Pith. sign in

REVIEW 4 major objections 5 minor 8 cited by

Enhanced Recommendation Combining Collaborative Filtering and Large Language Models

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper proposes a hybrid recommender that adds an LLM's semantic item score to a collaborative filtering dot product and reports better precision, recall, coverage, and user satisfaction on MovieLens and Amazon data.

desk verdict A poorly specified hybrid recommendation paper whose central equation is undefined; the reader's reject verdict is correct, and the dimensional inconsistency is an even more basic flaw than the missing code. read the letter →

arxiv 2412.18713 v1 pith:WSL67VIZ submitted 2024-12-25 cs.AI cs.IR

Xueting Lin , Zhan Cheng , Longfei Yun , Qingyi Lu , Yuanshuai Luo This is my paper · ORCID
classification cs.AIcs.IR
keywords collaborativefilteringlargelanguagemodelshybridrecommendationmatrixfactorizationcoldstartdatasparsitysemanticitemembeddingevaluation
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper is trying to show that a recommendation model gets measurably better when a large language model's semantic reading of item text is added, as a weighted term, to the classic collaborative filtering score. The proposed score is $\hat{y}_{ui} = P_u \cdot Q_i + \alpha F_{\mathrm{LLM}}(i)$, where the first term captures user-item behavior and the second captures item meaning from descriptions and reviews. On the MovieLens and Amazon Product Review datasets, the hybrid model is reported to beat both a pure collaborative filtering model and a pure LLM model on precision, recall, coverage, and a simulated user-satisfaction score. The practical stake is a simple fusion recipe that also supplies a cold-start path, recommending items through the LLM text embedding when interaction history is missing.

What carries the argument

The load-bearing object is the additive fusion equation $\hat{y}_{ui} = P_u \cdot Q_i + \alpha F_{\mathrm{LLM}}(i)$, together with the cold-start variant $\hat{y}_{ui} = P_u \cdot E_i$ built from the LLM text embedding $E_i = F_{\mathrm{LLM}}(T_i)$. This equation is what lets the model keep collaborative filtering's behavioral latent factors while injecting semantic information from item text; the single hyperparameter $\alpha$ controls the trade-off, and the MSE-plus-regularization loss is what trains the combined parameters. All of the paper's reported gains are attributed to this additive combination, so the identity itself is the machinery carrying the argument.

What would settle it

Re-run the MovieLens and Amazon comparisons with real held-out user ratings as the satisfaction measure, or run a small human study where users choose between hybrid and collaborative-filtering-only lists; if the hybrid does not beat pure collaborative filtering under real feedback, the paper's central claim is falsified.

Watch

Extended reading notes

Core claim

The central claim is that the additive score $\hat{y}_{ui} = P_u \cdot Q_i + \alpha F_{\mathrm{LLM}}(i)$ is a better predictor of user preference than either term alone, where $P_u$ is the user latent factor, $Q_i$ is the item latent factor, $F_{\mathrm{LLM}}(i)$ is an LLM-derived semantic feature vector for item $i$, and $\alpha$ balances the two. In the paper's experiments the hybrid model reaches 75.6% precision, 72.1% recall, 52.6% coverage, and 4.3/5 simulated satisfaction on MovieLens at $\alpha = 0.5$, against 72.3%, 68.9%, 43.2%, and 4.0 for pure collaborative filtering; the Amazon results follow the same pattern at $\alpha = 0.7$. The paper also claims that for cold-start items the prediction $\hat{y}_{ui} = P_u \cdot E_i$, using the LLM text embedding $E_i$ in place of the learned item factor, can produce recommendations without historical interaction data, with the whole model trained by minimizing MSE plus an $L_2$ regularization term over $P_u$, $Q_i$, and $E_i$.

Load-bearing premise

The evaluation assumes that the simulated user-interaction scores, which are generated rather than measured from real users, faithfully reflect genuine user preferences; if that simulation is unrealistic, the reported satisfaction gains for the hybrid model do not follow.

Editorial extensions

If this is right

  • If the hybrid score is better on MovieLens and Amazon, the same weighted addition should improve recommendation in any domain with item text, such as news, e-commerce, or video descriptions.
  • Because coverage rises from 43.2% to 52.6% on MovieLens, the model can surface long-tail items that pure collaborative filtering misses, changing the diversity of what users are shown.
  • For cold-start items, the $E_i = F_{\mathrm{LLM}}(T_i)$ path allows a new item to be scored against an existing user embedding without waiting for ratings to accumulate.
  • The reported sensitivity to $\alpha$ implies operators can tune the balance per scenario, leaning more on text in sparse-data settings and more on behavior in dense-data settings.
  • Joint training of $P_u$, $Q_i$, and LLM-related parameters under the MSE loss means the semantic features are adapted to the rating signal rather than frozen, so the two information sources are combined with feedback from the recommendation objective.

Reading between the lines

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

  • A testable extension the paper does not pursue is comparing a frozen pretrained LLM embedding with the jointly fine-tuned version inside $\hat{y}_{ui} = P_u \cdot Q_i + \alpha F_{\mathrm{LLM}}(i)$; the outcome would show whether joint training is what drives the gain.
  • Because the additive equation combines two quantities that may live on different scales, a learned normalization or gating layer is a natural next step the paper leaves implicit.
  • The satisfaction metric is simulated, so the direct human-subject follow-up is the obvious way to see whether the reported 4.3 versus 4.0 gap survives outside the simulation.
  • The architecture's item embeddings can be precomputed offline, so the practical deployment question the paper does not quantify is the added inference latency of the LLM term relative to the collaborative filtering score.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes a hybrid recommendation method that adds an LLM-derived semantic feature term to a collaborative filtering score. Section 3.2 defines the prediction as y_hat_ui = P_u · Q_i + α · F_LLM(i), with F_LLM(i) described as an LLM-extracted semantic feature vector, and includes a purported cold-start variant based on text embeddings. The experimental section (Section 4) reports precision, recall, coverage, and a simulated 'user satisfaction' metric on MovieLens and Amazon Product Review datasets, claiming that the hybrid model outperforms both pure collaborative filtering and pure LLM recommendation. The conclusion (Section 5) repeats the claim that the hybrid model significantly improves accuracy, diversity, and user satisfaction.

Significance. If the central construction and evaluation were sound, the paper would present a straightforward and potentially useful fusion strategy for combining collaborative filtering with LLM embeddings. The paper does state a concrete fusion formula and reports a small α-sweep, and it is explicit about the datasets used. However, the claimed result is not supported by the evidence as presented. The core formula is dimensionally inconsistent, the experimental setup omits the LLM, embedding model, training details, and any measure of variance, and the user satisfaction metric is based on an undescribed and unvalidated simulation. These are not presentation issues; they prevent the empirical claims from being attached to the stated model. The contribution, as it stands, is not reproducible from the manuscript.

major comments (4)
  1. [Section 3.2, Eq. (3)] Eq. (3) defines y_hat_ui = P_u · Q_i + α · F_LLM(i), but P_u · Q_i is a scalar while F_LLM(i) is explicitly called a semantic feature vector for item i. Adding a vector to a scalar is not well defined. The cold-start equation immediately following, y_hat_ui = P_u · E_i with E_i = F_LLM(T_i), suggests that the intended hybrid term may have been α(P_u · E_i), but the paper never states this, never specifies a projection or scalarization of F_LLM(i), and never defines the dimensionality of the vectors involved. Consequently, the quantitative results in Tables 1 and 2 cannot be traced to the formula given in Eq. (3), and the α-sweep in Table 2 is uninterpretable because the meaning of α is undefined.
  2. [Section 4, Tables 1 and 2] The experimental evaluation reports single point estimates for precision, recall, coverage, and user satisfaction, with no standard deviations, confidence intervals, significance tests, or repeated runs. The reported improvements, e.g., precision 75.6% vs. 72.3% and 70.1% on MovieLens, may be within noise, but the paper provides no way to assess this. Moreover, Table 1 reports only the best α for each dataset (α=0.5 for MovieLens and α=0.7 for Amazon) after Table 2 shows that performance varies with α; this post hoc selection on the test data is not described as a validation procedure, and the manuscript does not report which α values would be chosen by a principled model-selection rule.
  3. [Section 4, Evaluation Metrics] The paper states that 'User Satisfaction was rated by simulating user interactions with the recommended items' but provides no simulator, no interaction model, no definition of the simulation procedure, and no validation against actual user judgments. This is a load-bearing metric because the abstract and conclusion claim that the hybrid model improves 'user satisfaction,' yet the manuscript gives no evidence that the simulated ratings correspond to real user preferences. Without such evidence, the satisfaction claims in Tables 1 and 2 and in Section 5 are unsupported.
  4. [Section 4, Model Training] The experimental section does not identify the LLM used, the embedding method or dimensionality, the matrix factorization implementation, the optimization algorithm, the learning rate, the regularization coefficient λ, the latent vector dimensionality, or the training schedule. The only hyperparameter discussed is α. This is insufficient for reproducibility and makes it impossible to determine whether the reported hybrid results are due to the proposed fusion or to unspecified implementation choices.
minor comments (5)
  1. [Section 1, paragraph 1] The phrase 'digitalization,,' contains a doubled comma; please revise.
  2. [Section 2, paragraph 2] The sentence beginning 'Wen Jun Gu et al.'s[7] approach...' is grammatically incomplete and should be rewritten.
  3. [Section 3.2, Eq. (6)] The regularization term mixes notation: ||P_u||^2 + ||Q_i||^2 + ||E_i||^2_2 uses a subscript '2' only on the last term, and the intended norms should be stated consistently.
  4. [Section 4, Table 1] The table caption and text refer to 'User Satisfaction (/5)' but do not explain how a simulated rating on a 1–5 scale was produced; adding a reference to a concrete simulation protocol or removing the metric would clarify the claim.
  5. [References] Several references cited in the introduction and related work sections (e.g., Refs. [1], [2], [5], [7]) concern topics such as machine translation, autonomous driving, fraud detection, and stock prediction without a clear connection to recommendation systems; the motivation for citing them should be stated or the citations should be removed.

Circularity Check

1 steps flagged · score 6.0 of 10

The reported hybrid-model gains are selected maxima of the α sweep on the same evaluation data, so the central empirical claim is partially circular; the model formula itself is not definitionally circular.

  1. fitted input called prediction [Section 4, 'Evaluation Metrics' and 'Impact of Weight Parameter', Tables 1 and 2]
    "We experimented with different weight parameters α to adjust the contribution of collaborative filtering and LLM to find the best combination strategy. ... From the data in Table 2, it is clear that the hybrid model combining collaborative filtering and LLM outperforms individual models across several key metrics. ... When α = 0.5, the model achieved the best balance in terms of precision, recall, and coverage."

    The headline Hybrid Model rows in Table 1 are numerically identical to the best rows of the α sweep in Table 2 (α=0.5 for MovieLens, α=0.7 for Amazon). The paper describes the sweep as a search ('to find the best combination strategy') and draws its conclusion that the hybrid outperforms CF and LLM from that same swept table. Hence the reported 'hybrid model' is not a fixed model tested out-of-sample; it is the argmax over α on the same metrics used to declare success. The improvement over baselines is therefore a fitted maximum, not an independent prediction. This is partial circularity of the empirical claim, not of the model definition (Eq. 3).

full rationale

The paper's mathematical core is an explicit construction, not a derivation from its own outputs; Eq. (3) is a proposed weighted fusion, and the vector/scalar mismatch noted by the skeptic is a correctness defect rather than a circular reduction. No load-bearing self-citation is present: the many references are to unrelated prior work. The only identifiable circular step is the evaluation of the hybrid model: α is tuned over a grid and the best row is then presented as 'the hybrid model' and used as evidence that the hybrid significantly improves precision, recall, coverage, and satisfaction. Because the same metrics used to select α are the metrics used to validate the claim, the reported superiority is in-sample. The undefined 'User Satisfaction was rated by simulating user interactions' is a serious validity concern, but the paper gives no detail showing the simulator is derived from the model's own outputs, so it cannot be scored as circularity without speculation.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The paper introduces no new entities. Its central claim relies on standard but unstated modeling choices: that LLM embeddings are linearly additive to CF scores, that MSE is the right loss, and that a simulated user interaction can stand in for real user satisfaction. The alpha and lambda parameters are fitted or unspecified, and no independent evidence supports the simulation.

free parameters (4)
  • alpha (fusion weight) = 0.5 for MovieLens, 0.7 for Amazon Product Review
    Introduced in Eq. (3) to balance collaborative filtering and LLM contributions. Values were chosen per dataset, with only the best-performing values reported in Table 1.
  • lambda (regularization coefficient) = unspecified
    Introduced in Eq. (6) to control model complexity, but no value or selection procedure is reported.
  • Latent vector dimensionality for P_u and Q_i = unspecified
    The dimension of the user and item embeddings is never stated, though it determines the behavior of the dot products in Eqs. (2) and (3).
  • LLM choice and embedding configuration = unspecified
    The paper refers to a pretrained Transformer but does not identify the model, prompt format, or embedding dimension. These choices affect F_LLM(i) and therefore Eq. (3).
assumptions (4)
  • domain assumption MSE loss between predicted and actual ratings is an appropriate training objective for recommendation quality.
    Invoked in Section 3.2, Eq. (5). No justification is given for why MSE on ratings reflects user satisfaction or ranking quality.
  • domain assumption LLM text embeddings F_LLM(i) capture preference-relevant semantic features of items.
    Used throughout Section 3 to justify adding F_LLM(i) to the collaborative filtering score. The paper provides no evidence that these embeddings are aligned with user preference geometry.
  • ad hoc to paper User satisfaction can be measured by simulated user interactions.
    Section 4 states that user satisfaction was rated by simulating user interactions, but no simulator or validation is described. This is a paper-specific assumption with no independent support.
  • domain assumption A 70/15/15 train/validation/test split guarantees fair comparison across models.
    Section 4, Data Preparation. The paper assumes this split is sufficient for fair comparison without reporting variance or significance testing.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhanced Recommendation Combining Collaborative Filtering and Large Language Models." pith.science (2026). https://pith.science/paper/WSL67VIZ

@misc{pith2026241218713,
  author       = {Pith},
  title        = {Pith review of: Enhanced Recommendation Combining Collaborative Filtering and Large Language Models},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WSL67VIZ}},
  note         = {Machine review of arXiv:2412.18713}
}
read the original abstract

With the advent of the information explosion era, the importance of recommendation systems in various applications is increasingly significant. Traditional collaborative filtering algorithms are widely used due to their effectiveness in capturing user behavior patterns, but they encounter limitations when dealing with cold start problems and data sparsity. Large Language Models (LLMs), with their strong natural language understanding and generation capabilities, provide a new breakthrough for recommendation systems. This study proposes an enhanced recommendation method that combines collaborative filtering and LLMs, aiming to leverage collaborative filtering's advantage in modeling user preferences while enhancing the understanding of textual information about users and items through LLMs to improve recommendation accuracy and diversity. This paper first introduces the fundamental theories of collaborative filtering and LLMs, then designs a recommendation system architecture that integrates both, and validates the system's effectiveness through experiments. The results show that the hybrid model based on collaborative filtering and LLMs significantly improves precision, recall, and user satisfaction, demonstrating its potential in complex recommendation scenarios.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 8 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Instructional Prompt Optimization for Few-Shot LLM-Based Recommendations on Cold-Start Users

    cs.AI 2025-09 reject novelty 3.0 of 10

    A manuscript claims instructional prompt engineering improves LLM-based cold-start recommendation, but provides no reproducible evidence.

  2. Multimodal Foundation Model-Driven User Interest Modeling and Behavior Analysis on Short Video Platforms

    cs.IR 2025-09 reject novelty 3.0 of 10

    A standard attention-fusion plus Transformer sequence model is applied to short-video recommendation, with claimed gains over weak baselines and no reproducible artifacts.

  3. Meta-Learning for Cold-Start Personalization in Prompt-Tuned LLMs

    cs.LG 2025-07 reject novelty 3.0 of 10

    A meta-learned prompt-tuning method for cold-start LLM recommendations reports better Hit@10 and nDCG@10 on MovieLens-1M, but with no code, no error bars, and no shown results for Amazon or Recbole.

  4. Research on Low-Latency Inference and Training Efficiency Optimization for Graph Neural Network and Large Language Model-Based Recommendation Systems

    cs.LG 2025-06 reject novelty 3.0 of 10

    A hybrid GNN-LLM recommender with FPGA, DeepSpeed, and LoRA reportedly reaches NDCG@10 of 0.75 at 40-60ms latency while cutting training time by 66%, but the supporting artifacts are absent.

  5. LLM-Augmented Symptom Analysis for Cardiovascular Disease Risk Prediction: A Clinical NLP

    cs.CL 2025-07 reject novelty 2.0 of 10

    A small synthetic study reports that Bio_ClinicalBERT embeddings with Random Forest classify CVD risk in about 20 hand-written symptom texts, but the claims of MIMIC-III and CARDIO-NLP evaluation are unsupported.

  6. Research on Model Parallelism and Data Parallelism Optimization Methods in Large Language Model-Based Recommendation Systems

    cs.DC 2025-06 reject novelty 2.0 of 10

    A hybrid model-plus-data parallel scheme is reported to boost training throughput and GPU utilization for LLM-based recommenders, but the supporting experiments are not reproducible from the paper.

  7. Deep Learning Model Acceleration and Optimization Strategies for Real-Time Recommendation Systems

    cs.IR 2025-06 reject novelty 2.0 of 10

    A standard combination of model compression and serving optimization gives 2.4x throughput on a GPU benchmark, but the headline claims of <30% latency and preserved accuracy are not supported by the paper's own data.

  8. Research on Personalized Financial Product Recommendation by Integrating Large Language Models and Graph Neural Networks

    cs.IR 2025-06 reject novelty 2.0 of 10

    A hybrid LLM-plus-GNN recommender is claimed to beat collaborative filtering, LLM-only, and GNN-only baselines on financial product ranking, with NDCG@10 of 0.372.

Reference graph

Works this paper leans on

25 extracted references · 14 canonical work pages · cited by 8 Pith papers

  1. [1]

    Attention mechanism and context modeling system for text mining machine translation

    Bo, Shi, et al. "Attention mechanism and context modeling system for text mining machine translation." 2024 6th International Conference on Data-driven Optimization of Complex Systems (DOCS). IEEE, 2024

    work page 2024

  2. [2]

    Research on autonomous driving decision -making strategies based deep reinforcement learning

    Wang, Zixiang, et al. "Research on autonomous driving decision -making strategies based deep reinforcement learning." Proceedings of the 2024 4th International Conference on Internet of Things and Machine Learning. 2024

    work page 2024

  3. [3]

    Applications of explainable ai in natural language processing

    Xu, Qiming, et al. "Applications of explainable ai in natural language processing." Global Academic Frontiers 2.3 (2024): 51 -64

    work page 2024

  4. [4]

    An ensemble approach to stock price prediction using deep learning and time series models

    Sui, Mujie, et al. "An ensemble approach to stock price prediction using deep learning and time series models." (2024)

    2024

  5. [5]

    Dynamic fraud detection: Integrating reinforcement learning into graph neural networks

    Dong, Yuxin, et al. "Dynamic fraud detection: Integrating reinforcement learning into graph neural networks." 2024 6th International Conference on Data-driven Optimization of Complex Systems (DOCS). IEEE, 2024

    work page 2024

  6. [6]

    A Neural Matrix Decomposition Recommender System Model based on the Multimodal Large Language Model

    Xiang, Ao, et al. "A neural matrix decomposition recommender system model based on the multimodal large language model." arXiv preprint arXiv:2407.08942 (2024)

    work page Pith review arXiv 2024

  7. [7]

    Predicting stock prices with finbert-lstm: Integrating news sentiment analysis

    jun Gu, Wen, et al. "Predicting stock prices with finbert-lstm: Integrating news sentiment analysis." Proceedings of the 2024 8th International Conference on Cloud and Big Data Computing. 2024

    work page 2024

  8. [8]

    Research on improved u-net based remote sensing image segmentation algorithm

    Yang, Qiming, et al. "Research on improved u-net based remote sensing image segmentation algorithm." 2024 6th International Conference on Internet of Things, Automation and Artificial Intelligence (IoTAAI). IEEE, 2024

    work page 2024

Show all 25 references
  1. [9]

    Machine learning -based research on the adaptability of adolescents to online education

    Wang, Mingwei, and Sitong Liu. "Machine learning -based research on the adaptability of adolescents to online education." arXiv preprint arXiv:2408.16849 (2024)

    2024 arXiv

  2. [10]

    Deep learning solutions for pneumonia detection: Performance comparison of custom and transfer learning models

    Zhong, Yihao, et al. "Deep learning solutions for pneumonia detection: Performance comparison of custom and transfer learning models." International Conference on Automation and Intelligent Technology (ICAIT 2024). Vol. 13401. SPIE, 2024.1

    2024

  3. [11]

    Exploiting Diffusion Prior for Out -of-Distribution Detection

    Zhu, Armando, et al. "Exploiting Diffusion Prior for Out -of-Distribution Detection." arXiv preprint arXiv:2406.11105 (2024)

    2024 arXiv

  4. [12]

    Predicting Employee Turnover in the Financial Company: A Comparative Study of CatBoost and XGBoost Model s[J]

    Yin Z, Hu B, Chen S. Predicting Employee Turnover in the Financial Company: A Comparative Study of CatBoost and XGBoost Model s[J]. 2024

    2024

  5. [13]

    A Mixed-Heuristic Quantum-Inspired Simplified Swarm Optimization Algorithm for scheduling of real -time tasks in the multiprocessor system

    Su, Pei-Chiang, et al. "A Mixed-Heuristic Quantum-Inspired Simplified Swarm Optimization Algorithm for scheduling of real -time tasks in the multiprocessor system." Applied Soft Computing 131 (2022): 109807

    2022

  6. [14]

    Deep Learning for Weather Forecasting: A CNN -LSTM Hybrid Model for Predicting Historical Temperature Data

    Gong, Yuhao, et al. "Deep Learning for Weather Forecasting: A CNN -LSTM Hybrid Model for Predicting Historical Temperature Data." arXiv preprint arXiv:2410.14963 (2024)

    2024 arXiv

  7. [15]

    Research on Effectiveness Evaluation and Optimization of Baseball Teaching Method Based on Machine Learning[J]

    Sun S, Yuan J, Yang Y. Research on Effectiveness Evaluation and Optimization of Baseball Teaching Method Based on Machine Learning[J]. arXiv preprint arXiv:2411.15721, 2024

    2024 arXiv

  8. [16]

    Adaptive receptive field U -shaped temporal convolutional network for vulgar action segmentation

    Cao, Jin, et al. "Adaptive receptive field U -shaped temporal convolutional network for vulgar action segmentation." Neural Computing and Applications 35.13 (2023): 9593-9606

    2023

  9. [17]

    Fine -grained imbalanced leukocyte classification with global -local attention transformer

    Chen, Ben, et al. "Fine -grained imbalanced leukocyte classification with global -local attention transformer." Journal of King Saud University - Computer and Information Sciences 35.8 (2023): 101661

    2023

  10. [18]

    Harnessing llms for api interactions: A framework for classification and synthetic data generation[J]

    Tao C, Fan X, Yang Y. Harnessing llms for api interactions: A framework for classification and synthetic data generation[J]. arXiv preprint arXiv:2409.11703, 2024

    2024 arXiv

  11. [19]

    Towards resilient and efficient llms: A comparative study of efficiency, performance, and adversarial robustness[J]

    Fan X, Tao C. Towards resilient and efficient llms: A comparative study of efficiency, performance, and adversarial robustness[J]. arXiv preprint arXiv:2408.04585, 2024

    2024 arXiv

  12. [20]

    Application of Natural Language Processing in Financial Risk Detection

    Wang, Liyang, et al. "Application of Natural Language Processing in Financial Risk Detection." arXiv preprint arXiv:2406.09765 (2024)

    2024 arXiv

  13. [21]

    Research on splicing image detection algorithms based on natural image statistical characteristics

    Xiang, Ao, et al. "Research on splicing image detection algorithms based on natural image statistical characteristics." arXiv preprint arXiv:2404.16296 (2024)

    2024 arXiv

  14. [22]

    Autonomous navigation of unmanned vehicle through deep reinforcement learning

    Xu, Letian, et al. "Autonomous navigation of unmanned vehicle through deep reinforcement learning." arXiv preprint arXiv:2407.18962 (2024)

    2024 arXiv

  15. [23]

    Deep Learning Powered Estimate of The Extrinsic Parameters on Unmanned Surface Vehicles

    Shen, Yi, et al. "Deep Learning Powered Estimate of The Extrinsic Parameters on Unmanned Surface Vehicles." arXiv preprint arXiv:2406.04821 (2024)

    2024 arXiv

  16. [24]

    Mitigating Knowledge Conflicts in Language Model -Driven Question Answering

    Cao, Han, et al. "Mitigating Knowledge Conflicts in Language Model -Driven Question Answering." arXiv preprint arXiv:2411.11344 (2024)

    2024 arXiv

  17. [25]

    Improving text -image matching with adversarial learning and circle loss for multi -modal steganography

    Hu, Yuting, et al. "Improving text -image matching with adversarial learning and circle loss for multi -modal steganography." International Workshop on Digital Watermarking. Cham: Springer International Publishing, 2020

    2020

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.