arxiv: 2605.07314 · v1 · submitted 2026-05-08 · 💻 cs.IR · cs.AI

Recognition: 2 theorem links

· Lean Theorem

DCGL: Dual-Channel Graph Learning with Large Language Models for Knowledge-Aware Recommendation

Xinchi Zou , Tongzhenzhi Su , Jianjun Li , Yuan Fu , Chang Liu , Zhiying Deng , Zhiwei Shen

Authors on Pith no claims yet

Pith reviewed 2026-05-11 01:00 UTC · model grok-4.3

classification 💻 cs.IR cs.AI

keywords knowledge-aware recommendationlarge language modelsknowledge graphsdual-channel learningcontrastive learninggraph neural networkssparse datadynamic fusion

0 comments

The pith

DCGL uses dual channels to decouple LLM semantics from user behavior patterns, improving knowledge-aware recommendations especially when data is sparse.

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

The paper proposes the DCGL framework to fix three problems in current KG-plus-LLM recommendation systems: weak capture of implicit semantics, interference when fusing ID and language embeddings in one channel, and ignoring how often users interact with items. It separates semantic knowledge into one channel and behavioral patterns into another to stop early mixing of signals. Multi-level contrastive learning then makes the model more robust to noisy graph links and aligns the two channels, while a fusion step adapts the balance according to each user's interaction count.

Core claim

The DCGL framework features three innovations: a dual-channel architecture that structurally decouples rich semantic information from user behavioral patterns preventing early interference, a multi-level contrastive learning mechanism that enhances robustness against KG noise through intra-view contrasts and bridges semantic gaps between channels via inter-view alignment, and a dynamic fusion mechanism that adaptively balances semantic generalization and behavioral specificity based on interaction frequency.

What carries the argument

Dual-channel graph learning architecture that decouples semantic embeddings from behavioral ID patterns, combined with multi-level contrastive alignment and interaction-frequency-based dynamic fusion.

If this is right

Consistent outperformance of state-of-the-art methods across four real-world datasets.
Substantial gains specifically for users with limited interactions while preserving accuracy for frequent users.
Better capture of implicit semantic relationships beyond explicit KG links.
Reduced impact of KG noise through intra- and inter-view contrastive learning.
Adaptive balancing of semantic and behavioral signals without fixed hyperparameters.

Where Pith is reading between the lines

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

The separation principle may transfer to other settings where language model outputs must stay distinct from collaborative signals, such as conversational recommenders.
Interaction frequency as a dynamic control knob suggests similar adaptive weighting could help in session-based or cold-start recommendation tasks.
The framework implies that explicit channel alignment steps will become standard when scaling LLM-augmented graphs to larger catalogs.

Load-bearing premise

The assumption that dual-channel decoupling, multi-level contrastive alignment, and interaction-frequency-based dynamic fusion will prevent signal interference and resolve the three limitations without introducing new trade-offs or needing extensive tuning.

What would settle it

Ablation experiments on the same four real-world datasets showing that removing the dual-channel separation yields equal or higher accuracy in sparse user groups would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.07314 by Chang Liu, Jianjun Li, Tongzhenzhi Su, Xinchi Zou, Yuan Fu, Zhiwei Shen, Zhiying Deng.

**Figure 2.** Figure 2: The framework of our proposed DCGL. Progressing further, GNN-based methods [16, 18, 25, 26, 32, 33, 39, 40], at the research forefront, refine entity and relation representations by aggregating multi-hop neighbor embeddings. Notably, KGAT [25] employs graph attention mechanisms to propagate embeddings based on neighbor importance, enabling end-to-end recommendation score generation; KGRec [32] evaluates… view at source ↗

**Figure 3.** Figure 3: Prompt template for LLM-based entity description, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

**Figure 4.** Figure 4: Performance of DCGL vs. Baselines under different user & item groups (Book-Crossing Dataset). with a fixed-weight scheme (w/o FREQ) leads to clear performance decline, verifying that dynamic gating is essential for balancing semantic and behavioural information according to interaction frequency. Furthermore, removing either intra-view augmentation (w/o AUG) or inter-view alignment (w/o ALIGN) significant… view at source ↗

**Figure 6.** Figure 6: Visualization of the learned gating weights [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗

read the original abstract

Knowledge Graphs (KGs) have proven highly effective for recommendation systems by capturing latent item relationships, while recent integration of Large Language Models (LLMs) has further enhanced semantic understanding and addressed knowledge sparsity issues. Nevertheless, current KG-and-LLM-based methods still face three main limitations: 1) inadequate modeling of implicit semantic relationships beyond explicit KG links; 2) suboptimal single-channel fusion of ID and LLM embeddings, which often leads to signal interference and blurred representations; and 3) insufficient consideration of user-item interaction frequency variations in recommendation strategies. To address these challenges, we propose the Dual-Channel Graph Learning (DCGL) framework, featuring three key innovations: 1) a dual-channel architecture that structurally decouples rich semantic information from user behavioral patterns, preventing early interference; 2) a multi-level contrastive learning mechanism that enhances robustness against KG noise through intra-view contrasts and bridges semantic gaps between channels via inter-view alignment; and 3) a dynamic fusion mechanism that adaptively balances semantic generalization and behavioral specificity based on interaction frequency, resolving the cascading limitation. Extensive experiments on four real-world datasets show that DCGL consistently outperforms state-of-the-art methods, yielding substantial improvements in sparse scenarios while maintaining precision for active users. Our code is available at https://github.com/XinchiZou/DCGL.

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

DCGL's dual-channel split plus frequency-adaptive fusion is a practical engineering move for KG-LLM recommenders, but the sparse-user gains rest on unablated claims about the fusion step.

read the letter

The paper's core idea is to keep LLM-derived semantic embeddings in one channel and user-item behavioral patterns in another, then align them with multi-level contrastive learning before fusing based on interaction frequency. This directly targets the interference and sparsity issues the authors list, and the combination of structural decoupling with frequency-dependent weighting has not appeared in exactly this form in the prior KG-LLM work they cite. The four-dataset experiments report consistent outperformance, with larger lifts for low-interaction users, and the public code is a real plus for anyone who wants to test it themselves. That is the useful part: a concrete architecture that tries to balance semantic generalization against behavioral specificity without early mixing. The soft spot is the dynamic fusion module. The abstract and method treat interaction frequency as a reliable proxy that prevents interference and delivers the sparse gains, yet there is no sign of ablations that isolate the fusion on frequency-stratified user subsets or compare against a fixed-fusion baseline. Without those checks it is hard to know whether the dual-channel design alone explains the results or whether the frequency heuristic is load-bearing. Statistical significance, hyperparameter sensitivity, and controls for the contrastive losses are also not detailed in what is visible. This is incremental but cleanly scoped work aimed at recsys researchers who already combine KGs with LLMs. It is worth a serious referee to verify the experiments and see whether the fusion step holds up under closer inspection.

Referee Report

2 major / 2 minor

Summary. The manuscript proposes the DCGL framework for knowledge-aware recommendation. It integrates knowledge graphs with large language models to address three stated limitations of prior work: inadequate modeling of implicit semantic relationships, signal interference from single-channel fusion of ID and LLM embeddings, and lack of adaptation to user-item interaction frequency variations. The framework introduces a dual-channel architecture to decouple semantic and behavioral signals, multi-level contrastive learning (intra-view for robustness against KG noise and inter-view for channel alignment), and an interaction-frequency-based dynamic fusion mechanism. Extensive experiments on four real-world datasets are reported to show consistent outperformance over state-of-the-art methods, with particular gains in sparse scenarios while preserving precision for active users. Code is released at https://github.com/XinchiZou/DCGL.

Significance. If the central empirical claims hold after targeted validation, the work would advance LLM-augmented KG recommendation by offering a principled separation of semantic generalization from behavioral specificity and an adaptive fusion strategy. The open-source code is a clear strength that supports reproducibility and follow-on research in the field.

major comments (2)

[Experiments] Experiments section: the abstract and results claim substantial improvements in sparse scenarios attributable to the dual-channel decoupling, multi-level contrastive alignment, and interaction-frequency-based dynamic fusion, yet no ablation studies are described that isolate the dynamic fusion module on frequency-stratified user subsets (e.g., low-frequency vs. high-frequency users). Without these controls, it is not possible to confirm that the frequency heuristic itself drives the reported gains rather than the dual-channel architecture alone.
[Method] Method section (dynamic fusion description): the mechanism is asserted to adaptively balance semantic generalization and behavioral specificity based on interaction frequency, but the manuscript provides no explicit formulation, threshold selection procedure, or sensitivity analysis for the fusion weights or frequency proxy. This is load-bearing for the claim that the approach resolves signal interference without introducing new trade-offs or hyperparameter sensitivity.

minor comments (2)

[Abstract] Abstract: the phrase 'substantial improvements' is used without any numerical quantification (e.g., relative gains in Recall@K or NDCG@K on sparse subsets); adding concrete metrics would improve clarity.
[Method] Notation: the distinction between 'intra-view contrasts' and 'inter-view alignment' is introduced without an accompanying equation or diagram reference in the method overview, which could be clarified for readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. We address each major comment below and outline the revisions we will make to strengthen the paper.

read point-by-point responses

Referee: [Experiments] Experiments section: the abstract and results claim substantial improvements in sparse scenarios attributable to the dual-channel decoupling, multi-level contrastive alignment, and interaction-frequency-based dynamic fusion, yet no ablation studies are described that isolate the dynamic fusion module on frequency-stratified user subsets (e.g., low-frequency vs. high-frequency users). Without these controls, it is not possible to confirm that the frequency heuristic itself drives the reported gains rather than the dual-channel architecture alone.

Authors: We agree that the current experiments do not include ablations that isolate the dynamic fusion module specifically on frequency-stratified user subsets. The manuscript reports overall ablations for the dual-channel and contrastive components as well as results on sparse vs. dense scenarios, but lacks the targeted controls requested. In the revised manuscript we will add these experiments: we will stratify users by interaction frequency (e.g., low-frequency as the bottom quartile and high-frequency as the top quartile), report performance with and without the dynamic fusion module on each stratum, and quantify the incremental contribution of the frequency-adaptive fusion to the observed gains in sparse settings. revision: yes
Referee: [Method] Method section (dynamic fusion description): the mechanism is asserted to adaptively balance semantic generalization and behavioral specificity based on interaction frequency, but the manuscript provides no explicit formulation, threshold selection procedure, or sensitivity analysis for the fusion weights or frequency proxy. This is load-bearing for the claim that the approach resolves signal interference without introducing new trade-offs or hyperparameter sensitivity.

Authors: We acknowledge that the manuscript does not supply an explicit mathematical formulation of the dynamic fusion weights, a clear description of how the frequency proxy or any thresholds are chosen, or a sensitivity analysis. These details are necessary to substantiate the claims. In the revision we will insert the full formulation of the fusion function (including how interaction frequency is normalized and mapped to channel weights), specify the threshold/proxy selection procedure based on dataset statistics, and add a sensitivity study on the controlling hyperparameter(s) either in the main experiments section or in an expanded appendix. revision: yes

Circularity Check

0 steps flagged

No circularity: framework design and empirical claims are independent of self-referential reductions.

full rationale

The DCGL paper proposes an architectural framework (dual-channel decoupling, multi-level contrastive alignment, frequency-based dynamic fusion) motivated by three external limitations in prior KG+LLM recommenders. No equations, derivations, or first-principles predictions are presented that reduce to the inputs by construction; performance gains are asserted via experiments on four real-world datasets rather than tautological fits or self-citations. Design choices address stated problems without renaming known results or smuggling ansatzes via self-citation chains. The central claims remain falsifiable against external benchmarks and do not collapse into the framework's own definitions.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The central claim rests on the effectiveness of KGs for capturing item relationships and LLMs for semantic understanding (standard domain assumptions), plus the unproven premise that early fusion causes interference and that frequency-based dynamic weighting will resolve it without side effects. No free parameters or invented entities are explicitly quantified in the abstract.

axioms (2)

domain assumption Knowledge graphs effectively capture latent item relationships and LLMs address knowledge sparsity in recommendation.
Stated in the opening sentence of the abstract as established background.
domain assumption Single-channel fusion of ID and LLM embeddings leads to signal interference.
Presented as one of the three main limitations current methods face.

pith-pipeline@v0.9.0 · 5556 in / 1446 out tokens · 27530 ms · 2026-05-11T01:00:08.107966+00:00 · methodology

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/Cost/FunctionalEquation.lean washburn_uniqueness_aczel unclear

?

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

dual-channel architecture that structurally decouples rich semantic information from user behavioral patterns... frequency-aware gated fusion mechanism that adaptively balances semantic generalization and behavioral specificity based on interaction frequency
IndisputableMonolith/Foundation/ArithmeticFromLogic.lean LogicNat.equivNat unclear

?

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

multi-level contrastive learning... intra-view contrasts and... inter-view alignment

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

45 extracted references · 2 canonical work pages · 1 internal anchor

[1]

Qingyao Ai, Vahid Azizi, Xu Chen, and Yongfeng Zhang. 2018. Learning heteroge- neous knowledge base embeddings for explainable recommendation.Algorithms 11, 9 (2018), 137

2018
[2]

Keqin Bao, Jizhi Zhang, Yang Zhang, Wenjie Wang, Fuli Feng, and Xiangnan He. 2023. Tallrec: An effective and efficient tuning framework to align large language model with recommendation. InProceedings of the 17th ACM conference on recommender systems. 1007–1014

2023
[3]

Antoine Bordes, Nicolas Usunier, Alberto Garcia-Duran, Jason Weston, and Ok- sana Yakhnenko. 2013. Translating embeddings for modeling multi-relational data.Advances in neural information processing systems26 (2013)

2013
[4]

Yixin Cao, Xiang Wang, Xiangnan He, Zikun Hu, and Tat-Seng Chua. 2019. Unifying knowledge graph learning and recommendation: Towards a better understanding of user preferences. InThe world wide web conference. 151–161

2019
[5]

Jin Chen, Zheng Liu, Xu Huang, Chenwang Wu, Qi Liu, Gangwei Jiang, Yuanhao Pu, Yuxuan Lei, Xiaolong Chen, Xingmei Wang, et al. 2024. When large language models meet personalization: Perspectives of challenges and opportunities.World Wide Web27, 4 (2024), 42

2024
[6]

Ziqiang Cui, Yunpeng Weng, Xing Tang, Fuyuan Lyu, Dugang Liu, Xiuqiang He, and Chen Ma. 2025. Comprehending knowledge graphs with large language models for recommender systems. InProceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval. 1229– 1239

2025
[7]

Xin Dong, Lei Yu, Zhonghuo Wu, Yuxia Sun, Lingfeng Yuan, and Fangxi Zhang
[8]

InProceedings of the AAAI Conference on artificial intelligence, Vol

A hybrid collaborative filtering model with deep structure for recommender systems. InProceedings of the AAAI Conference on artificial intelligence, Vol. 31
[9]

Chongming Gao, Ruijun Chen, Shuai Yuan, Kexin Huang, Yuanqing Yu, and Xiangnan He. 2025. Sprec: Self-play to debias llm-based recommendation. In Proceedings of the ACM on Web Conference 2025. 5075–5084

2025
[10]

Chen Gao, Yu Zheng, Nian Li, Yinfeng Li, Yingrong Qin, Jinghua Piao, Yuhan Quan, Jianxin Chang, Depeng Jin, Xiangnan He, et al. 2023. A survey of graph neural networks for recommender systems: Challenges, methods, and directions. ACM Transactions on Recommender Systems1, 1 (2023), 1–51

2023
[11]

Ruining He and Julian McAuley. 2016. Ups and downs: Modeling the visual evolution of fashion trends with one-class collaborative filtering. Inproceedings of the 25th international conference on world wide web. 507–517

2016
[12]

Xiangnan He and Tat-Seng Chua. 2017. Neural factorization machines for sparse predictive analytics. InProceedings of the 40th International ACM SIGIR conference on Research and Development in Information Retrieval. 355–364

2017
[13]

Xiangnan He, Kuan Deng, Xiang Wang, Yan Li, Yongdong Zhang, and Meng Wang. 2020. Lightgcn: Simplifying and powering graph convolution network for recommendation. InProceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval. 639–648

2020
[14]

Yupeng Hou, Junjie Zhang, Zihan Lin, Hongyu Lu, Ruobing Xie, Julian McAuley, and Wayne Xin Zhao. 2024. Large language models are zero-shot rankers for recommender systems. InEuropean Conference on Information Retrieval. Springer, 364–381

2024
[15]

Zheng Hu, Zhe Li, Ziyun Jiao, Satoshi Nakagawa, Jiawen Deng, Shimin Cai, Tao Zhou, and Fuji Ren. 2025. Bridging the user-side knowledge gap in knowledge- aware recommendations with large language models. InProceedings of the AAAI Conference on Artificial Intelligence, Vol. 39. 11799–11807

2025
[16]

Yangqin Jiang, Yuhao Yang, Lianghao Xia, Da Luo, Kangyi Lin, and Chao Huang
[17]

RecLM: Recommendation Instruction Tuning. InProceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), Wanxiang Che, Joyce Nabende, Ekaterina Shutova, and Mohammad Taher Pilehvar (Eds.). Association for Computational Linguistics, 15443–15459. https: //aclanthology.org/2025.acl-long.751/

2025
[18]

Anchen Li, Bo Yang, Huan Huo, Farookh Hussain, and Guandong Xu. 2025. Hy- percomplex knowledge graph-aware recommendation. InProceedings of the 48th international ACM SIGIR conference on research and development in information retrieval. 2017–2026

2025
[19]

Lei Li, Yongfeng Zhang, Dugang Liu, and Li Chen. 2024. Large language models for generative recommendation: A survey and visionary discussions. InProceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 10146–10159

2024
[20]

Yanhui Li, Dongxia Wang, Zhu Sun, Haonan Zhang, and Huizhong Guo. 2025. LightKG: Efficient Knowledge-Aware Recommendations with Simplified GNN Architecture. InProceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2. 1577–1588

2025
[21]

Yuhan Li, Xinni Zhang, Linhao Luo, Heng Chang, Yuxiang Ren, Irwin King, and Jia Li. 2025. G-refer: Graph retrieval-augmented large language model for explainable recommendation. InProceedings of the ACM on Web Conference 2025. 240–251

2025
[22]

Tommaso Di Noia, Vito Claudio Ostuni, Paolo Tomeo, and Eugenio Di Sciascio
[23]

Sprank: Semantic path-based ranking for top-n recommendations using linked open data.ACM Transactions on Intelligent Systems and Technology (TIST) 8, 1 (2016), 1–34

2016
[24]

Xubin Ren, Wei Wei, Lianghao Xia, Lixin Su, Suqi Cheng, Junfeng Wang, Dawei Yin, and Chao Huang. 2024. Representation learning with large language models for recommendation. InProceedings of the ACM web conference 2024. 3464–3475

2024
[25]

Steffen Rendle, Christoph Freudenthaler, Zeno Gantner, and Lars Schmidt-Thieme
[26]

BPR: Bayesian personalized ranking from implicit feedback.arXiv preprint arXiv:1205.2618(2012)

work page internal anchor Pith review arXiv 2012
[27]

Riku Togashi, Mayu Otani, and Shin’ichi Satoh. 2021. Alleviating cold-start problems in recommendation through pseudo-labelling over knowledge graph. DCGL: Dual-Channel Graph Learning with Large Language Models for Knowledge-Aware Recommendation SIGIR ’26, July 20–24, 2026, Melbourne, VIC, Australia InProceedings of the 14th ACM international conference o...

2021
[28]

Hongwei Wang, Fuzheng Zhang, Xing Xie, and Minyi Guo. 2018. DKN: Deep knowledge-aware network for news recommendation. InProceedings of the 2018 world wide web conference. 1835–1844

2018
[29]

Xiang Wang, Xiangnan He, Yixin Cao, Meng Liu, and Tat-Seng Chua. 2019. Kgat: Knowledge graph attention network for recommendation. InProceedings of the 25th ACM SIGKDD international conference on knowledge discovery & data mining. 950–958

2019
[30]

Xiang Wang, Tinglin Huang, Dingxian Wang, Yancheng Yuan, Zhenguang Liu, Xiangnan He, and Tat-Seng Chua. 2021. Learning intents behind interactions with knowledge graph for recommendation. InProceedings of the web conference

2021
[31]

Xiang Wang, Dingxian Wang, Canran Xu, Xiangnan He, Yixin Cao, and Tat-Seng Chua. 2019. Explainable Reasoning over Knowledge Graphs for Recommendation. Proceedings of the AAAI Conference on Artificial Intelligence33, 01 (Jul. 2019), 5329–5336. doi:10.1609/aaai.v33i01.33015329

work page doi:10.1609/aaai.v33i01.33015329 2019
[32]

Wei Wei, Xubin Ren, Jiabin Tang, Qinyong Wang, Lixin Su, Suqi Cheng, Jun- feng Wang, Dawei Yin, and Chao Huang. 2024. Llmrec: Large language models with graph augmentation for recommendation. InProceedings of the 17th ACM international conference on web search and data mining. 806–815

2024
[33]

Jiancan Wu, Xiang Wang, Fuli Feng, Xiangnan He, Liang Chen, Jianxun Lian, and Xing Xie. 2021. Self-supervised graph learning for recommendation. InProceed- ings of the 44th international ACM SIGIR conference on research and development in information retrieval. 726–735

2021
[34]

Likang Wu, Zhi Zheng, Zhaopeng Qiu, Hao Wang, Hongchao Gu, Tingjia Shen, Chuan Qin, Chen Zhu, Hengshu Zhu, Qi Liu, et al . 2024. A survey on large language models for recommendation.World Wide Web27, 5 (2024), 60

2024
[35]

Xu Xie, Fei Sun, Zhaoyang Liu, Shiwen Wu, Jinyang Gao, Jiandong Zhang, Bolin Ding, and Bin Cui. 2022. Contrastive learning for sequential recommendation. In 2022 IEEE 38th international conference on data engineering (ICDE). IEEE, 1259– 1273

2022
[36]

Yuhao Yang, Chao Huang, Lianghao Xia, and Chunzhen Huang. 2023. Knowledge graph self-supervised rationalization for recommendation. InProceedings of the 29th ACM SIGKDD conference on knowledge discovery and data mining. 3046–3056

2023
[37]

Yuhao Yang, Chao Huang, Lianghao Xia, and Chenliang Li. 2022. Knowledge graph contrastive learning for recommendation. InProceedings of the 45th in- ternational ACM SIGIR conference on research and development in information retrieval. 1434–1443

2022
[38]

Junliang Yu, Hongzhi Yin, Xin Xia, Tong Chen, Lizhen Cui, and Quoc Viet Hung Nguyen. 2022. Are graph augmentations necessary? simple graph contrastive learning for recommendation. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval. 1294–1303

2022
[39]

Xiao Yu, Xiang Ren, Yizhou Sun, Quanquan Gu, Bradley Sturt, Urvashi Khandel- wal, Brandon Norick, and Jiawei Han. 2014. Personalized entity recommendation: A heterogeneous information network approach. InProceedings of the 7th ACM international conference on Web search and data mining. 283–292

2014
[40]

Fuzheng Zhang, Nicholas Jing Yuan, Defu Lian, Xing Xie, and Wei-Ying Ma
[41]

In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining

Collaborative knowledge base embedding for recommender systems. In Proceedings of the 22nd ACM SIGKDD international conference on knowledge discovery and data mining. 353–362
[42]

Qian Zhao, Hao Qian, Ziqi Liu, Gong-Duo Zhang, and Lihong Gu. 2024. Break- ing the barrier: utilizing large language models for industrial recommendation systems through an inferential knowledge graph. InProceedings of the 33rd ACM International Conference on Information and Knowledge Management. 5086–5093

2024
[43]

Zihuai Zhao, Wenqi Fan, Jiatong Li, Yunqing Liu, Xiaowei Mei, Yiqi Wang, Zhen Wen, Fei Wang, Xiangyu Zhao, Jiliang Tang, et al. 2024. Recommender systems in the era of large language models (llms).IEEE Transactions on Knowledge and Data Engineering36, 11 (2024), 6889–6907

2024
[44]

Ding Zou, Wei Wei, Xian-Ling Mao, Ziyang Wang, Minghui Qiu, Feida Zhu, and Xin Cao. 2022. Multi-level cross-view contrastive learning for knowledge- aware recommender system. InProceedings of the 45th international ACM SIGIR conference on research and development in information retrieval. 1358–1368

2022
[45]

Ding Zou, Wei Wei, Feida Zhu, Chuanyu Xu, Tao Zhang, and Chengfu Huo. 2024. Knowledge enhanced multi-intent transformer network for recommendation. In Companion proceedings of the ACM web conference 2024. 1–9

2024