Generative Ghost: Investigating Ranking Bias Hidden in AI-Generated Videos

Haowen Gao; Huawei Shen; Leigang Qu; Liang Pang; Shicheng Xu; Tat-Seng Chua; Xueqi Cheng

REVIEW 4 major objections 6 minor 40 references

Generative Ghost: Investigating Ranking Bias Hidden in AI-Generated Videos

T0 review · 4 major / 6 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read This paper claims that text-video retrieval models rank AI-generated videos above equally relevant real videos, calls the effect Visual-Temporal Induced Source Bias, and shows that mixing AI videos into training data strengthens it while…

desk verdict Useful benchmark, likely real effect, but the headline bias magnitudes rest on an unmet equal-relevance assumption and the strongest model shows mixed results. read the letter →

arxiv 2502.07327 v2 pith:A2XFNXTH submitted 2025-02-11 cs.IR cs.CV

Haowen Gao , Liang Pang , Shicheng Xu , Leigang Qu , Tat-Seng Chua , Huawei Shen , Xueqi Cheng This is my paper · ORCID

classification cs.IRcs.CV

keywords text-videoretrievalAI-generatedcontentsourcebiasVisual-TemporalInducedfairnesscontrastivedebiasingbenchmarkconstructiontemporalinformation

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 argues that text-video retrieval models systematically rank AI-generated videos above real videos that carry the same relevance label, and that this preference comes from extra visual and temporal signals that generative models embed in their output. The authors construct a benchmark pairing real videos from MSR-VTT with AI videos made by two open-source generators, and they measure retrieval with a new metric, Normalized $\Delta$, designed to separate source bias from quality and semantic gaps. Across three retrieval models and four test sets, the AI videos rank higher, and the preference strengthens as the share of AI-generated videos in the training set grows from 0% to 80%. A contrastive fine-tuning step mostly reverses the preference. If the claim holds, any retrieval system indexing a mixed real-and-AI library will silently over-expose synthetic content and may feed that bias back into future training data.

What carries the argument

The quantitative core is the Normalized $\Delta$ metric, defined as Relative $\Delta$ minus Location $\Delta$. Relative $\Delta$ compares the mixed-retrieval rankings of real and AI videos, while Location $\Delta$ simulates a no-interference baseline by interleaving independently computed rankings with a random offset, so the subtraction is meant to isolate genuine source bias from semantic or quality gaps. The supporting machinery includes frame-shuffle and single-frame ablations that separate the temporal and visual components of the bias, and a contrastive objective $\Delta r = E_V(V_G) - E_V(V_R)$ that is added to the retrieval loss to push real videos above AI-generated ones.

What would settle it

Filter the benchmark to pairs that human annotators judge equally relevant (or replace equal-relevance labels with human relevance scores), recompute Normalized $\Delta$ on that subset for the same three retrieval models, and check whether AI-generated videos still rank above real videos; if the preference disappears, the paper's central claim fails on its own benchmark.

Watch

Extended reading notes

Core claim

On the paper's own terms, text-video retrieval models are biased toward AI-generated videos: for a query, an AI video that shares the relevance label of its real counterpart is ranked above that counterpart across essentially all model-dataset combinations. The authors name this effect Visual-Temporal Induced Source Bias and locate its causes in both visual content and temporal structure, showing that randomizing frame order reduces the temporal contribution and that single-frame retrieval leaves a residual visual preference. The bias grows monotonically as the proportion of AI-generated videos in the training set increases, with a visible effect already at 20% AI content. The authors attribute the bias to additional, highly consistent information that video generators encode into their outputs, and they demonstrate a contrastive debiasing objective that substantially reduces the preference by shifting scores toward real videos.

Load-bearing premise

The benchmark labels every AI-generated video as equally relevant to its real counterpart, but the paper's own human evaluation judged the real video more relevant in 32% to 47% of pairs (versus 13% to 18% for the AI video), so the equal-relevance premise underlies all reported bias magnitudes.

Editorial extensions

If this is right

Video search engines that index mixed libraries will systematically place AI-generated clips above equally relevant real footage, shaping which content users see first.
As AI videos accumulate online and enter training sets, the preference self-amplifies: even a 20% training share changes ranking behavior, so the bias compounds across model generations.
Retrieval evaluation on mixed real-and-AI corpora needs relevance-matched human judgment or metrics like Normalized $\Delta$; otherwise measured bias is conflated with quality differences.
Contrastive debiasing works as a post-hoc fine-tuning step, and the extracted debiasing vector can be transferred to other videos, offering a practical mitigation without architectural changes.

Reading between the lines

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

Because the paper's human evaluation contradicts its equal-relevance premise, the headline bias magnitudes are likely inflated; the direction of the effect would probably survive a relevance-matched re-test, but that is an editorial reading, not the paper's claim.
The clustering of the debiasing vector $p$ suggests AI video generators imprint a common statistical signature on their output, one that could serve as a provenance detector or watermark outside retrieval tasks.
The same visual-temporal preference likely extends to video recommendation and autoplay ranking, not just text-video retrieval; a logged-interaction study on a mixed catalogue would test that.
If better commercial generators embed richer temporal information, the bias may grow rather than shrink as generation quality improves, since the paper's mechanism treats temporal richness as high-relevance signal.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Quick take: this is a useful empirical paper, likely right about the direction of the effect, but the headline magnitudes rest on an assumption the paper's own data contradict, and the strongest model shows mixed results.

The real contributions are the benchmark and the measurement toolkit. Thirteen thousand generated videos from two open-source generators across four conditions, tested on three retrieval models, plus a PIKA robustness check: that is a solid base. The NormalizedΔ and MixR metrics are a reasonable attempt to separate source bias from semantic mismatch, and the training-loop experiment (Figure 3) is a nice demonstration that AI-generated training data amplifies the effect. The visual/temporal decomposition via frame shuffling and single-frame retrieval is genuinely informative. Public code and data are a plus.

The soft spots are real. First, the equal-relevance premise (Benchmark Requirement 1) is not met. Table 2 shows human raters judge the real video more relevant in 32–47% of pairs, versus 13–18% for the AI video. The headline RelativeΔ and NormalizedΔ are computed under the assumption that each pair shares the query relevance. The authors acknowledge the difficulty in §2.3 but still present 'even when both have the same relevance level' as a finding. The direction of the bias may survive on an equal-relevance subset—if anything, ranking AI videos above semantically superior real videos makes the bias more striking—but the magnitudes, and the 'same relevance' framing, are not established. Second, NormalizedΔ's correction depends on a random-interleaving null model that is asserted rather than validated. Third, InternVideo shows positive RelativeΔ for R@1 on three of four datasets, so the 'clear preference for AI' is not uniform across models; the paper overstates. Fourth, the debiasing method in Table 7 pushes AI videos to R@1=0 with median ranks over 200—an inverted anti-AI bias that is not acknowledged. Fifth, the ImageCond keyframe position was selected on the test data, and the statistical appendix is a one-paragraph assertion of p<0.05 with no effect sizes or test details. Finally, §5.2 calls the clustering of the p-vector a 'direct cause'; the shuffle and t-SNE evidence support association, not causation, and Appendix D itself concedes the mechanisms need more work.

Who should read this: anyone working on AIGC bias in retrieval or building mixed real/AI video collections. The benchmark is a reusable asset and the phenomenon is worth knowing about. But the quantitative claims should be treated as provisional until the equal-relevance issue is addressed with a human-labeled equal-relevance subset.

Recommendation: yes, send to serious peer review. The core measurement is externally grounded and the benchmark is a contribution, but the authors need to fix the framing, add error bars, and either demonstrate the equal-relevance subset or soften the claim.

Referee Report

4 major / 6 minor

Summary. This paper introduces a benchmark of 13,000 AI-generated videos derived from MSR-VTT using CogVideoX and OpenSora, with four test conditions (text-conditioned, image-conditioned, video-extended, plus a 9,000-video training split), and proposes RelativeΔ, NormalizedΔ, and MixR metrics to quantify whether text-video retrieval models rank AI-generated videos above real videos under an 'identical semantics' assumption. Using Alpro, Frozen in Time, and InternVideo, the paper reports predominantly negative metric values, interprets them as Visual-Temporal Induced Source Bias, reports that training with increasing AI-video proportions intensifies the bias, attributes the bias to extra visual and temporal information, and proposes a contrastive-learning debiasing method plus a debiasing vector p.

Significance. If the central claim held, this would be a timely and useful contribution: video retrieval on mixed real-and-AI libraries is an emerging problem, the benchmark is large and publicly released, and the use of three retrieval models, two generators, and a PIKA spot-check provides valuable external grounding. The paper is also honest in its limitation appendices. The directional finding—AI-generated videos tend to be ranked above real counterparts in mixed retrieval—is likely robust. However, the headline 'same relevance level' claim and the quantitative bias magnitudes rest on an unverified equal-relevance premise that the paper's own human evaluation contradicts. At present the contribution is a promising measurement framework plus a directional finding, rather than an established quantitative bias result.

major comments (4)

[§2 Requirement (1) and Table 2; abstract and §3.2] The benchmark's first requirement, 'Identical Semantics,' is load-bearing: the claim that retrieval models rank AI-generated videos higher 'even when both have the same relevance level' depends on every real/AI pair sharing the query's relevance label. Table 2 shows that human evaluators judged the real video more relevant in 32%–47% of sampled pairs and the AI video in 13%–18%, with equal judgments in 40%–53%. Because RelativeΔ, NormalizedΔ, and MixR (Eqs. 1–5) assign equal relevance to every pair, the headline comparison is made under a label assignment that the authors' own data contradict. The paper needs a human-labeled equal-relevance subset (or per-pair relevance stratification) and must report the bias metrics on that subset; without this, the reported magnitudes conflate source bias with relevance misspecification. The direction of the effect may survive, but the quantitative claims and the 'same relevance level' framing do not.
[§2.4, Eqs. (2)–(3)] NormalizedΔ is introduced as a correction for semantic discrepancy, but it relies on the random-interleaving null model in Eqs. (2)–(3), in which mixed rankings are simulated by doubling independent ranks and subtracting a random offset c. No justification is given that this interleaving matches the candidate-pool size and score competition of the actual mixed list, and the distribution of c is unspecified (the notation c ∈ 0,1 is ambiguous between {0,1} and the interval [0,1]). Without validation of this null model, the signed NormalizedΔ values—which flip across models and datasets (e.g., InternVideo R@1 +34.67 on OpenSora TextCond versus Frozen's –64.89)—cannot be interpreted as calibrated bias measurements.
[§3.3 and Figure 3] The training-loop corollary, that the model's preference for AI-generated content strengthens as the AI share grows from 0% to 80%, is supported by a figure plus a detailed 20% example rather than by a full quantitative comparison. The text's summary numbers are also hard to parse: it says NormalizedΔ R@1 'increases by 49.29 points' when fine-tuning on real videos, and then reports a 'decrease of 89.52' when 20% AI is added. The paper should report the complete 0/20/40/60/80 table with significance and variability, and state explicitly whether the trend is monotonic for every metric. As written, the central training-loop claim is not fully established.
[§5.1, Eq. (10); §5.2, Eq. (11)] The debiasing loss in Eq. (10) is constructed from Δr = score(AI) – score(real) and is applied only when Δr ≥ 0, so it directly penalizes the model for ranking AI videos above real videos. Table 7 then reports mixed-AI R@1 = 0 and RelativeΔ = 200, a saturated outcome; this demonstrates that the objective was optimized, not that an independently measured source bias was reduced. Relatedly, the p vector in Eq. (11) is the difference between the original encoder and this same debiased encoder, so the claim that p captures 'additional information embedded by generation encoders' risks circularity: adding p_avg to real representations is equivalent to adding the learned debiasing direction. Independent evidence, such as probing generation encoders directly or controlling for relevance, is needed before p can be interpreted as the cause of the bias.

minor comments (6)

[Figure 1] Figure 1 contains the typo 'addition imformation,' and the caption states the causal mechanism ('extra visual and temporal information embedded by the generation model') more strongly than the experiments independently support.
[§2.4, Eqs. (2)–(3)] Please specify whether c is drawn from {0,1} or from the interval [0,1], and report sensitivity of LocationΔ and NormalizedΔ to the choice of c.
[§2.3, Tables 1 and 2] Table 1 reports CLIP similarity between generated and real videos (0.72–0.87), while Table 2 shows that human raters often prefer the real video's query relevance; the paper should clarify that these measure different constructs and discuss the apparent tension.
[Appendix A] The statistical significance appendix reports 'paired T-tests' but gives no sample size, no correction for multiple comparisons, and no effect sizes; please add these details.
[§5.1] The sentence 'when Δr < 0, we do not apply this loss function, ensuring that during contrastive learning, the model still favors generated videos' appears to contradict the stated goal of reducing preference for AI-generated videos and should be rephrased.
[Appendix B] The PIKA check uses only 100 videos and a single retrieval model; it is useful as a spot check, but the statement that 'Visual-Temporal Induced Source Bias might be further amplified' in commercial models goes beyond what this check can support.

Circularity Check

2 steps flagged · score 5.0 of 10

Debiasing result is the training objective restated as a rank metric; the headline 'same relevance level' is assumed by the benchmark's identical-semantics labeling rule rather than established.

fitted input called prediction [Section 5.1, Eqs. (9)-(10), evaluated in Table 7]
"The overall training objective is: 𝜃VC = arg min 𝜃vc L(𝑟VC,𝑦,𝜃 VC)+ Δ𝑟."

Δr is defined in Eq. (9) as the model score for the AI-generated video minus the score for the paired real video. Eq. (10) minimizes L + Δr, i.e., it directly trains the model to give real videos higher scores than AI videos. Table 7 then reports that after this training mixed-AI R@1 collapses to 0 and NormalizedΔ becomes strongly positive (~+184 to +200). This 'debiasing' result is the optimization objective itself expressed as a rank-based metric, not an independent test or prediction. The mitigation conclusion is therefore forced by the chosen training loss, although transfer to the ImageCond and CogVideoX test sets provides some generalization evidence.
self definitional [Section 2, Benchmark Requirement (1) and Section 2.4, Eq. (1), with Table 2 human evaluation]
"(1) Identical Semantics: Ensuring generated videos have the same semantics as original ones, so they share the same relevant labels as the query, which helps prevent abnormal retrieval rankings due to excessive video-query similarity."

The benchmark labels every AI-generated video as exactly as relevant as its paired real video, so in Eq. (1) any higher rank of the AI video is counted as source bias. That is the same claim the paper says it discovers: models rank AI-generated videos higher than real videos 'even when both have the same relevance level' (abstract and Section 3.2). The paper's own Table 2, however, reports human evaluators choosing the real video as more relevant in 32–47% of pairs and the AI video in only 13–18%, with 40–53% equal. The equal-relevance condition is thus assumed by the labeling rule, not established; the reported bias magnitudes include the effect of unequal relevance. This makes the 'same relevance level' framing definitional rather than empirical.

full rationale

The core empirical finding is not merely a rename or a self-citation artifact: the benchmark uses real MSR-VTT videos, 13,000 videos from two third-party generators, and three off-the-shelf retrieval models, and the reported preference for AI-generated videos is measured from actual ranking distributions. So the paper has substantial independent content. However, two steps in the derivation chain are partially circular. First, the debiasing section defines a loss Δr that directly rewards ranking the real video above its AI counterpart, then reports that the rank-based bias metric collapsed after training; that is the optimization target restated, not independent confirmation. Second, the benchmark's 'Identical Semantics' requirement assigns equal relevance labels to real/AI pairs by construction, and the paper's own human evaluation in Table 2 shows that equality is not achieved; the headline 'same relevance level' therefore rests on the benchmark's labeling convention rather than measured equivalence. This makes the magnitudes of RelativeΔ and NormalizedΔ depend on an assumption the paper itself contradicts, although the direction of the effect would probably survive on a true equal-relevance subset. No load-bearing self-citation or imported uniqueness theorem was found; citations to prior source-bias work are contextual rather than definitional.

Assumptions & free parameters 3 free parameters · 6 assumptions · 3 invented entities

The benchmark derives its labels from the assumption that caption-paired real and AI videos are relevance-equivalent, which the paper's own human evaluation (Table 2) partially refutes. The ImageCond keyframe position was selected post hoc, the LocationDelta metric uses a randomly chosen interleaving offset, and the debiasing loss applies only above a hand-set threshold. The visual-versus-temporal decomposition assumes frame shuffling and single-frame retrieval are clean interventions, and the p-vector explanation is constructed from the debiased model whose bias it explains.

free parameters (3)

ImageCond keyframe position = 20% of video duration
Selected post hoc as the position yielding the best generation similarity (Section 2.2); applies to the test set, so it is a test-time parameter choice.
LocationDelta interleaving offset c = random draw from {0, 1}
Eq. (2)-(3) simulate the mixed retrieval list with a randomly chosen offset; LocationDelta then subtracts into NormalizedDelta, making the headline metric stochastic at the level of list simulation.
Debiasing loss activation threshold = 0, applied only when Delta-r > 0
Eq. (10) applies the contrastive penalty only when the AI video scores above the real video; this hand-set asymmetry defines the post-fine-tuning ranking behavior.

assumptions (6)

domain assumption MSR-VTT captions and relevance annotations are treated as ground truth for both real and generated videos
Used in Sections 2.1-2.2 to pair each generated video with its real counterpart's query labels.
domain assumption Caption-paired AI-generated and real videos are relevance-equivalent ('Identical Semantics')
Requirement (1) in Section 2 is load-bearing for RelativeDelta and NormalizedDelta; Table 2's own human evaluation shows real videos judged more relevant in 32% to 47% of pairs, so equal relevance holds only partially.
domain assumption CLIP ViT-B/32 cosine similarity measures real-vs-generated semantic similarity
Used in Section 2.3 to validate dataset quality; the CLIP family also underlies the retrieval model features, which may make the quality check favorable.
domain assumption GPT-4 caption fusion produces comprehensive prompts
Used in Section 2.2; caption errors propagate into the generated videos and hence into the relevance labeling.
domain assumption Frame shuffling and single-frame retrieval isolate temporal from visual contributions
Sections 4.1-4.2 interpret these interventions causally; shuffling also creates out-of-distribution inputs for models expecting temporal structure, affecting real and AI videos asymmetrically.
standard math Paired t-tests as reported are valid
Appendix A states p < 0.05 for all comparisons without test statistics, sample sizes, degrees of freedom, or how the paired groups were constructed.

invented entities (3)

Visual-Temporal Induced Source Bias
purpose: Named phenomenon for retrieval models ranking AI-generated videos above real ones; the paper's central explanatory construct.
Inferred from ranking measurements on this benchmark; the frame-shuffle and single-frame experiments are in-paper evidence only, with no outside falsifiable handle yet.
Extra visual-temporal information embedded by generation encoders
purpose: Postulated latent cause of the bias, claimed to be consistent across videos and transferable.
Supported by p-vector clustering (Figure 5) and the transfer experiment (Table 8), all within the same models used to define the bias.
Debiasing vector p (p_avg)
purpose: Difference between debiased and original video embeddings, used to transfer and visualize the bias.
Constructed from the debiased model whose behavior defines the bias; Table 8 provides an in-paper intervention check, but there is no external validation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Generative Ghost: Investigating Ranking Bias Hidden in AI-Generated Videos." pith.science (2026). https://pith.science/paper/A2XFNXTH

@misc{pith2026250207327,
  author       = {Pith},
  title        = {Pith review of: Generative Ghost: Investigating Ranking Bias Hidden in AI-Generated Videos},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/A2XFNXTH}},
  note         = {Machine review of arXiv:2502.07327}
}

read the original abstract

With the rapid development of AI-generated content (AIGC), the creation of high-quality AI-generated videos has become faster and easier, resulting in the Internet being flooded with all kinds of video content. However, the impact of these videos on the content ecosystem remains largely unexplored. Video information retrieval remains a fundamental approach for accessing video content. Building on the observation that retrieval models often favor AI-generated content in ad-hoc and image retrieval tasks, we investigate whether similar biases emerge in the context of challenging video retrieval, where temporal and visual factors may further influence model behavior. To explore this, we first construct a comprehensive benchmark dataset containing both real and AI-generated videos, along with a set of fair and rigorous metrics to assess bias. This benchmark consists of 13,000 videos generated by two state-of-the-art open-source video generation models. We meticulously design a suite of rigorous metrics to accurately measure this preference, accounting for potential biases arising from the limited frame rate and suboptimal quality of AIGC videos. We then applied three off-the-shelf video retrieval models to perform retrieval tasks on this hybrid dataset. Our findings reveal a clear preference for AI-generated videos in retrieval. Further investigation shows that incorporating AI-generated videos into the training set of retrieval models exacerbates this bias. Unlike the preference observed in image modalities, we find that video retrieval bias arises from both unseen visual and temporal information, making the root causes of video bias a complex interplay of these two factors. To mitigate this bias, we fine-tune the retrieval models using a contrastive learning approach. The results of this study highlight the potential implications of AI-generated videos on retrieval systems.

Figures

Figures reproduced from arXiv: 2502.07327 by the authors.

**Figure 2.** Using the CLIP model to compute the similarity between AI-generated video datasets and real video datasets (X-axis), [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗

**Figure 3.** Evaluation results on a training set containing a mix of AI-generated videos. We vary the proportion of AI-generated [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗

Figures from the paper (2 more)

**Figure 4.** Figure 4: Changes in retrieval metrics of the model ( [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]

**Figure 5.** Figure 5: t-SNE visualization of image representations and [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 26 canonical work pages

[1]

Azzah Al-Maskari and Mark Sanderson. 2010. A review of factors influencing user satisfaction in information retrieval. Journal of the American Society for Information Science and Technology 61, 5 (2010), 859–868

work page 2010
[2]

Lisa Anne Hendricks, Oliver Wang, Eli Shechtman, Josef Sivic, Trevor Darrell, and Bryan Russell. 2017. Localizing moments in video with natural language. In Proceedings of the IEEE international conference on computer vision . 5803–5812

2017
[3]

Max Bain, Arsha Nagrani, Gül Varol, and Andrew Zisserman. 2021. Frozen in time: A joint video and image encoder for end-to-end retrieval. In Proceedings of the IEEE/CVF international conference on computer vision . 1728–1738

work page 2021
[4]

Fabian Caba Heilbron, Victor Escorcia, Bernard Ghanem, and Juan Carlos Niebles
[5]

David Chen and William B Dolan. 2011. Collecting highly parallel data for paraphrase evaluation. In Proceedings of the 49th annual meeting of the association for computational linguistics: human language technologies . 190–200

2011
[6]

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Interna- tional conference on machine learning . PMLR, 1597–1607

2020
[7]

Sunhao Dai, Weihao Liu, Yuqi Zhou, Liang Pang, Rongju Ruan, Gang Wang, Zhenhua Dong, Jun Xu, and Ji-Rong Wen. 2024. Cocktail: A Comprehensive Information Retrieval Benchmark with LLM-Generated Documents Integration. Findings of the Association for Computational Linguistics: ACL 2024 (2024)

work page 2024
[8]

Sunhao Dai, Chen Xu, Shicheng Xu, Liang Pang, Zhenhua Dong, and Jun Xu

Show all 40 references

[9]

Sunhao Dai, Yuqi Zhou, Liang Pang, Weihao Liu, Xiaolin Hu, Yong Liu, Xiao Zhang, Gang Wang, and Jun Xu. 2024. Neural Retrievers are Biased Towards LLM-Generated Content. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (2024)

2024
[10]

Sunhao Dai, Yuqi Zhou, Liang Pang, Weihao Liu, Xiaolin Hu, Yong Liu, Xiao Zhang, and Jun Xu. 2023. Llms may dominate information access: Neural re- trievers are biased towards llm-generated texts. arXiv preprint arXiv:2310.20501 (2023)

2023 arXiv
[11]

Lin Geng Foo, Hossein Rahmani, and Jun Liu. 2023. Ai-generated content (aigc) for various data modalities: A survey. arXiv preprint arXiv:2308.14177 2 (2023), 2

2023 arXiv
[12]

Ruoyuan Gao. 2021. Toward a fairer information retrieval system . Ph. D. Disserta- tion. Rutgers The State University of New Jersey, School of Graduate Studies

2021
[13]

Ruoyuan Gao and Chirag Shah. 2021. Addressing bias and fairness in search systems. In Proceedings of the 44th international ACM SIGIR conference on research and development in information retrieval . 2643–2646

2021
[14]

Yifei Gao, Jiaqi Wang, Zhiyu Lin, and Jitao Sang. 2024. AIGCs Confuse AI Too: Investigating and Explaining Synthetic Image-induced Hallucinations in Large Vision-Language Models. arXiv preprint arXiv:2403.08542 (2024)

2024 arXiv
[15]

Sahin Cem Geyik, Stuart Ambler, and Krishnaram Kenthapadi. 2019. Fairness- aware ranking in search & recommendation systems with application to linkedin talent search. In Proceedings of the 25th acm sigkdd international conference on knowledge discovery & data mining . 2221–2231

2019
[16]

Gerard Goggin. 2010. Global mobile media. Routledge

2010
[17]

Dongxu Li, Junnan Li, Hongdong Li, Juan Carlos Niebles, and Steven CH Hoi
[18]

Abbe Mowshowitz and Akira Kawaguchi. 2002. Assessing bias in search engines. Information Processing & Management 38, 1 (2002), 141–156

2002
[19]

Abbe Mowshowitz and Akira Kawaguchi. 2005. Measuring search engine bias. Information processing & management 41, 5 (2005), 1193–1205

2005
[20]

Open-Sora: Democratizing Efficient Video Production for All

Open-Sora 2024. Open-Sora: Democratizing Efficient Video Production for All . https://github.com/hpcaitech/Open-Sora

2024
[21]

Evaggelia Pitoura, Panayiotis Tsaparas, Giorgos Flouris, Irini Fundulaki, Panagi- otis Papadakos, Serge Abiteboul, and Gerhard Weikum. 2018. On measuring bias in online information. ACM SIGMOD Record 46, 4 (2018), 16–21

2018
[22]

Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, et al. 2021. Learning transferable visual models from natural language supervision. In International conference on machine learni...

2021
[23]

Anna Rohrbach, Marcus Rohrbach, Niket Tandon, and Bernt Schiele. 2015. A dataset for movie description. In Proceedings of the IEEE conference on computer vision and pattern recognition . 3202–3212

2015
[24]

Christian Tarchi, Sonia Zaccoletti, and Lucia Mason. 2021. Learning from text, video, or subtitles: A comparative analysis. Computers & Education 160 (2021), 104034

2021
[25]

Laurens Van der Maaten and Geoffrey Hinton. 2008. Visualizing data using t-SNE. Journal of machine learning research 9, 11 (2008)

2008
[26]

Haoyu Wang, Sunhao Dai, Haiyuan Zhao, Liang Pang, Xiao Zhang, Gang Wang, Zhenhua Dong, Jun Xu, and Ji-Rong Wen. 2025. Perplexity-Trap: PLM-Based Retrievers Overrate Low Perplexity Documents. The Thirteenth International Conference on Learning Representations (2025). MM ’25, Oc...

2025
[27]

Yi Wang, Kunchang Li, Yizhuo Li, Yinan He, Bingkun Huang, Zhiyu Zhao, Hongjie Zhang, Jilan Xu, Yi Liu, Zun Wang, et al . 2022. Internvideo: General video foundation models via generative and discriminative learning. arXiv preprint arXiv:2212.03191 (2022)

2022 arXiv
[28]

Colin Wilkie and Leif Azzopardi. 2014. A retrievability analysis: Exploring the relationship between retrieval bias and retrieval performance. InProceedings of the 23rd ACM International Conference on Conference on Information and Knowledge Management. 81–90

2014
[29]

Jiayang Wu, Wensheng Gan, Zefeng Chen, Shicheng Wan, and Hong Lin. 2023. Ai-generated content (aigc): A survey. arXiv preprint arXiv:2304.06632 (2023)

2023 arXiv
[30]

Zhen Xing, Qijun Feng, Haoran Chen, Qi Dai, Han Hu, Hang Xu, Zuxuan Wu, and Yu-Gang Jiang. 2024. A survey on video diffusion models. Comput. Surveys 57, 2 (2024), 1–42

2024
[31]

Jun Xu, Tao Mei, Ting Yao, and Yong Rui. 2016. Msr-vtt: A large video description dataset for bridging video and language. In Proceedings of the IEEE conference on computer vision and pattern recognition . 5288–5296

2016
[32]

Shicheng Xu, Danyang Hou, Liang Pang, Jingcheng Deng, Jun Xu, Huawei Shen, and Xueqi Cheng. 2024. Ai-generated images introduce invisible relevance bias to text-image retrieval. Proceedings of the 47th international ACM SIGIR conference on research and development in informati...

2024
[33]

Shicheng Xu, Danyang Hou, Liang Pang, Jingcheng Deng, Jun Xu, Huawei Shen, and Xueqi Cheng. 2024. Invisible relevance bias: Text-image retrieval models prefer ai-generated images. In Proceedings of the 47th international ACM SIGIR conference on research and development in info...

2024
[34]

Ke Yang and Julia Stoyanovich. 2017. Measuring fairness in ranked outputs. In Proceedings of the 29th international conference on scientific and statistical database management. 1–6

2017
[35]

Zhuoyi Yang, Jiayan Teng, Wendi Zheng, Ming Ding, Shiyu Huang, Jiazheng Xu, Yuanming Yang, Wenyi Hong, Xiaohan Zhang, Guanyu Feng, et al . 2024. Cogvideox: Text-to-video diffusion models with an expert transformer. arXiv preprint arXiv:2408.06072 (2024)

2024 arXiv
[36]

Sirui Yao and Bert Huang. 2017. New fairness metrics for recommendation that embrace differences. arXiv preprint arXiv:1706.09838 (2017)

2017 arXiv
[37]

Visual-Temporal Induced Source Bias

Yuqi Zhou, Sunhao Dai, Liang Pang, Gang Wang, Zhenhua Dong, Jun Xu, and Ji-Rong Wen. 2024. Source Echo Chamber: Exploring the Escalation of Source Bias in User, Data, and Recommender System Feedback Loop. arXiv preprint arXiv:2405.17998 (2024). Generative Ghost: Investigating ...

2024 arXiv
[2015]

In Proceedings of the ieee conference on computer vision and pattern recognition

Activitynet: A large-scale video benchmark for human activity understand- ing. In Proceedings of the ieee conference on computer vision and pattern recognition. 961–970
[2022]

In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

Align and prompt: Video-and-language pre-training with entity prompts. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 4953–4963
[2024]

In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining

Bias and Unfairness in Information Retrieval Systems: New Challenges in the LLM Era. In Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining . 6437–6447

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.

[1] [1]

Azzah Al-Maskari and Mark Sanderson. 2010. A review of factors influencing user satisfaction in information retrieval. Journal of the American Society for Information Science and Technology 61, 5 (2010), 859–868

work page 2010

[2] [2]

Lisa Anne Hendricks, Oliver Wang, Eli Shechtman, Josef Sivic, Trevor Darrell, and Bryan Russell. 2017. Localizing moments in video with natural language. In Proceedings of the IEEE international conference on computer vision . 5803–5812

2017

[3] [3]

Max Bain, Arsha Nagrani, Gül Varol, and Andrew Zisserman. 2021. Frozen in time: A joint video and image encoder for end-to-end retrieval. In Proceedings of the IEEE/CVF international conference on computer vision . 1728–1738

work page 2021

[4] [4]

Fabian Caba Heilbron, Victor Escorcia, Bernard Ghanem, and Juan Carlos Niebles

[5] [5]

David Chen and William B Dolan. 2011. Collecting highly parallel data for paraphrase evaluation. In Proceedings of the 49th annual meeting of the association for computational linguistics: human language technologies . 190–200

2011

[6] [6]

Ting Chen, Simon Kornblith, Mohammad Norouzi, and Geoffrey Hinton. 2020. A simple framework for contrastive learning of visual representations. In Interna- tional conference on machine learning . PMLR, 1597–1607

2020

[7] [7]

Sunhao Dai, Weihao Liu, Yuqi Zhou, Liang Pang, Rongju Ruan, Gang Wang, Zhenhua Dong, Jun Xu, and Ji-Rong Wen. 2024. Cocktail: A Comprehensive Information Retrieval Benchmark with LLM-Generated Documents Integration. Findings of the Association for Computational Linguistics: ACL 2024 (2024)

work page 2024

[8] [8]

Sunhao Dai, Chen Xu, Shicheng Xu, Liang Pang, Zhenhua Dong, and Jun Xu