A Standardized Re-evaluation of Conversational Recommender Systems on the ReDial Dataset

Ivica Kostric; Krisztian Balog

arxiv: 2605.13053 · v2 · pith:YVDWE34Unew · submitted 2026-05-13 · 💻 cs.IR

A Standardized Re-evaluation of Conversational Recommender Systems on the ReDial Dataset

Ivica Kostric , Krisztian Balog This is my paper

Pith reviewed 2026-05-21 08:56 UTC · model grok-4.3

classification 💻 cs.IR

keywords conversational recommender systemsReDial datasetreproducibility studyevaluation metricsnoveltyLLM backbonerecall metricsuser-centric evaluation

0 comments

The pith

Standardized re-evaluation of conversational recommenders on ReDial shows nearly half of reported accuracy comes from repetition shortcuts rather than genuine recommendations.

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

The paper re-runs seven conversational recommender systems from different architectural families on the ReDial dataset using identical preprocessing steps, ground-truth definitions, and evaluation metrics. It demonstrates that models frequently achieve high scores by repeating items already mentioned in the conversation history, and that removing these repetition shortcuts halves the apparent accuracy in novelty-focused tests. The work also finds that gains between methods shrink or disappear when the underlying large language model is held fixed, indicating that backbone capacity explains more variance than specific design choices. Finally, user-centric utility measures reveal that standard recall metrics overstate real conversational value because they reward echoing prior turns instead of introducing useful new items.

Core claim

When seven prominent conversational recommender systems are evaluated under fixed preprocessing, a uniform definition of ground-truth items, and metrics that ignore items already present in the conversation, nearly 50 percent of previously reported accuracy vanishes because it originated in repetition shortcuts. Once the capacity of the LLM backbone is controlled, differences attributable to architectural innovations become small or inconsistent. Traditional recall@K scores prove misleading for conversational effectiveness because they credit systems that simply echo conversation history rather than deliver novel, relevant suggestions.

What carries the argument

The controlled re-evaluation protocol that fixes preprocessing, ground-truth item selection, and novelty-aware metrics across methods from three architectural families.

If this is right

Architectural comparisons in CRS must hold the LLM backbone fixed to isolate the contribution of model design.
Novelty-aware metrics that exclude repeated items should supplement or replace standard recall to measure true recommendation quality.
User-centric utility metrics give a more realistic view of conversational effectiveness than aggregate recall alone.
Fine-grained ranking metrics like Recall@1 are highly sensitive to small implementation differences and require strict standardization for reliable comparisons.

Where Pith is reading between the lines

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

Similar standardization efforts on other conversational recommendation datasets could expose comparable overstatements in published results.
Future architecture papers should report results across multiple backbone sizes to separate model capacity from design innovations.
Simple history-based baselines may match more complex systems when novelty is not enforced, suggesting a need to test against such controls.
Extending the protocol to full multi-turn simulations could highlight differences in interaction efficiency that static metrics miss.

Load-bearing premise

The authors' specific choices for preprocessing, ground-truth definition, and novelty-focused metrics faithfully represent the original ReDial dataset intent and earlier studies without introducing new selection biases.

What would settle it

Re-running the same seven methods with an alternative ground-truth definition that treats repeated conversation items as valid or with metrics that allow repetition and checking whether the 50 percent accuracy reduction and backbone-dominance pattern still hold.

Figures

Figures reproduced from arXiv: 2605.13053 by Ivica Kostric, Krisztian Balog.

**Figure 2.** Figure 2: Example dialogue from the ReDial dataset. The [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗

**Figure 1.** Figure 1: Finally, our evaluation using novel user-centric metrics [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗

**Figure 3.** Figure 3: Commonly used pipelines in modern CRSs. Green signifies the components that use an LLM. (Left) Modular Fusion Pipelines (KBRD and KGSF) use different, disjoint components user and dialogue modeling. Outputs from the recommender are integrated with the output from the modeling components. (Middle) Shared-Backbone Pipelines (UniCRS and ECR) use the same model for recommendation and dialogue generation using … view at source ↗

**Figure 4.** Figure 4: Recall@1 under reported, standardized, and dedu [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗

read the original abstract

Recent years have seen a surge of research into conversational recommender systems (CRS). Among existing datasets, ReDial is the most widely used benchmark, cited in hundreds of studies. However, variations in how the dataset is preprocessed and used in experiments, particularly in the definition of ground-truth items, make it difficult to compare results across studies. These comparisons are further complicated by confounding factors such as the choice of the underlying large language model (LLM) and the use of external data sources. In this work, we revisit seven prominent CRS methods across three architectural families and evaluate them under standardized conditions. Our reproducibility study reveals a ``granularity gap,'' where fine-grained ranking (Recall@1) is highly sensitive to implementation details, while our replicability analysis shows that nearly 50% of reported accuracy stems from ``repetition shortcuts'' that are absent in novelty-focused evaluation. Furthermore, we find that performance gains are often driven more by the capacity of the LLM backbone than by specific architectural innovations. Finally, by applying user-centric utility metrics, we demonstrate that traditional recall frequently overstates a system's actual conversational effectiveness. This work establishes a transparent, controlled baseline and promotes evaluation practices that prioritize novelty and interaction efficiency.

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

This re-evaluation finds that repetition shortcuts explain nearly half the accuracy on ReDial and that LLM size often matters more than the proposed CRS architectures.

read the letter

The main takeaway is that a lot of what gets reported as progress in conversational recommenders on ReDial is actually models picking up on repeated items from earlier in the dialogue. The paper puts a number on it—around 50 percent of the accuracy in their standardized runs—and shows that switching to novelty-aware metrics changes the rankings noticeably. They also document a granularity gap where Recall@1 is brittle to small preprocessing choices while coarser metrics look more stable. On the positive side, re-running seven methods from three families under one controlled pipeline gives a cleaner picture than the scattered numbers in the literature. The addition of user-centric utility metrics is a straightforward way to show that raw recall can overstate how useful the conversation actually is. The work is honest about the role of the underlying LLM backbone, which is a useful corrective when many papers claim architectural wins without isolating that factor. The soft spot is the standardization itself. Their preprocessing and ground-truth rules may not line up exactly with how each of the original seven papers treated repeats or item sets, so the 50 percent shortcut figure could partly reflect the new protocol rather than pure overstatement in prior results. Without side-by-side tables matching their ground-truth lists to the ones used in the re-implemented papers, it is hard to rule that out. This paper is for anyone who builds or benchmarks CRS models on ReDial. Readers who care about evaluation hygiene will get concrete numbers they can check. It is worth sending to peer review so referees can verify the preprocessing details and see whether the headline claims survive that check.

Referee Report

2 major / 2 minor

Summary. The paper conducts a standardized re-evaluation of seven prominent conversational recommender systems (CRS) across three architectural families on the ReDial dataset. It identifies a granularity gap where fine-grained ranking (e.g., Recall@1) is sensitive to implementation details, attributes nearly 50% of reported accuracy to repetition shortcuts absent under novelty-focused evaluation, finds that performance gains are driven more by LLM backbone capacity than specific architectural innovations, and shows via user-centric utility metrics that traditional recall overstates actual conversational effectiveness.

Significance. If the standardization of preprocessing, ground-truth definitions, and metrics accurately preserves the conventions of the original studies, this work establishes a valuable controlled baseline for CRS research. The explicit quantitative findings on repetition shortcuts and LLM dominance, together with the promotion of novelty-focused and interaction-efficiency metrics, could improve reproducibility and discourage overstated claims based on flawed evaluation practices.

major comments (2)

[§4.2] §4.2 (Replicability Analysis): The central claim that nearly 50% of reported accuracy stems from repetition shortcuts depends on the authors' chosen novelty-focused evaluation and ground-truth item mapping. Without a direct side-by-side comparison to the exact repetition handling and ground-truth lists used in the seven re-implemented original papers, it remains unclear whether this percentage reflects overstatement in prior work or differences introduced by the new standardization protocol.
[§3.3] §3.3 (Ground-Truth Definition): The standardization of ground-truth items and dialogue-to-item granularity is load-bearing for both the granularity-gap observation and the shortcut attribution. The manuscript does not report explicit validation that these choices match the item sets and repetition allowances employed in the baseline methods, which risks turning the headline quantitative result into an artifact of the re-evaluation design.

minor comments (2)

[Table 2] Table 2: The ablation results isolating LLM backbone effects would benefit from additional columns showing variance across multiple random seeds to strengthen the claim that architecture is secondary to backbone capacity.
[§5] §5 (User-Centric Metrics): The transition from traditional recall to utility metrics is promising but would be clearer if the exact formulation of the utility function were provided as an equation rather than described in prose.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed and constructive comments on our reproducibility study. We address each major comment below, providing clarifications on our standardization choices and indicating where revisions will strengthen the manuscript.

read point-by-point responses

Referee: [§4.2] §4.2 (Replicability Analysis): The central claim that nearly 50% of reported accuracy stems from repetition shortcuts depends on the authors' chosen novelty-focused evaluation and ground-truth item mapping. Without a direct side-by-side comparison to the exact repetition handling and ground-truth lists used in the seven re-implemented original papers, it remains unclear whether this percentage reflects overstatement in prior work or differences introduced by the new standardization protocol.

Authors: We agree that a direct side-by-side comparison would further strengthen the claim. Our re-implementations followed the ground-truth definitions, preprocessing steps, and repetition allowances as described in each of the seven original papers and their released code (where available). The 50% figure arises from an internal comparison within our standardized framework: the same models evaluated under the conventional protocol (which permits recommending previously mentioned items) versus a novelty-focused protocol that excludes such items from the ground-truth. This isolates the contribution of repetition shortcuts without altering the underlying model implementations. We will revise the manuscript to include an explicit table contrasting the evaluation settings reported in the originals with our protocol, along with additional details on how repetition was handled in each baseline. revision: partial
Referee: [§3.3] §3.3 (Ground-Truth Definition): The standardization of ground-truth items and dialogue-to-item granularity is load-bearing for both the granularity-gap observation and the shortcut attribution. The manuscript does not report explicit validation that these choices match the item sets and repetition allowances employed in the baseline methods, which risks turning the headline quantitative result into an artifact of the re-evaluation design.

Authors: We acknowledge the importance of explicit validation for the ground-truth construction. Section 3.3 details our mapping of dialogues to recommended items, derived from the ReDial dataset structure and aligned with the conventions described in the original papers for each architectural family. Where code was available, we cross-checked against the baselines' item sets and allowed repetitions in the standard evaluation to match their reported setups. The novelty-focused variant is presented as a diagnostic analysis rather than a replacement for prior protocols. In the revision, we will add an appendix providing per-method validation of the ground-truth rules and repetition handling to make the alignment transparent. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical re-evaluation with external benchmarks

full rationale

The paper conducts a reproducibility study by re-implementing seven existing CRS methods under a standardized preprocessing and evaluation protocol on the ReDial dataset. All quantitative claims (e.g., repetition-shortcut contribution, granularity gap, LLM backbone dominance) are obtained by direct measurement against previously published results rather than by fitting parameters to the authors' own outputs or by deriving predictions that reduce to self-defined inputs. No equations, ansatzes, or uniqueness theorems are introduced that would make any result tautological by construction. Self-citations, if present, are not load-bearing for the central replicability findings, which rest on transparent protocol choices and external comparisons.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claims rest on the assumption that the chosen standardization of ReDial preprocessing and novelty metrics is the correct reference point. No free parameters are introduced. No new entities are postulated. Background assumptions are standard IR evaluation practices.

axioms (1)

domain assumption Standard IR metrics such as Recall@K and novelty-aware variants can be applied uniformly after fixing preprocessing and ground-truth definitions.
Invoked when the authors standardize conditions across methods to enable comparison.

pith-pipeline@v0.9.0 · 5744 in / 1339 out tokens · 32112 ms · 2026-05-21T08:56:15.702668+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

standardized evaluation settings across all methods

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

22 extracted references · 22 canonical work pages · 1 internal anchor

[1]

Nolwenn Bernard and Krisztian Balog. 2025. Limitations of Current Evaluation Practices for Conversational Recommender Systems and the Potential of User Simulation. InProceedings of the 2025 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region (SIGIR-AP ’25). 261–271. arXiv:2510.05624

work page arXiv 2025
[2]

Qibin Chen, Junyang Lin, Yichang Zhang, Ming Ding, Yukuo Cen, Hongxia Yang, and Jie Tang. 2019. Towards Knowledge-Based Recommender Dialog System. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Pro- cessing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) (EMNLP ’19). 1803–1813

work page 2019
[3]

Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. QLoRA: Efficient Finetuning of Quantized LLMs. arXiv:2305.14314

work page internal anchor Pith review Pith/arXiv arXiv 2023
[4]

Maurizio Ferrari Dacrema, Simone Boglio, Paolo Cremonesi, and Dietmar Jannach

work page
[5]

A Troubling Analysis of Reproducibility and Progress in Recommender Systems Research.ACM Trans. Inf. Syst.39, 2 (2021), 20:1–20:49

work page 2021
[6]

Shirley Anugrah Hayati, Dongyeop Kang, Qingxiaoyang Zhu, Weiyan Shi, and Zhou Yu. 2020. INSPIRED: Toward Sociable Recommendation Dialog Systems. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (EMNLP ’20). 8142–8152

work page 2020
[7]

Zhankui He, Zhouhang Xie, Rahul Jha, Harald Steck, Dawen Liang, Yesu Feng, Bodhisattwa Prasad Majumder, Nathan Kallus, and Julian Mcauley. 2023. Large Language Models as Zero-Shot Conversational Recommenders. InProceedings of the 32nd ACM International Conference on Information and Knowledge Manage- ment (CIKM ’23). 720–730

work page 2023
[8]

Dietmar Jannach. 2022. Evaluating conversational recommender systems: A landscape of research.Artificial Intelligence Review(2022)

work page 2022
[9]

Dongyeop Kang, Anusha Balakrishnan, Pararth Shah, Paul Crook, Y-Lan Boureau, and Jason Weston. 2019. Recommendation as a Communication Game: Self- Supervised Bot-Play for Goal-oriented Dialogue. InProceedings of the 2019 Con- ference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language Proces...

work page 2019
[10]

Raymond Li, Samira Ebrahimi Kahou, Hannes Schulz, Vincent Michalski, Laurent Charlin, and Chris Pal. 2018. Towards Deep Conversational Recommendations. InAdvances in Neural Information Processing Systems (NIPS ’18, Vol. 31)

work page 2018
[11]

Tingting Liang, Chenxin Jin, Lingzhi Wang, Wenqi Fan, Congying Xia, Kai Chen, and Yuyu Yin. 2024. LLM-REDIAL: A Large-Scale Dataset for Conversational Recommender Systems Created from User Behaviors with LLMs. InFindings of the Association for Computational Linguistics: ACL 2024 (Findings ’24). 8926–8939

work page 2024
[12]

Zeming Liu, Haifeng Wang, Zheng-Yu Niu, Hua Wu, and Wanxiang Che. 2021. DuRecDial 2.0: A Bilingual Parallel Corpus for Conversational Recommendation. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP ’21). 4335–4347

work page 2021
[13]

Mathieu Ravaut, Hao Zhang, Lu Xu, Aixin Sun, and Yong Liu. 2024. Parameter- Efficient Conversational Recommender System as a Language Processing Task. InProceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers) (EACL ’24). 152–165

work page 2024
[14]

Ting-Chun Wang, Shang-Yu Su, and Yun-Nung Chen. 2022. BARCOR: To- wards A Unified Framework for Conversational Recommendation Systems. arXiv:2203.14257

work page arXiv 2022
[15]

Xiaolei Wang, Kun Zhou, Ji-Rong Wen, and Wayne Xin Zhao. 2022. Towards Unified Conversational Recommender Systems via Knowledge-Enhanced Prompt Learning. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’22). 1929–1937

work page 2022
[16]

Bowen Yang, Cong Han, Yu Li, Lei Zuo, and Zhou Yu. 2022. Improving Con- versational Recommendation Systems’ Quality with Context-Aware Item Meta- Information. InFindings of the Association for Computational Linguistics: NAACL 2022 (NAACL ’22). 38–48

work page 2022
[17]

Ting Yang and Li Chen. 2024. Unleashing the Retrieval Potential of Large Lan- guage Models in Conversational Recommender Systems. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). 43–52

work page 2024
[18]

Xiaoyu Zhang, Ruobing Xie, Yougang Lyu, Xin Xin, Pengjie Ren, Mingfei Liang, Bo Zhang, Zhanhui Kang, Maarten de Rijke, and Zhaochun Ren. 2024. Towards Empathetic Conversational Recommender Systems. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). 84–93

work page 2024
[19]

Kun Zhou, Xiaolei Wang, Yuanhang Zhou, Chenzhan Shang, Yuan Cheng, Wayne Xin Zhao, Yaliang Li, and Ji-Rong Wen. 2021. CRSLab: An Open-Source Toolkit for Building Conversational Recommender System. InProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Proces...

work page 2021
[20]

Kun Zhou, Wayne Xin Zhao, Shuqing Bian, Yuanhang Zhou, Ji-Rong Wen, and Jingsong Yu. 2020. Improving Conversational Recommender Systems via Knowl- edge Graph based Semantic Fusion. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD ’20). 1006– 1014

work page 2020
[21]

Kun Zhou, Yuanhang Zhou, Wayne Xin Zhao, Xiaoke Wang, and Ji-Rong Wen

work page
[22]

InProceed- ings of the 28th International Conference on Computational Linguistics (COLING ’20)

Towards Topic-Guided Conversational Recommender System. InProceed- ings of the 28th International Conference on Computational Linguistics (COLING ’20). 4128–4139

work page

[1] [1]

Nolwenn Bernard and Krisztian Balog. 2025. Limitations of Current Evaluation Practices for Conversational Recommender Systems and the Potential of User Simulation. InProceedings of the 2025 Annual International ACM SIGIR Conference on Research and Development in Information Retrieval in the Asia Pacific Region (SIGIR-AP ’25). 261–271. arXiv:2510.05624

work page arXiv 2025

[2] [2]

Qibin Chen, Junyang Lin, Yichang Zhang, Ming Ding, Yukuo Cen, Hongxia Yang, and Jie Tang. 2019. Towards Knowledge-Based Recommender Dialog System. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Pro- cessing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP) (EMNLP ’19). 1803–1813

work page 2019

[3] [3]

Tim Dettmers, Artidoro Pagnoni, Ari Holtzman, and Luke Zettlemoyer. 2023. QLoRA: Efficient Finetuning of Quantized LLMs. arXiv:2305.14314

work page internal anchor Pith review Pith/arXiv arXiv 2023

[4] [4]

Maurizio Ferrari Dacrema, Simone Boglio, Paolo Cremonesi, and Dietmar Jannach

work page

[5] [5]

A Troubling Analysis of Reproducibility and Progress in Recommender Systems Research.ACM Trans. Inf. Syst.39, 2 (2021), 20:1–20:49

work page 2021

[6] [6]

Shirley Anugrah Hayati, Dongyeop Kang, Qingxiaoyang Zhu, Weiyan Shi, and Zhou Yu. 2020. INSPIRED: Toward Sociable Recommendation Dialog Systems. InProceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP) (EMNLP ’20). 8142–8152

work page 2020

[7] [7]

Zhankui He, Zhouhang Xie, Rahul Jha, Harald Steck, Dawen Liang, Yesu Feng, Bodhisattwa Prasad Majumder, Nathan Kallus, and Julian Mcauley. 2023. Large Language Models as Zero-Shot Conversational Recommenders. InProceedings of the 32nd ACM International Conference on Information and Knowledge Manage- ment (CIKM ’23). 720–730

work page 2023

[8] [8]

Dietmar Jannach. 2022. Evaluating conversational recommender systems: A landscape of research.Artificial Intelligence Review(2022)

work page 2022

[9] [9]

Dongyeop Kang, Anusha Balakrishnan, Pararth Shah, Paul Crook, Y-Lan Boureau, and Jason Weston. 2019. Recommendation as a Communication Game: Self- Supervised Bot-Play for Goal-oriented Dialogue. InProceedings of the 2019 Con- ference on Empirical Methods in Natural Language Processing and the 9th Interna- tional Joint Conference on Natural Language Proces...

work page 2019

[10] [10]

Raymond Li, Samira Ebrahimi Kahou, Hannes Schulz, Vincent Michalski, Laurent Charlin, and Chris Pal. 2018. Towards Deep Conversational Recommendations. InAdvances in Neural Information Processing Systems (NIPS ’18, Vol. 31)

work page 2018

[11] [11]

Tingting Liang, Chenxin Jin, Lingzhi Wang, Wenqi Fan, Congying Xia, Kai Chen, and Yuyu Yin. 2024. LLM-REDIAL: A Large-Scale Dataset for Conversational Recommender Systems Created from User Behaviors with LLMs. InFindings of the Association for Computational Linguistics: ACL 2024 (Findings ’24). 8926–8939

work page 2024

[12] [12]

Zeming Liu, Haifeng Wang, Zheng-Yu Niu, Hua Wu, and Wanxiang Che. 2021. DuRecDial 2.0: A Bilingual Parallel Corpus for Conversational Recommendation. InProceedings of the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP ’21). 4335–4347

work page 2021

[13] [13]

Mathieu Ravaut, Hao Zhang, Lu Xu, Aixin Sun, and Yong Liu. 2024. Parameter- Efficient Conversational Recommender System as a Language Processing Task. InProceedings of the 18th Conference of the European Chapter of the Association for Computational Linguistics (Volume 1: Long Papers) (EACL ’24). 152–165

work page 2024

[14] [14]

Ting-Chun Wang, Shang-Yu Su, and Yun-Nung Chen. 2022. BARCOR: To- wards A Unified Framework for Conversational Recommendation Systems. arXiv:2203.14257

work page arXiv 2022

[15] [15]

Xiaolei Wang, Kun Zhou, Ji-Rong Wen, and Wayne Xin Zhao. 2022. Towards Unified Conversational Recommender Systems via Knowledge-Enhanced Prompt Learning. InProceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD ’22). 1929–1937

work page 2022

[16] [16]

Bowen Yang, Cong Han, Yu Li, Lei Zuo, and Zhou Yu. 2022. Improving Con- versational Recommendation Systems’ Quality with Context-Aware Item Meta- Information. InFindings of the Association for Computational Linguistics: NAACL 2022 (NAACL ’22). 38–48

work page 2022

[17] [17]

Ting Yang and Li Chen. 2024. Unleashing the Retrieval Potential of Large Lan- guage Models in Conversational Recommender Systems. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). 43–52

work page 2024

[18] [18]

Xiaoyu Zhang, Ruobing Xie, Yougang Lyu, Xin Xin, Pengjie Ren, Mingfei Liang, Bo Zhang, Zhanhui Kang, Maarten de Rijke, and Zhaochun Ren. 2024. Towards Empathetic Conversational Recommender Systems. InProceedings of the 18th ACM Conference on Recommender Systems (RecSys ’24). 84–93

work page 2024

[19] [19]

Kun Zhou, Xiaolei Wang, Yuanhang Zhou, Chenzhan Shang, Yuan Cheng, Wayne Xin Zhao, Yaliang Li, and Ji-Rong Wen. 2021. CRSLab: An Open-Source Toolkit for Building Conversational Recommender System. InProceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Proces...

work page 2021

[20] [20]

Kun Zhou, Wayne Xin Zhao, Shuqing Bian, Yuanhang Zhou, Ji-Rong Wen, and Jingsong Yu. 2020. Improving Conversational Recommender Systems via Knowl- edge Graph based Semantic Fusion. InProceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (KDD ’20). 1006– 1014

work page 2020

[21] [21]

Kun Zhou, Yuanhang Zhou, Wayne Xin Zhao, Xiaoke Wang, and Ji-Rong Wen

work page

[22] [22]

InProceed- ings of the 28th International Conference on Computational Linguistics (COLING ’20)

Towards Topic-Guided Conversational Recommender System. InProceed- ings of the 28th International Conference on Computational Linguistics (COLING ’20). 4128–4139

work page