in this domain is a plausible extension of prior summarization work, but its value hinges entirely on the missing empirical support.

major comments (2)

[Abstract] Abstract: the claim that 'we compare the accuracy of these two models' is unsupported; the manuscript supplies no datasets, metrics, baselines, experimental setup, or numerical results, rendering the central performance claim unevaluable.
[Abstract] Abstract (methods description): the reward signal is derived directly from raw search click-through logs without any described debiasing step (e.g., inverse propensity scoring or examination model) for position or popularity bias; this directly undermines the claim that the ranked sentences improve true search relevance rather than observed click rates.

minor comments (1)

The two methods are referenced but never distinguished in the provided text; explicit algorithmic or architectural differences should be stated.

Simulated Author's Rebuttal

2 responses · 1 unresolved

We thank the referee for the comments. We address each major point below.

read point-by-point responses

Referee: [Abstract] Abstract: the claim that 'we compare the accuracy of these two models' is unsupported; the manuscript supplies no datasets, metrics, baselines, experimental setup, or numerical results, rendering the central performance claim unevaluable.

Authors: We agree. The manuscript text consists solely of the abstract and contains no experimental section, datasets, metrics, baselines, or results. The abstract claim is therefore unsupported. We will revise the abstract to remove the sentence stating that we compare the accuracy of the two models. revision: yes
Referee: [Abstract] Abstract (methods description): the reward signal is derived directly from raw search click-through logs without any described debiasing step (e.g., inverse propensity scoring or examination model) for position or popularity bias; this directly undermines the claim that the ranked sentences improve true search relevance rather than observed click rates.

Authors: This observation is correct. The manuscript describes no debiasing procedure for the raw click logs. We will revise the text to explicitly acknowledge that the reward signal may reflect position/popularity biases rather than true relevance and to note the absence of techniques such as inverse propensity scoring. revision: partial

standing simulated objections not resolved

Absence of any experimental results, datasets, or evaluation details in the manuscript, which prevents supplying the requested empirical validation.

Circularity Check

0 steps flagged

No circularity; methods use external logs and titles as independent supervision signals

full rationale

The paper describes two RL-based extractive summarization methods that take product titles and raw search click-through logs as external inputs to produce sentence rankings. No equations, derivations, or self-citations are presented that reduce the claimed output to a fitted parameter or renamed input by construction. The central approach is an empirical pipeline whose success depends on the quality of the external logs rather than any self-referential loop. This is the normal case of a non-circular applied ML paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only abstract available; no free parameters, axioms, or invented entities are described.

pith-pipeline@v0.9.0 · 5634 in / 967 out tokens · 48429 ms · 2026-05-24T21:38:21.534359+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages · 16 internal anchors

[1]

Mehdi Allahyari, Seyedamin Pouriyeh, Mehdi Asse/f_i, Saeid Safaei, Elizabeth D Trippe, Juan B Gutierrez, and Krys Kochut. 2017. Text summarization techniques: a brief survey. arXiv preprint arXiv:1707.02268 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017
[2]

Qian Chen, Xiaodan Zhu, Zhenhua Ling, Si Wei, and Hui Jiang. 2016. Distraction-based neural networks for document summarization. arXiv preprint arXiv:1610.08462 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016
[3]

Jianpeng Cheng and Mirella Lapata. 2016. Neural summarization by extracting sentences and words. arXiv preprint arXiv:1603.07252 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016
[4]

Ronan Collobert, Jason Weston, L ´eon Bo/t_tou, Michael Karlen, Koray Kavukcuoglu, and Pavel Kuksa. 2011. Natural language processing (almost) from scratch. Journal of machine learning research 12, Aug (2011), 2493–2537

work page 2011
[5]

G¨unes Erkan and Dragomir R Radev. 2004. Lexrank: Graph-based lexical cen- trality as salience in text summarization. Journal of arti/f_icial intelligence research 22 (2004), 457–479

work page 2004
[6]

Clinton Gormley and Zachary Tong. 2015. Elasticsearch: /T_he De/f_initive Guide: A Distributed Real-Time Search and Analytics Engine . ” O’Reilly Media, Inc. ”

work page 2015
[7]

2014.Solr in action

Trey Grainger, Timothy Po/t_ter, and Yonik Seeley. 2014.Solr in action. Manning Cherry Hill

work page 2014
[8]

Nal Kalchbrenner, Edward Grefenste/t_te, and Phil Blunsom. 2014. A convolutional neural network for modelling sentences. arXiv preprint arXiv:1404.2188 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[9]

Yoon Kim. 2014. Convolutional neural networks for sentence classi/f_ication.arXiv preprint arXiv:1408.5882 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014
[10]

Aliasgar Kutiyanawala, Prateek Verma, and Zheng Yan. 2018. Towards a sim- pli/f_ied ontology for be/t_ter e-commerce search.CoRR abs/1807.02039 (2018). arXiv:1807.02039 h/t_tp://arxiv.org/abs/1807.02039

work page internal anchor Pith review Pith/arXiv arXiv 2018
[11]

Jiwei Li, Minh-/T_hang Luong, and Dan Jurafsky. 2015. A hierarchical neural autoencoder for paragraphs and documents. arXiv preprint arXiv:1506.01057 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015
[12]

Jiwei Li, Will Monroe, Alan Ri/t_ter, Michel Galley, Jianfeng Gao, and Dan Jurafsky

work page
[13]

Deep Reinforcement Learning for Dialogue Generation

Deep reinforcement learning for dialogue generation. arXiv preprint arXiv:1606.01541 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016
[14]

Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. Text Summarization Branches Out (2004)

work page 2004
[15]

Ramesh Nallapati, Feifei Zhai, and Bowen Zhou. 2017. Summarunner: A re- current neural network based sequence model for extractive summarization of documents. In /T_hirty-First AAAI Conference on Arti/f_icial Intelligence

work page 2017
[16]

Ramesh Nallapati, Bowen Zhou, Caglar Gulcehre, Bing Xiang, et al. 2016. Ab- stractive text summarization using sequence-to-sequence rnns and beyond.arXiv preprint arXiv:1602.06023 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016
[17]

Shashi Narayan, Shay B Cohen, and Mirella Lapata. 2018. Ranking sentences for extractive summarization with reinforcement learning. arXiv preprint arXiv:1802.08636 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018
[18]

Shashi Narayan, Nikos Papasarantopoulos, Shay B Cohen, and Mirella Lapata

work page
[19]

Neural Extractive Summarization with Side Information

Neural extractive summarization with side information. arXiv preprint arXiv:1704.04530 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017
[20]

Ani Nenkova, Lucy Vanderwende, and Kathleen McKeown. 2006. A compo- sitional context sensitive multi-document summarizer: exploring the factors that in/f_luence summarization. InProceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval. ACM, 573–580

work page 2006
[21]

Joel Larocca Neto, Alexandre D Santos, Celso AA Kaestner, Neto Alexandre, D Santos, et al. 2000. Document clustering and text summarization. (2000)

work page 2000
[22]

Romain Paulus, Caiming Xiong, and Richard Socher. 2017. A deep reinforced model for abstractive summarization. arXiv preprint arXiv:1705.04304 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017
[23]

Dragomir R Radev, Timothy Allison, Sasha Blair-Goldensohn, John Blitzer, Arda Celebi, Stanko Dimitrov, Ellio/t_t Drabek, Ali Hakim, Wai Lam, Danyu Liu, et al

work page
[24]

MEAD-a platform for multidocument multilingual text summarization. (2004)

work page 2004
[25]

Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba

work page
[26]

Sequence Level Training with Recurrent Neural Networks

Sequence level training with recurrent neural networks. arXiv preprint arXiv:1511.06732 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015
[27]

Alexander M Rush, Sumit Chopra, and Jason Weston. 2015. A neural a/t_tention model for abstractive sentence summarization. arXiv preprint arXiv:1509.00685 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015
[28]

Abigail See, Peter J Liu, and Christopher D Manning. 2017. Get to the point: Summarization with pointer-generator networks. arXiv preprint arXiv:1704.04368 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017
[29]

Ilya Sutskever, Oriol Vinyals, and /Q_uoc V Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing systems. 3104–3112

work page 2014
[30]

Ryen W White, Ian Ruthven, and Joemon M Jose. 2002. Finding relevant docu- ments using top ranking sentences: an evaluation of two alternative schemes. In Proceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval . ACM, 57–64

work page 2002
[31]

Ronald J Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine learning 8, 3-4 (1992), 229–256. Ranking sentences from product description & bullets for be/t_ter search SIGIR 2019 eCom, July 2019, Paris, France

work page 1992
[32]

Michihiro Yasunaga, Rui Zhang, Kshitijh Meelu, Ayush Pareek, Krishnan Srini- vasan, and Dragomir Radev. 2017. Graph-based neural multi-document summa- rization. arXiv preprint arXiv:1706.06681 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017

[1] [1]

Mehdi Allahyari, Seyedamin Pouriyeh, Mehdi Asse/f_i, Saeid Safaei, Elizabeth D Trippe, Juan B Gutierrez, and Krys Kochut. 2017. Text summarization techniques: a brief survey. arXiv preprint arXiv:1707.02268 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017

[2] [2]

Qian Chen, Xiaodan Zhu, Zhenhua Ling, Si Wei, and Hui Jiang. 2016. Distraction-based neural networks for document summarization. arXiv preprint arXiv:1610.08462 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016

[3] [3]

Jianpeng Cheng and Mirella Lapata. 2016. Neural summarization by extracting sentences and words. arXiv preprint arXiv:1603.07252 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016

[4] [4]

Ronan Collobert, Jason Weston, L ´eon Bo/t_tou, Michael Karlen, Koray Kavukcuoglu, and Pavel Kuksa. 2011. Natural language processing (almost) from scratch. Journal of machine learning research 12, Aug (2011), 2493–2537

work page 2011

[5] [5]

G¨unes Erkan and Dragomir R Radev. 2004. Lexrank: Graph-based lexical cen- trality as salience in text summarization. Journal of arti/f_icial intelligence research 22 (2004), 457–479

work page 2004

[6] [6]

Clinton Gormley and Zachary Tong. 2015. Elasticsearch: /T_he De/f_initive Guide: A Distributed Real-Time Search and Analytics Engine . ” O’Reilly Media, Inc. ”

work page 2015

[7] [7]

2014.Solr in action

Trey Grainger, Timothy Po/t_ter, and Yonik Seeley. 2014.Solr in action. Manning Cherry Hill

work page 2014

[8] [8]

Nal Kalchbrenner, Edward Grefenste/t_te, and Phil Blunsom. 2014. A convolutional neural network for modelling sentences. arXiv preprint arXiv:1404.2188 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014

[9] [9]

Yoon Kim. 2014. Convolutional neural networks for sentence classi/f_ication.arXiv preprint arXiv:1408.5882 (2014)

work page internal anchor Pith review Pith/arXiv arXiv 2014

[10] [10]

Aliasgar Kutiyanawala, Prateek Verma, and Zheng Yan. 2018. Towards a sim- pli/f_ied ontology for be/t_ter e-commerce search.CoRR abs/1807.02039 (2018). arXiv:1807.02039 h/t_tp://arxiv.org/abs/1807.02039

work page internal anchor Pith review Pith/arXiv arXiv 2018

[11] [11]

Jiwei Li, Minh-/T_hang Luong, and Dan Jurafsky. 2015. A hierarchical neural autoencoder for paragraphs and documents. arXiv preprint arXiv:1506.01057 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015

[12] [12]

Jiwei Li, Will Monroe, Alan Ri/t_ter, Michel Galley, Jianfeng Gao, and Dan Jurafsky

work page

[13] [13]

Deep Reinforcement Learning for Dialogue Generation

Deep reinforcement learning for dialogue generation. arXiv preprint arXiv:1606.01541 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016

[14] [14]

Chin-Yew Lin. 2004. Rouge: A package for automatic evaluation of summaries. Text Summarization Branches Out (2004)

work page 2004

[15] [15]

Ramesh Nallapati, Feifei Zhai, and Bowen Zhou. 2017. Summarunner: A re- current neural network based sequence model for extractive summarization of documents. In /T_hirty-First AAAI Conference on Arti/f_icial Intelligence

work page 2017

[16] [16]

Ramesh Nallapati, Bowen Zhou, Caglar Gulcehre, Bing Xiang, et al. 2016. Ab- stractive text summarization using sequence-to-sequence rnns and beyond.arXiv preprint arXiv:1602.06023 (2016)

work page internal anchor Pith review Pith/arXiv arXiv 2016

[17] [17]

Shashi Narayan, Shay B Cohen, and Mirella Lapata. 2018. Ranking sentences for extractive summarization with reinforcement learning. arXiv preprint arXiv:1802.08636 (2018)

work page internal anchor Pith review Pith/arXiv arXiv 2018

[18] [18]

Shashi Narayan, Nikos Papasarantopoulos, Shay B Cohen, and Mirella Lapata

work page

[19] [19]

Neural Extractive Summarization with Side Information

Neural extractive summarization with side information. arXiv preprint arXiv:1704.04530 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017

[20] [20]

Ani Nenkova, Lucy Vanderwende, and Kathleen McKeown. 2006. A compo- sitional context sensitive multi-document summarizer: exploring the factors that in/f_luence summarization. InProceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval. ACM, 573–580

work page 2006

[21] [21]

Joel Larocca Neto, Alexandre D Santos, Celso AA Kaestner, Neto Alexandre, D Santos, et al. 2000. Document clustering and text summarization. (2000)

work page 2000

[22] [22]

Romain Paulus, Caiming Xiong, and Richard Socher. 2017. A deep reinforced model for abstractive summarization. arXiv preprint arXiv:1705.04304 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017

[23] [23]

Dragomir R Radev, Timothy Allison, Sasha Blair-Goldensohn, John Blitzer, Arda Celebi, Stanko Dimitrov, Ellio/t_t Drabek, Ali Hakim, Wai Lam, Danyu Liu, et al

work page

[24] [24]

MEAD-a platform for multidocument multilingual text summarization. (2004)

work page 2004

[25] [25]

Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba

work page

[26] [26]

Sequence Level Training with Recurrent Neural Networks

Sequence level training with recurrent neural networks. arXiv preprint arXiv:1511.06732 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015

[27] [27]

Alexander M Rush, Sumit Chopra, and Jason Weston. 2015. A neural a/t_tention model for abstractive sentence summarization. arXiv preprint arXiv:1509.00685 (2015)

work page internal anchor Pith review Pith/arXiv arXiv 2015

[28] [28]

Abigail See, Peter J Liu, and Christopher D Manning. 2017. Get to the point: Summarization with pointer-generator networks. arXiv preprint arXiv:1704.04368 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017

[29] [29]

Ilya Sutskever, Oriol Vinyals, and /Q_uoc V Le. 2014. Sequence to sequence learning with neural networks. In Advances in neural information processing systems. 3104–3112

work page 2014

[30] [30]

Ryen W White, Ian Ruthven, and Joemon M Jose. 2002. Finding relevant docu- ments using top ranking sentences: an evaluation of two alternative schemes. In Proceedings of the 25th annual international ACM SIGIR conference on Research and development in information retrieval . ACM, 57–64

work page 2002

[31] [31]

Ronald J Williams. 1992. Simple statistical gradient-following algorithms for connectionist reinforcement learning. Machine learning 8, 3-4 (1992), 229–256. Ranking sentences from product description & bullets for be/t_ter search SIGIR 2019 eCom, July 2019, Paris, France

work page 1992

[32] [32]

Michihiro Yasunaga, Rui Zhang, Kshitijh Meelu, Ayush Pareek, Krishnan Srini- vasan, and Dragomir Radev. 2017. Graph-based neural multi-document summa- rization. arXiv preprint arXiv:1706.06681 (2017)

work page internal anchor Pith review Pith/arXiv arXiv 2017