pith. sign in

arxiv: 2502.16759 · v3 · submitted 2025-02-24 · 💻 cs.IR

Can Explanations Improve Recommendations? Evidence from Prediction-Informed Explanations

Yuyan Wang , Pan Li , Minmin Chen This is my paper

Pith reviewed 2026-05-23 03:00 UTC · model grok-4.3

classification 💻 cs.IR
keywords recommender systemsexplainable AIlarge language modelsprediction-informed explanationsalternating trainingpoint-of-interest recommendationhuman evaluation
0
0 comments X

The pith

Explanations aligned with predictions can improve both recommendation accuracy and user preference.

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

The paper argues that explanations do not have to trade off against accuracy in recommender systems. By embedding LLM-generated explanations into an alternating training loop where predictions guide explanations and explanations refine predictions, the two components reinforce each other. A sympathetic reader would care because this reframes explainability as a performance lever rather than a constraint. The approach also shows strong data efficiency, matching top models with far less training data while producing explanations that humans prefer. The work grounds the mutual reinforcement in multi-environment statistical learning theory.

Core claim

RecPIE jointly optimizes recommendation predictions and natural-language explanations generated by LLMs by alternating between prediction-informed explanations and explanation-informed predictions. The LLM is fine-tuned via LoRA and reinforcement learning with a reward tied to recommendation accuracy. Drawing on multi-environment statistical learning theory, the framework shows that explanation generation and prediction can be mutually reinforcing rather than competing.

What carries the argument

The RecPIE alternating training loop that embeds LLM explanation generation into the prediction learning process, with predictions guiding explanations and explanations feeding back to refine predictions.

If this is right

  • Predictive accuracy rises 3-4% over state-of-the-art baselines on large-scale POI recommendation data.
  • The model matches the best baseline while using only 12% of the training data.
  • Human raters prefer RecPIE explanations 61.5% of the time versus 16.6% for the strongest baseline.
  • Explanations receive ratings closer to human-generated text than those from post-hoc methods.

Where Pith is reading between the lines

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

  • The alternating loop could be tested in other generative-plus-discriminative settings such as personalized search or content moderation where natural-language feedback might refine the underlying model.
  • The data-efficiency result suggests joint prediction-explanation training may lower labeling costs in additional recommendation or ranking domains.
  • If the mutual reinforcement holds, similar loops might reduce the need for separate post-hoc explanation modules in deployed AI systems.

Load-bearing premise

The LLM-generated explanations remain accurate and non-hallucinated enough throughout training to supply useful signal instead of noise.

What would settle it

Train the model with deliberately inaccurate or randomized explanations in the feedback loop and check whether accuracy still rises above baselines or falls back to them.

Figures

Figures reproduced from arXiv: 2502.16759 by Minmin Chen, Pan Li, Yuyan Wang.

Figure 3
Figure 3. Figure 3: Fig.3. As an example, here is the LLM-generated profile given the prompt above and the product name “JW [PITH_FULL_IMAGE:figures/full_fig_p018_3.png] view at source ↗
read the original abstract

Recommender systems are central to digital platforms, yet they face a fundamental trade-off between accuracy and explainability. Black-box models achieve strong performance but lack interpretability needed for trust and adoption. Existing explainable AI approaches either treat explanations as post-hoc or at the cost of accuracy. We challenge this view, proposing that explanations, when designed as an integral component of a system and aligned with prediction outcomes, can improve both interpretability and performance. We introduce RecPIE (Recommendation with Prediction-Informed Explanations), a framework that jointly optimizes recommendation predictions and natural-language explanations generated by LLMs. RecPIE embeds explanation generation into the learning loop: predictions guide explanation generation (prediction-informed explanations), which are fed back to refine subsequent predictions (explanation-informed predictions) via alternating training. The LLM is fine-tuned using LoRA and reinforcement learning with a customized reward derived from recommendation accuracy. Drawing on multi-environment statistical learning theory, we formally ground why explanation generation and prediction can be mutually reinforcing. We evaluate RecPIE on large-scale point-of-interest recommendation data from Google Maps, where user preferences span diverse place categories. RecPIE improves predictive accuracy by 3-4% over state-of-the-art baselines and matches the best performing model using only 12% of the training data. In human evaluations with 566 participants, RecPIE explanations are preferred 61.5% of the time (versus 16.6% for the best baseline) and rated closer to human-generated explanations. These results reframe explainability not as a constraint on performance but as a design lever for improving AI systems, with implications for trust, data efficiency, and marketplace deployment.

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.

Referee Report

3 major / 1 minor

Summary. The paper introduces RecPIE, a framework that jointly optimizes recommendation predictions and LLM-generated natural-language explanations via an alternating loop: predictions inform explanations, which are fed back to refine predictions through LoRA fine-tuning and RL with a reward derived from recommendation accuracy. It invokes multi-environment statistical learning theory to argue that the components are mutually reinforcing, and reports 3-4% accuracy gains over baselines, 12% data efficiency, and 61.5% human preference (vs. 16.6% for best baseline) on Google Maps POI data with 566 participants.

Significance. If the empirical gains and theoretical grounding hold after verification, the work would reframe explainability as a performance-enhancing design choice rather than a trade-off in recommender systems, with potential implications for data efficiency and user trust. The integration of LLM explanations into the training loop via RL is a timely contribution to cs.IR, though its impact hinges on resolving the gaps in theory mapping and experimental rigor noted below.

major comments (3)
  1. [Abstract] Abstract: the claim that multi-environment statistical learning theory formally grounds the alternating prediction-explanation loop supplies no explicit mapping of the two environments, no named theorem, and no verification that the LoRA+RL setup with accuracy-derived reward satisfies the theory's conditions on feedback variance or environment shift; this mapping is load-bearing for the mutual-reinforcement premise.
  2. [Abstract] Abstract (results): the reported 3-4% predictive accuracy lift and 12% data-efficiency result are presented without error bars, statistical significance tests, ablation studies isolating the explanation-feedback component, or confirmation that LLM explanations remain non-hallucinated and correlated with user preferences rather than injecting noise.
  3. [Abstract] Abstract: the reward signal is derived directly from recommendation accuracy, creating a potential circularity where explanation quality is judged by the same metric it is intended to improve; the multi-environment citation is invoked but not shown to establish independence of the prediction and explanation environments.
minor comments (1)
  1. [Abstract] Abstract: the human evaluation reports 61.5% preference but does not specify how explanations were presented to participants or whether controls for explanation length and style were applied.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and outline revisions where appropriate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that multi-environment statistical learning theory formally grounds the alternating prediction-explanation loop supplies no explicit mapping of the two environments, no named theorem, and no verification that the LoRA+RL setup with accuracy-derived reward satisfies the theory's conditions on feedback variance or environment shift; this mapping is load-bearing for the mutual-reinforcement premise.

    Authors: We agree the abstract omits the explicit mapping. The full manuscript references multi-environment statistical learning theory to justify mutual reinforcement, but we will revise by adding a dedicated subsection that (1) maps the prediction task to environment 1 and LLM explanation generation to environment 2, (2) names the invoked theorem on alternating optimization bounds, and (3) verifies that the LoRA+RL setup with accuracy reward meets the theory's conditions on feedback variance and limited environment shift. This addition will appear in the methodology section of the revision. revision: yes

  2. Referee: [Abstract] Abstract (results): the reported 3-4% predictive accuracy lift and 12% data-efficiency result are presented without error bars, statistical significance tests, ablation studies isolating the explanation-feedback component, or confirmation that LLM explanations remain non-hallucinated and correlated with user preferences rather than injecting noise.

    Authors: The experimental section already reports error bars, paired statistical tests, and ablations isolating the explanation-feedback component. Human preference results (61.5% vs. 16.6%) provide evidence that explanations align with user preferences rather than noise. To address the abstract specifically, we will add a clause noting statistical significance and ablation support, plus a brief statement on validation against hallucination via the preference study. No new experiments are required. revision: partial

  3. Referee: [Abstract] Abstract: the reward signal is derived directly from recommendation accuracy, creating a potential circularity where explanation quality is judged by the same metric it is intended to improve; the multi-environment citation is invoked but not shown to establish independence of the prediction and explanation environments.

    Authors: We maintain there is no circularity. The accuracy-derived reward optimizes only the explanation-generation policy within the RL step; the subsequent alternating step uses those explanations to update the separate prediction model. Multi-environment theory is invoked precisely to treat the two as distinct environments whose distributions differ, with the feedback loop creating reinforcement without metric identity. We will expand the theory paragraph to explicitly state this independence and the role of environment shift. revision: no

Circularity Check

0 steps flagged

No significant circularity in the derivation chain

full rationale

The paper presents RecPIE as an empirical framework using alternating prediction-explanation training and RL fine-tuning with a reward derived from recommendation accuracy, while citing multi-environment statistical learning theory for mutual reinforcement. No load-bearing steps reduce the claimed 3-4% accuracy gains or data-efficiency results to the inputs by construction. There are no self-definitional equations, fitted parameters renamed as predictions, or self-citations whose content is shown to be the sole justification for the central claim. The results rest on external evaluations (Google Maps data and 566-participant human study) rather than tautological mappings, satisfying the criteria for a self-contained derivation.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The central claim rests on the applicability of multi-environment statistical learning theory to the alternating loop and on the assumption that LLM explanations provide non-noisy supervisory signal; no new physical entities are introduced.

free parameters (2)
  • LoRA adaptation rank and learning rate
    Standard hyperparameters for LLM fine-tuning whose specific values affect the explanation quality and are chosen to optimize the joint objective.
  • Custom RL reward scaling coefficients
    Weights that balance recommendation accuracy against explanation properties; chosen during training.
axioms (1)
  • domain assumption Multi-environment statistical learning theory establishes that explanation generation and prediction can be mutually reinforcing
    Invoked to formally ground the alternating training procedure.

pith-pipeline@v0.9.0 · 5829 in / 1256 out tokens · 31238 ms · 2026-05-23T03:00:15.068998+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

105 extracted references · 105 canonical work pages · 14 internal anchors

  1. [1]

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

    ENTRY address author booktitle chapter doi edition editor eid howpublished institution isbn issn journal key month note number organization pages publisher school series title type url volume year label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block FUNCTION init.state.consts #0 'before.all := #1...

    work page

  2. [2]

    write newline

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

    work page

  3. [3]

    GPT-4 Technical Report

    Achiam J, Adler S, Agarwal S, Ahmad L, Akkaya I, Aleman FL, Almeida D, Altenschmidt J, Altman S, Anadkat S, et al. (2023) Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    work page internal anchor Pith review Pith/arXiv arXiv 2023

  4. [4]

    IEEE access 6:52138--52160

    Adadi A, Berrada M (2018) Peeking inside the black-box: a survey on explainable artificial intelligence (xai). IEEE access 6:52138--52160

    work page 2018

  5. [5]

    An efficient reduction of ranking to classification

    Ailon N, Mohri M (2007) An efficient reduction of ranking to classification. arXiv preprint arXiv:0710.2889

    work page internal anchor Pith review Pith/arXiv arXiv 2007

  6. [6]

    Machine learning 72:139--153

    Balcan MF, Bansal N, Beygelzimer A, Coppersmith D, Langford J, Sorkin GB (2008) Robust reductions from ranking to classification. Machine learning 72:139--153

    work page 2008

  7. [7]

    Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, 717--725

    Bauman K, Liu B, Tuzhilin A (2017) Aspect based recommendations: Recommending items with the most valuable aspects based on user reviews. Proceedings of the 23rd ACM SIGKDD international conference on knowledge discovery and data mining, 717--725

    work page 2017

  8. [8]

    Advances in neural information processing systems 24

    Bergstra J, Bardenet R, Bengio Y, K \'e gl B (2011) Algorithms for hyper-parameter optimization. Advances in neural information processing systems 24

    work page 2011

  9. [9]

    The Annals of Statistics 37(4):1705--1732

    Bickel PJ, Ritov Y, Tsybakov AB (2009) Simultaneous analysis of lasso and dantzig selector. The Annals of Statistics 37(4):1705--1732

    work page 2009

  10. [10]

    UAI '98 , 43--52

    Breese JS, Heckerman D, Kadie C (1998) Empirical analysis of predictive algorithms for collaborative filtering. UAI '98 , 43--52

    work page 1998

  11. [11]

    Language Models are Few-Shot Learners

    Brown TB (2020) Language models are few-shot learners. arXiv preprint arXiv:2005.14165

    work page internal anchor Pith review Pith/arXiv arXiv 2020

  12. [12]

    International conference on machine learning, 883--892 (PMLR)

    Chen J, Song L, Wainwright M, Jordan M (2018) Learning to explain: An information-theoretic perspective on model interpretation. International conference on machine learning, 883--892 (PMLR)

    work page 2018

  13. [13]

    WSDM '19 , 456--464

    Chen M, Beutel A, Covington P, Jain S, Belletti F, Chi EH (2019) Top-k off-policy correction for a reinforce recommender system. WSDM '19 , 456--464

    work page 2019

  14. [14]

    Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 385--396

    Chen Y, Truong QT, Shen X, Li J, King I (2024) Shopping trajectory representation learning with pre-training for e-commerce customer understanding and recommendation. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 385--396

    work page 2024

  15. [15]

    ACM Transactions on Information Systems (TOIS) 37(2):1--28

    Cheng Z, Chang X, Zhu L, Kanjirathinkal RC, Kankanhalli M (2019) Mmalfm: Explainable recommendation by leveraging reviews and images. ACM Transactions on Information Systems (TOIS) 37(2):1--28

    work page 2019

  16. [16]

    IJCAI, 3748--3754

    Cheng Z, Ding Y, He X, Zhu L, Song X, Kankanhalli MS (2018) A\^ 3ncf: An adaptive aspect attention model for rating prediction. IJCAI, 3748--3754

    work page 2018

  17. [17]

    Proceedings of the 27th ACM International conference on information and knowledge management, 147--156

    Chin JY, Zhao K, Joty S, Cong G (2018) Anr: Aspect-based neural recommender. Proceedings of the 27th ACM International conference on information and knowledge management, 147--156

    work page 2018

  18. [18]

    Learning Phrase Representations using RNN Encoder-Decoder for Statistical Machine Translation

    Cho K (2014) Learning phrase representations using rnn encoder-decoder for statistical machine translation. arXiv preprint arXiv:1406.1078

    work page internal anchor Pith review Pith/arXiv arXiv 2014

  19. [19]

    Companion Proceedings of the 23rd International Conference on Intelligent User Interfaces, 1--2

    Costa F, Ouyang S, Dolog P, Lawlor A (2018) Automatic generation of natural language explanations. Companion Proceedings of the 23rd International Conference on Intelligent User Interfaces, 1--2

    work page 2018

  20. [20]

    Recsys '16

    Covington P, Adams J, Sargin E (2016) Deep neural networks for youtube recommendations. Recsys '16

    work page 2016

  21. [21]

    Proceedings of the 15th ACM conference on recommender systems, 143--153

    de Souza Pereira Moreira G, Rabhi S, Lee JM, Ak R, Oldridge E (2021) Transformers4rec: Bridging the gap between nlp and sequential/session-based recommendation. Proceedings of the 15th ACM conference on recommender systems, 143--153

    work page 2021

  22. [22]

    BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin J, Chang MW, Lee K, Toutanova K (2019) Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805

    work page internal anchor Pith review Pith/arXiv arXiv 2019

  23. [23]

    Marketing science 40(6):1059--1080

    Dhillon PS, Aral S (2021) Modeling dynamic user interests: A neural matrix factorization approach. Marketing science 40(6):1059--1080

    work page 2021

  24. [24]

    The Llama 3 Herd of Models

    Dubey A, Jauhri A, Pandey A, Kadian A, Al-Dahle A, Letman A, Mathur A, Schelten A, Yang A, Fan A, et al. (2024) The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  25. [25]

    arXiv preprint arXiv:2303.03092

    Fan J, Fang C, Gu Y, Zhang T (2023) Environment invariant linear least squares. arXiv preprint arXiv:2303.03092

    work page arXiv 2023

  26. [26]

    National science review 1(2):293--314

    Fan J, Han F, Liu H (2014) Challenges of big data analysis. National science review 1(2):293--314

    work page 2014

  27. [27]

    Marketing Science

    Fong H, Kumar V, Sudhir K (2024) A theory-based explainable deep learning architecture for music emotion. Marketing Science

    work page 2024

  28. [28]

    Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 454--464

    Gholami E, Motamedi M, Aravindakshan A (2022) Parsrec: Explainable personalized attention-fused recurrent sequential recommendation using session partial actions. Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 454--464

    work page 2022

  29. [29]

    ACM Transactions on Mgmt Information Systems (TMIS) 6(4):1--19

    Gomez-Uribe CA, Hunt N (2015) The netflix recommender system: Algorithms, business value, and innovation. ACM Transactions on Mgmt Information Systems (TMIS) 6(4):1--19

    work page 2015

  30. [30]

    The Twelfth International Conference on Learning Representations

    Gu Y, Dong L, Wei F, Huang M (2024 a ) Minillm: Knowledge distillation of large language models. The Twelfth International Conference on Learning Representations

    work page 2024

  31. [31]

    arXiv preprint arXiv:2405.04715

    Gu Y, Fang C, B \"u hlmann P, Fan J (2024 b ) Causality pursuit from heterogeneous environments via neural adversarial invariance learning. arXiv preprint arXiv:2405.04715

    work page arXiv 2024

  32. [32]

    ACM Transactions on Information Systems (TOIS) 37(3):1--27

    Guan X, Cheng Z, He X, Zhang Y, Zhu Z, Peng Q, Chua TS (2019) Attentive aspect modeling for review-aware recommendation. ACM Transactions on Information Systems (TOIS) 37(3):1--27

    work page 2019

  33. [33]

    Recommender Systems Handbook: Third Edition, 547--601 (Springer US)

    Gunawardana A, Shani G, Yogev S (2022) Evaluating recommender systems. Recommender Systems Handbook: Third Edition, 547--601 (Springer US)

    work page 2022

  34. [34]

    Radiology 143(1):29--36

    Hanley JA, McNeil BJ (1982) The meaning and use of the area under a receiver operating characteristic (roc) curve. Radiology 143(1):29--36

    work page 1982

  35. [35]

    Proceedings of the 26th international conference on world wide web, 173--182

    He X, Liao L, Zhang H, Nie L, Hu X, Chua TS (2017) Neural collaborative filtering. Proceedings of the 26th international conference on world wide web, 173--182

    work page 2017

  36. [36]

    Proceedings of the 32nd ACM international conference on information and knowledge management, 720--730

    He Z, Xie Z, Jha R, Steck H, Liang D, Feng Y, Majumder BP, Kallus N, McAuley J (2023) Large language models as zero-shot conversational recommenders. Proceedings of the 32nd ACM international conference on information and knowledge management, 720--730

    work page 2023

  37. [37]

    Distilling the Knowledge in a Neural Network

    Hinton G (2015) Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  38. [38]

    science 313(5786):504--507

    Hinton GE, Salakhutdinov RR (2006) Reducing the dimensionality of data with neural networks. science 313(5786):504--507

    work page 2006

  39. [39]

    Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 585--593

    Hou Y, Mu S, Zhao WX, Li Y, Ding B, Wen JR (2022) Towards universal sequence representation learning for recommender systems. Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 585--593

    work page 2022

  40. [40]

    LoRA: Low-Rank Adaptation of Large Language Models

    Hu EJ, Shen Y, Wallis P, Allen-Zhu Z, Li Y, Wang S, Wang L, Chen W (2021) Lora: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685

    work page internal anchor Pith review Pith/arXiv arXiv 2021

  41. [41]

    arXiv preprint arXiv:2308.16505

    Huang X, Lian J, Lei Y, Yao J, Lian D, Xie X (2023) Recommender ai agent: Integrating large language models for interactive recommendations. arXiv preprint arXiv:2308.16505

    work page arXiv 2023

  42. [42]

    arXiv preprint arXiv:2407.02694

    Jeong DP, Lipton ZC, Ravikumar P (2024) Llm-select: Feature selection with large language models. arXiv preprint arXiv:2407.02694

    work page arXiv 2024

  43. [43]

    Mixtral of Experts

    Jiang AQ, Sablayrolles A, Roux A, Mensch A, Savary B, Bamford C, Chaplot DS, Casas Ddl, Hanna EB, Bressand F, et al. (2024) Mixtral of experts. arXiv preprint arXiv:2401.04088

    work page internal anchor Pith review Pith/arXiv arXiv 2024

  44. [44]

    ICDM '18 , 197--206 (IEEE)

    Kang WC, McAuley J (2018) Self-attentive sequential recommendation. ICDM '18 , 197--206 (IEEE)

    work page 2018

  45. [45]

    Auto-Encoding Variational Bayes

    Kingma DP (2013) Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  46. [46]

    Proceedings of the 14th ACM International Conference on Web Search and Data Mining, 967--975

    Le TH, Lauw HW (2021) Explainable recommendation with comparative constraints on product aspects. Proceedings of the 14th ACM International Conference on Web Search and Data Mining, 967--975

    work page 2021

  47. [47]

    Iui workshops

    Lee OJ, Jung JJ (2018) Explainable movie recommendation systems by using story-based similarity. Iui workshops

    work page 2018

  48. [48]

    Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 1248--1257

    Li J, He Z, Shang J, McAuley J (2023 a ) Ucepic: Unifying aspect planning and lexical constraints for generating explanations in recommendation. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 1248--1257

    work page 2023

  49. [49]

    arXiv preprint arXiv:2105.11601

    Li L, Zhang Y, Chen L (2021) Personalized transformer for explainable recommendation. arXiv preprint arXiv:2105.11601

    work page arXiv 2021

  50. [50]

    ACM Transactions on Information Systems 41(4):1--26

    Li L, Zhang Y, Chen L (2023 b ) Personalized prompt learning for explainable recommendation. ACM Transactions on Information Systems 41(4):1--26

    work page 2023

  51. [51]

    Proceedings of the 14th ACM Conference on Recommender Systems, 279--288

    Li P, Que M, Jiang Z, Hu Y, Tuzhilin A (2020) Purs: Personalized unexpected recommender system for improving user satisfaction. Proceedings of the 14th ACM Conference on Recommender Systems, 279--288

    work page 2020

  52. [52]

    Information Systems Research 35(3):1257--1273

    Li P, Tuzhilin A (2024) When variety seeking meets unexpectedness: Incorporating variety-seeking behaviors into design of unexpected recommender systems. Information Systems Research 35(3):1257--1273

    work page 2024

  53. [53]

    arXiv preprint arXiv:2306.01475

    Li P, Wang Y, Chi EH, Chen M (2023 c ) Prompt tuning large language models on personalized aspect extraction for recommendations. arXiv preprint arXiv:2306.01475

    work page arXiv 2023

  54. [54]

    (2023) How can recommender systems benefit from large language models: A survey

    Lin J, Dai X, Xi Y, Liu W, Chen B, Zhang H, Liu Y, Wu C, Li X, Zhu C, et al. (2023) How can recommender systems benefit from large language models: A survey. ACM Transactions on Information Systems

    work page 2023

  55. [55]

    arXiv preprint arXiv:2409.09415

    Lingo R, Arroyo M, Chhajer R (2024) Enhancing llm problem solving with reap: Reflection, explicit problem deconstruction, and advanced prompting. arXiv preprint arXiv:2409.09415

    work page arXiv 2024

  56. [56]

    (2009) Learning to rank for information retrieval

    Liu TY, et al. (2009) Learning to rank for information retrieval. Foundations and Trends in Information Retrieval 3(3):225--331

    work page 2009

  57. [57]

    A Unified Approach to Interpreting Model Predictions

    Lundberg S (2017) A unified approach to interpreting model predictions. arXiv preprint arXiv:1705.07874

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  58. [58]

    Meta AI Blog

    Meta A (2024) Introducing llama 3.1: Our most capable models to date. Meta AI Blog

    work page 2024

  59. [59]

    ACM Computing Surveys 55(13s):1--42

    Nauta M, Trienes J, Pathak S, Nguyen E, Peters M, Schmitt Y, Schl \"o tterer J, Van Keulen M, Seifert C (2023) From anecdotal evidence to quantitative evaluation methods: A systematic review on evaluating explainable ai. ACM Computing Surveys 55(13s):1--42

    work page 2023

  60. [60]

    Ni J, Li J, McAuley J (2019) Justifying recommendations using distantly-labeled reviews and fine-grained aspects. Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), 188--197

    work page 2019

  61. [61]

    Visual Explanation by Interpretation: Improving Visual Feedback Capabilities of Deep Neural Networks

    Oramas J, Wang K, Tuytelaars T (2017) Visual explanation by interpretation: Improving visual feedback capabilities of deep neural networks. arXiv preprint arXiv:1712.06302

    work page internal anchor Pith review Pith/arXiv arXiv 2017

  62. [62]

    Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, 2690--2696

    Pan D, Li X, Li X, Zhu D (2021) Explainable recommendation via interpretable feature mapping and evaluation of explainability. Proceedings of the Twenty-Ninth International Conference on International Joint Conferences on Artificial Intelligence, 2690--2696

    work page 2021

  63. [63]

    Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, 873--880

    Pathak Y, Laghuvarapu S, Mehta S, Priyakumar UD (2020) Chemically interpretable graph interaction network for prediction of pharmacokinetic properties of drug-like molecules. Proceedings of the AAAI Conference on Artificial Intelligence, volume 34, 873--880

    work page 2020

  64. [64]

    Normalization: A Preprocessing Stage

    Patro S (2015) Normalization: A preprocessing stage. arXiv preprint arXiv:1503.06462

    work page internal anchor Pith review Pith/arXiv arXiv 2015

  65. [65]

    Peng W, Xu D, Xu T, Zhang J, Chen E (2023) Are gpt embeddings useful for ads and recommendation? International Conference on Knowledge Science, Engineering and Management, 151--162 (Springer)

    work page 2023

  66. [66]

    Journal of the Royal Statistical Society Series B: Statistical Methodology 78(5):947--1012

    Peters J, B \"u hlmann P, Meinshausen N (2016) Causal inference by using invariant prediction: identification and confidence intervals. Journal of the Royal Statistical Society Series B: Statistical Methodology 78(5):947--1012

    work page 2016

  67. [67]

    (2019) Language models are unsupervised multitask learners

    Radford A, Wu J, Child R, Luan D, Amodei D, Sutskever I, et al. (2019) Language models are unsupervised multitask learners. OpenAI blog 1(8):9

    work page 2019

  68. [68]

    Marketing Science

    Rafieian O, Kapoor A, Sharma A (2024) Multiobjective personalization of marketing interventions. Marketing Science

    work page 2024

  69. [69]

    Available at SSRN

    Ragodos R, Wang T, Feng L, Hu YJ (2024) The risk of inferring data insights from post hoc explanations of machine learning models. Available at SSRN

    work page 2024

  70. [70]

    Proceedings of the ACM on Web Conference 2024, 3464--3475

    Ren X, Wei W, Xia L, Su L, Cheng S, Wang J, Yin D, Huang C (2024) Representation learning with large language models for recommendation. Proceedings of the ACM on Web Conference 2024, 3464--3475

    work page 2024

  71. [71]

    Recommender systems handbook, 1--35 (Springer)

    Ricci F, Rokach L, Shapira B (2010) Introduction to recommender systems handbook. Recommender systems handbook, 1--35 (Springer)

    work page 2010

  72. [72]

    Proceedings of the 1st ACM conference on Electronic commerce, 158--166

    Schafer JB, Konstan J, Riedl J (1999) Recommender systems in e-commerce. Proceedings of the 1st ACM conference on Electronic commerce, 158--166

    work page 1999

  73. [73]

    arXiv preprint arXiv:2009.07118

    Schick T, Sch \"u tze H (2020) It's not just size that matters: Small language models are also few-shot learners. arXiv preprint arXiv:2009.07118

    work page arXiv 2020

  74. [74]

    The Annals of Statistics 48(4):1875--1897, ://dx.doi.org/10.1214/19-AOS1875

    Schmidt-Hieber J (2020) Nonparametric regression using deep neural networks with relu activation function. The Annals of Statistics 48(4):1875--1897, ://dx.doi.org/10.1214/19-AOS1875

    work page doi:10.1214/19-aos1875 2020

  75. [75]

    Preprint, submitted August 3:2024

    Sisodia A, Burnap A, Kumar V (2024) Generative interpretable visual design: Using disentanglement for visual conjoint analysis. Preprint, submitted August 3:2024

    work page 2024

  76. [76]

    CNET, January 10

    Solsman JE (2018) Youtube’s ai is the puppet master over most of what you watch. CNET, January 10

    work page 2018

  77. [77]

    arXiv preprint arXiv:2005.00724

    Subramanian S, Bogin B, Gupta N, Wolfson T, Singh S, Berant J, Gardner M (2020) Obtaining faithful interpretations from compositional neural networks. arXiv preprint arXiv:2005.00724

    work page arXiv 2020

  78. [78]

    CIKM '19 , 1441--1450

    Sun F, Liu J, Wu J, Pei C, Lin X, Ou W, Jiang P (2019) Bert4rec: Sequential recommendation with bidirectional encoder representations from transformer. CIKM '19 , 1441--1450

    work page 2019

  79. [79]

    arXiv preprint arXiv:2402.09702

    Sun Y, Chen Z, Orlandi V, Wang T, Rudin C (2024) Sparse and faithful explanations without sparse models. arXiv preprint arXiv:2402.09702

    work page arXiv 2024

  80. [80]

    Proceedings of the 27th International Conference on Neural Information Processing Systems-Volume 2, 3104--3112

    Sutskever I, Vinyals O, Le QV (2014) Sequence to sequence learning with neural networks. Proceedings of the 27th International Conference on Neural Information Processing Systems-Volume 2, 3104--3112

    work page 2014

Showing first 80 references.