Datasets:

AbeerMostafa
/

Novelty-Reviewer-Dataset

paper_id stringlengths 9 12	title stringlengths 10 183	venue stringclasses 2 values	year int16 2.02k 2.03k	split stringclasses 1 value	in_eval_subset bool 1 class	abstract stringlengths 3 3.6k	num_reviews int8 1 7	reviews listlengths 1 7	review_template stringclasses 7 values	novelty_excerpts listlengths 0 77	novelty_extraction_raw stringlengths 24 3.64k ⌀	novelty_score int8 -1 2 ⌀	novelty_score_justification stringlengths 143 1.93k ⌀	novelty_detailed_assessment stringlengths 99 1.59k ⌀	novelty_summary stringlengths 860 2.24k
-0Cjhnl-dhK	Towards Uncertainties in Deep Learning that Are Accurate and Calibrated	ICLR	2,022	train	false	Predictive uncertainties can be characterized by two properties---calibration and sharpness. This paper introduces algorithms that ensure the calibration of any model while maintaining sharpness. They apply in both classification and regression and guarantee the strong property of distribution calibration, while being ...	1	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThe authors propose to decompose different common loss functions into calibration and refinement terms and show theoretically that under some (strong) assumptions, such as universal approximation, models can become perfectly calibrated without sacrificing p...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "Unclear / small novelty compared to previous work", "Performance in higher dimensions: The current method seems to only work well in very low-dimensional settings (one to two dimensions or so).", "Empirical results: The current results are missing error bars, which makes it impossible to assess their statistic...	{ "novelty_excerpts": [ "Unclear / small novelty compared to previous work", "Performance in higher dimensions: The current method seems to only work well in very low-dimensional settings (one to two dimensions or so).", "Empirical results: The current results are missing error bars, which makes it imposs...	-1	The paper introduces a recalibration algorithm that builds upon existing methods with minimal innovation. The work presents a constrained optimization framework that, while theoretically sound, lacks significant novelty in its approach. The contribution lies in its application to specific tasks, but the method's limita...	The paper proposes a recalibration algorithm that decomposes common loss functions into calibration and refinement terms, but this idea is not novel. The method works well in low-dimensional settings, but its performance in higher dimensions remains unclear. The work extends existing methods by introducing a constraine...	Novelty Score: -1 Score Justification: The paper introduces a recalibration algorithm that builds upon existing methods with minimal innovation. The work presents a constrained optimization framework that, while theoretically sound, lacks significant novelty in its approach. The contribution lies in its application to...
-29uFS4FiDZ	Word Sense Induction with Knowledge Distillation from BERT	ICLR	2,022	train	false	Pre-trained contextual language models are ubiquitously employed for language understanding tasks, but are unsuitable for resource-constrained systems. Noncontextual word embeddings are an efficient alternative in these settings. Such methods typically use one vector to encode multiple different meanings of a word, an...	3	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThis paper proposes a two-stage method to distill word sense knowledge from BERT into a skip-gram-like framework. The authors show that the proposed method performs well on sense selection and induction (or representation). They also verify the efﬁcacy on t...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "One of the main weaknesses is that the results are not significant enough.", "In Table 3, the performance of the proposed model on both criteria is worse than the existing models.", "And the paper didn't compare with all existing sense models (see missing citations).", "In Figure 2, most of the improvements ...	{ "novelty_excerpts": [ "One of the main weaknesses is that the results are not significant enough.", "In Table 3, the performance of the proposed model on both criteria is worse than the existing models.", "And the paper didn't compare with all existing sense models (see missing citations).", "In Fig...	-1	The paper introduces a straightforward approach to word sense induction using BERT, though the methodological novelty is low due to its similarity to existing approaches. The contributions are largely incremental, building upon existing techniques without substantial conceptual or technical innovation.	The paper presents a two-stage method for distilling word sense knowledge from BERT, which demonstrates state-of-the-art results. However, the approach is not novel, as it relies on existing techniques such as attention mechanisms and knowledge distillation. The work extends existing methods for word sense induction, b...	Novelty Score: -1 Score Justification: The paper introduces a straightforward approach to word sense induction using BERT, though the methodological novelty is low due to its similarity to existing approaches. The contributions are largely incremental, building upon existing techniques without substantial conceptual o...
-3yxxvDis3L	How to Improve Sample Complexity of SGD over Highly Dependent Data?	ICLR	2,022	train	false	Conventional machine learning applications typically assume that data samples are independently and identically distributed (i.i.d.). However, many practical scenarios naturally involve a data-generating process that produces highly dependent data samples, which are known to heavily bias the stochastic optimization pro...	3	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThis work analyzes and provides high probability convergence guarantee for periodically subsampled and minibatch SGD for optimization of convex and Lipschitz continuous function when the data stream is $\\phi$-mixing.\n\n## Main Review\nStrength:\n\nAnalyze...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "There is no novelty here.", "almost no theoretical novelty", "The contributions are neither significant nor novel.", "The contributions are only marginally significant or novel." ]	{ "novelty_excerpts": [ "There is no novelty here.", "almost no theoretical novelty", "The contributions are neither significant nor novel.", "The contributions are only marginally significant or novel." ] }	0	The paper presents a competent analysis of SGD over dependent data, but its contributions remain largely incremental. The work builds on existing results and frameworks, providing some interesting insights but not fundamentally new ideas. The method extends existing approaches to subsampled and minibatch SGD, though th...	The paper introduces a framework for analyzing SGD over dependent data, leveraging the phi-mixing process to model data dependence. However, the analysis largely relies on existing results and frameworks, such as those presented in (Agarwal & Duchi, 2012). The work provides some interesting insights into the performanc...	Novelty Score: 0 Score Justification: The paper presents a competent analysis of SGD over dependent data, but its contributions remain largely incremental. The work builds on existing results and frameworks, providing some interesting insights but not fundamentally new ideas. The method extends existing approaches to ...
-6me0AsJVdu	Model Validation Using Mutated Training Labels: An Exploratory Study	ICLR	2,022	train	false	For out-of-sample validation, the sample set may be too small to be representative of the data distribution; the accuracy can have a large variance across different runs; excessive reuse of a fixed set of samples can lead to overfitting even if the samples are held out and not used in the training process. This paper i...	2	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThis paper introduces new approach for validation of ML models which is based on top of ideas imported from software development & QA (so-called mutation and metamorphic techniques). The core claimed advantage consist in the ability to avoid cross-validatio...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "A novel approach in the area of validation techniques, which are very conservative", "The contributions are significant, and do not exist in prior works.", "The contributions are significant, and do not exist in prior works.", "This work represents a first empirical study in this new direction, and a lot of ...	{ "novelty_excerpts": [ "A novel approach in the area of validation techniques, which are very conservative", "The contributions are significant, and do not exist in prior works.", "The contributions are significant, and do not exist in prior works.", "This work represents a first empirical study in t...	1	The paper introduces a novel approach to model validation, building on ideas from software development and QA. This approach presents a new direction in validation techniques, which are conservative in nature. However, the method's effectiveness and applicability are limited by its dependence on specific evaluation met...	The paper presents a moderately novel approach to model validation, combining mutation and metamorphic techniques from software development and QA. The work extends existing validation methods by introducing a new metric for evaluating model performance, which is based on the difference in training accuracy between mod...	Novelty Score: 1 Score Justification: The paper introduces a novel approach to model validation, building on ideas from software development and QA. This approach presents a new direction in validation techniques, which are conservative in nature. However, the method's effectiveness and applicability are limited by it...
-70L8lpp9DF	Hyperparameter Tuning with Renyi Differential Privacy	ICLR	2,022	train	false	For many differentially private algorithms, such as the prominent noisy stochastic gradient descent (DP-SGD), the analysis needed to bound the privacy leakage of a single training run is well understood. However, few studies have reasoned about the privacy leakage resulting from the multiple training runs needed to fin...	2	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThe authors make the following contributions regarding differentially private hyperparameter tuning :\n* As an example, the authors train an SVM with a weight penalty and show that, in presence of an outlier, a membership inference attack can be employed to...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "The authors make the following contributions regarding differentially private hyperparameter tuning :", "The authors propose a way to measure the utility of the algorithms by looking at the expected quantile of the output.", "The results of their utility analysis, coupled with an experiment on MNIST, show that...	{ "novelty_excerpts": [ "The authors make the following contributions regarding differentially private hyperparameter tuning :", "The authors propose a way to measure the utility of the algorithms by looking at the expected quantile of the output.", "The results of their utility analysis, coupled with an experiment on ...	1	The paper presents moderately novel contributions to the field of differentially private hyperparameter tuning. It builds upon existing ideas and provides new insights into the problem formulation and methodological innovations. The work extends and improves upon prior approaches, offering a more comprehensive understa...	The paper introduces a novel approach to private hyperparameter tuning, which combines existing methods and provides a more general framework for analyzing the trade-offs between privacy and utility. The work extends the results of Liu and Talwar (2019) and provides a more comprehensive understanding of the problem for...	Novelty Score: 1 Score Justification: The paper presents moderately novel contributions to the field of differentially private hyperparameter tuning. It builds upon existing ideas and provides new insights into the problem formulation and methodological innovations. The work extends and improves upon prior approaches,...
-7UeX2KPqs	State-Action Joint Regularized Implicit Policy for Offline Reinforcement Learning	ICLR	2,022	train	false	Offline reinforcement learning enables learning from a fixed dataset, without further interactions with the environment. The lack of environmental interactions makes the policy training vulnerable to state-action pairs far from the training dataset and prone to missing rewarding actions. For training more effective age...	3	[ { "review_text": "# Review\n\n## Summary Of The Paper\nOne of the existing Offline RL algorithms is to constrain the learned policy, such as constraining the learned policy to be consistent with the behavior policy itself or the action distribution based on state conditions, or adopting a Gaussian policy. Howev...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "The technical novelty is sufficient and significant.", "The contributions are significant and somewhat new.", "Aspects of the contributions exist in prior work.", "I believe the most valuable experiments are the toy experiments to illustrate their motivations as the findings in the toy experiments may help t...	{ "novelty_excerpts": [ "The technical novelty is sufficient and significant.", "The contributions are significant and somewhat new.", "Aspects of the contributions exist in prior work.", "I believe the most valuable experiments are the toy experiments to illustrate their motivations as the findings i...	1	The paper introduces a novel approach to offline reinforcement learning by leveraging implicit policies and state-action visitation matching. This method extends existing work by addressing the limitations of Gaussian policies and behavior policies. However, the contributions remain largely incremental, building upon e...	The paper presents a framework that supports learning a flexible while well-regularized policy in offline RL. The method uses an implicit policy to capture the multi-modality of the action value function and controls the state-action access frequency in the data set to be consistent with the learned policy. This approa...	Novelty Score: 1 Score Justification: The paper introduces a novel approach to offline reinforcement learning by leveraging implicit policies and state-action visitation matching. This method extends existing work by addressing the limitations of Gaussian policies and behavior policies. However, the contributions rema...
-8sBpe7rDiV	NETWORK INSENSITIVITY TO PARAMETER NOISE VIA PARAMETER ATTACK DURING TRAINING	ICLR	2,022	train	false	Neuromorphic neural network processors, in the form of compute-in-memory crossbar arrays of memristors, or in the form of subthreshold analog and mixed-signal ASICs, promise enormous advantages in compute density and energy efficiency for NN-based ML tasks. However, these technologies are prone to computational non-ide...	1	[ { "review_text": "# Review\n\n## Summary Of The Paper\nAuthors investigate the problem of network robustness in the presence of parameter variations. To that end, authors propose an adversarial parameter perturbation based robust training method. Proposed training approach iteratively performs adversarial attac...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "My main concern relates to the ambiguity on explicitly putting the novelty of this paper as the proposed problem is recently addressed by similar studies.", "In fact, adversarial parameter perturbations for regularization is an approach that was demonstrated in the context of model generalization with similar re...	{ "novelty_excerpts": [ "My main concern relates to the ambiguity on explicitly putting the novelty of this paper as the proposed problem is recently addressed by similar studies.", "In fact, adversarial parameter perturbations for regularization is an approach that was demonstrated in the context of model ge...	1	The paper introduces a moderately novel approach to model regularization through adversarial parameter perturbations, successfully applied to recurrent SNNs. The work presents a competent extension of existing ideas in adversarial training, but its novelty is somewhat limited by the fact that similar methods have been ...	The paper introduces a method for adversarial parameter perturbation based model regularization, which is successfully applied to recurrent SNNs. This approach builds on existing ideas in adversarial training, but its novelty is somewhat limited by the fact that similar methods have been proposed recently. The work ext...	Novelty Score: 1 Score Justification: The paper introduces a moderately novel approach to model regularization through adversarial parameter perturbations, successfully applied to recurrent SNNs. The work presents a competent extension of existing ideas in adversarial training, but its novelty is somewhat limited by t...
-9uy3c7b_ks	Learning Controllable Elements Oriented Representations for Reinforcement Learning	ICLR	2,022	train	false	Deep Reinforcement Learning (deep RL) has been successfully applied to solve various decision-making problems in recent years. However, the observations in many real-world tasks are often high dimensional and include much task-irrelevant information, limiting the applications of RL algorithms. To tackle this problem, w...	3	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThere has been a school thought lately in RL that good representations for control should focus on abstracting out on only the task-relevant features. In high-dimensional tasks, this is especially useful in making optimal use of a limited neural network cap...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "the idea of using controllability to inform compression of state representations -- after all, what is important to the agent is what it can control about the environment.", "Perhaps the most important situation is where there are non-controllable task-informative features.", "The authors do point out one exam...	{ "novelty_excerpts": [ "the idea of using controllability to inform compression of state representations -- after all, what is important to the agent is what it can control about the environment.", "Perhaps the most important situation is where there are non-controllable task-informative features.", "The...	0	The paper presents a representation learning framework that builds on existing ideas in mutual information and controllability, but the contributions remain largely incremental. The approach combines and extends previous work, introducing a new objective function and experimental results that demonstrate its effectiven...	The paper introduces a novel representation learning framework, LCER, which enables the agent to extract task-relevant representations from image-based observations. The work presents a combination of mutual information and controllability to guide representation learning, extending previous approaches like CURL and PI...	Novelty Score: 0 Score Justification: The paper presents a representation learning framework that builds on existing ideas in mutual information and controllability, but the contributions remain largely incremental. The approach combines and extends previous work, introducing a new objective function and experimental ...
-BTmxCddppP	Revisiting Out-of-Distribution Detection: A Simple Baseline is Surprisingly Effective	ICLR	2,022	train	false	It is an important problem in trustworthy machine learning to recognize out-of-distribution (OOD) inputs which are inputs unrelated to the in-distribution task. Many out-of-distribution detection methods have been suggested in recent years. The goal of this paper is to recognize common objectives as well as to identify...	2	[ { "review_text": "# Review\n\n## Summary Of The Paper\nThe paper analyzes different OOD detection methods and show that even if the formulation for many OOD methods were different, the binary discrimination is equivalent to those different types of methods when the rankings induced by Bayes optimal classifier a...	## Summary Of The Paper Provide a brief summary of the paper and its contributions. ## Main Review Please list both the strengths and weaknesses of the paper. When discussing weaknesses, please provide concrete, actionable feedback on the paper. ## Summary Of The Review Please provide a short summary justifying your ...	[ "The paper provides insights and findings of properties of different OOD detection approaches.", "Most of the previous work were mainly empirical and lacked mathematical foundation", "The finding of uncertainty based methods being equivalent under the hood with binary discriminators is interesting.", "It expl...	{ "novelty_excerpts": [ "The paper provides insights and findings of properties of different OOD detection approaches.", "Most of the previous work were mainly empirical and lacked mathematical foundation", "The finding of uncertainty based methods being equivalent under the hood with binary discriminator...	1	The paper introduces a novel connection between uncertainty-based methods and binary discriminators, providing insights into the properties of different OOD detection approaches. However, the contributions are somewhat limited by their focus on a specific set of methods and datasets, and the work builds upon existing i...	The paper presents a comprehensive analysis of existing OOD detection methods, highlighting their similarities and differences through mathematical derivations and empirical comparisons. The work extends the understanding of OOD detection by showing that uncertainty-based methods are equivalent to binary discriminators...	Novelty Score: 1 Score Justification: The paper introduces a novel connection between uncertainty-based methods and binary discriminators, providing insights into the properties of different OOD detection approaches. However, the contributions are somewhat limited by their focus on a specific set of methods and datase...
-Nf6TikpjQ	Multi-agent Performative Prediction: From Global Stability and Optimality to Chaos	ICLR	2,022	train	false	"The recent framework of performative prediction is aimed at capturing settings where predictions in(...TRUNCATED)	1	[{"review_text":"# Review\n\n## Summary Of The Paper\nThis paper studied the different behaviours of(...TRUNCATED)	"## Summary Of The Paper\nProvide a brief summary of the paper and its contributions.\n\n## Main Rev(...TRUNCATED)	["As claimed, this is the first time 'where such formal chaotic results are established in settings (...TRUNCATED)	"{\n \"novelty_excerpts\": [\n \"As claimed, this is the first time 'where such formal chaotic r(...TRUNCATED)	1	"The paper presents moderately novel contributions by establishing formal chaotic results in setting(...TRUNCATED)	"The paper introduces a novel application of exponentiated gradient descent in performative predicti(...TRUNCATED)	"Novelty Score: 1\n\nScore Justification:\nThe paper presents moderately novel contributions by esta(...TRUNCATED)

End of preview. Expand in Data Studio

Novelty Reviewer Dataset

A large-scale dataset of expert novelty judgments from the peer reviews of ICLR and NeurIPS submissions.

The dataset links 37,899 submissions to the novelty assessments in their 102,021 official reviews. For every submission it provides the reviews with their numerical scores, the passages in which reviewers assess novelty, and a paper-level novelty label with a written justification. It supports research on automated novelty assessment, review generation, related-work retrieval, and the analysis of peer review itself.


Paper	Are We Truly Innovating? A Qualitative and Quantitative Study of Originality in AI Research Papers
Code	github.com/AbeerMostafa/Novelty-Reviewer
Model	AbeerMostafa/Novelty_Reviewer
License	CC BY-NC 4.0
Language	English

Dataset at a glance


Submissions	37,899
Official reviews	102,021 (1–7 per submission, mean 2.7)
Venues	ICLR 2022–2025, NeurIPS 2022–2025
Labels	Expert-Consensus Novelty Alignment (ECNA) score in {−1, 0, 1, 2}, with justification
Splits	31,411 train / 6,488 test; 500-paper evaluation subset
Format	Parquet, one record per submission

Venue	2022	2023	2024	2025	Total
ICLR	2,420	3,431	6,644	10,474	22,969
NeurIPS	2,543	3,108	3,819	5,460	14,930

Quick start

from datasets import load_dataset

ds = load_dataset("AbeerMostafa/Novelty-Reviewer-Dataset")
paper = ds["test"][0]
print(paper["title"])
print(paper["novelty_score"], paper["novelty_score_justification"])

# The 500 papers evaluated in the paper
benchmark = ds["test"].filter(lambda r: r["in_eval_subset"])

The companion package adds helpers (load_novelty_dataset, eval_subset, reviews_frame), PDF download from OpenReview, and tutorial notebooks.

Novelty labels

Each submission carries an Expert-Consensus Novelty Alignment (ECNA), derived from the stance of its reviewers:

Score	Label	Definition	Submissions
−1	Not novel	The work is derivative and largely replicates existing approaches.	2,924 (7.7 %)
0	Limited novelty	The work makes minor variations of known methods.	7,606 (20.1 %)
1	Moderately novel	The work is original but overlaps notably with prior work or mainly extends existing ideas.	24,177 (63.8 %)
2	Highly novel	The work introduces fundamentally new ideas, methods, or problem formulations.	3,183 (8.4 %)

The labels reflect reviewing practice at selective venues: most submissions are judged moderately novel, and both extremes are rare. Reviewer agreement varies: all reviewers agree for 18,727 submissions, a majority agrees for 12,477, and opinions are evenly split for 6,695.

Dataset structure

`default`: one record per submission

Field	Type	Description
`paper_id`	string	OpenReview identifier; the paper is at `https://openreview.net/forum?id=<paper_id>`
`title`	string	Title from OpenReview
`abstract`	string	Abstract from OpenReview; links are replaced by `[URL]`
`venue`, `year`	string, int16	`ICLR` or `NeurIPS`; conference year
`split`	string	`train` or `test`
`in_eval_subset`	bool	Marks the 500 test submissions evaluated in the paper
`num_reviews`	int8	Number of official reviews
`reviews`	list of struct	Official reviews (see below)
`review_template`	string	The venue's review form: section names and score rubrics
`novelty_excerpts`	list of string	Verbatim review passages that assess novelty
`novelty_extraction_raw`	string	Raw output of the extraction model, from which `novelty_excerpts` is parsed
`novelty_score`	int8	ECNA score in {−1, 0, 1, 2}
`novelty_score_justification`	string	Short justification of the score
`novelty_detailed_assessment`	string	Detailed paper-level novelty assessment
`novelty_summary`	string	Complete generated assessment (score, justification and detailed assessment)

Each element of reviews:

Field	Type	Description
`review_text`	string	Full review in markdown, one `##` section per field of the review form
`rating`, `confidence`	int8	Overall rating (`Recommendation` at ICLR 2022–2023) and reviewer confidence
`soundness`, `presentation`, `contribution`	int8	ICLR 2024–2025, NeurIPS 2022–2024
`correctness`, `technical_novelty_and_significance`, `empirical_novelty_and_significance`	int8	ICLR 2022–2023
`originality`, `significance`, `quality`, `clarity`	int8	NeurIPS 2025

Scores that a venue's review form does not contain are null. NeurIPS 2025 reviews use a different form without overall rating and confidence, and no review template is provided for them.

`baseline_reviews`: model-generated reviews

Full reviews generated by five LLMs (Llama-OpenReviewer-8B, Llama-3.1-70B-Instruct, Llama-3.1-8B-Instruct, gpt-4o-2024-11-20, claude-3.5-sonnet) for the 400 ICLR 2025 and NeurIPS 2024 submissions of the OpenReviewer benchmark (Idahl & Ahmadi, NAACL 2025). Fields: paper_id, model, generated_review.

Splits

Split	Submissions	Description
`train`	31,411	25,951 submissions used to train the Novelty Reviewer model, plus NeurIPS 2025
`test`	6,488	Held-out submissions from ICLR 2022–2025 and NeurIPS 2022–2024
`test` with `in_eval_subset`	500	Evaluation subset of the paper

Splits are defined per submission, so all reviews of a paper fall into the same split. NeurIPS 2025 was collected after the experiments of the paper and is released as additional training data.

Dataset creation

Source data

Submissions and official reviews were collected through the OpenReview API. Titles and abstracts are the official OpenReview metadata. Full paper texts are not redistributed; they remain available on OpenReview, and the companion code downloads and converts them to the input format of the model.

Annotation process

Labels were generated in two stages with Llama-3.1-8B-Instruct:

Extraction. For each review, the model extracts every passage that bears on novelty, both explicit judgments and implicit signals such as comparisons with prior work or criticism of incremental contributions. Passages are copied verbatim (greedy decoding).
Aggregation. For each submission, the model reads all reviews together with per-review novelty stances and writes a paper-level assessment and an ECNA score (temperature 0.3, top-p 0.9). The score reflects only the reviewers' stance, not the model's own judgment.

All prompts are published in the code repository.

Quality validation

Fidelity. On 500 sampled submissions, each sentence of the generated assessment was matched to its most similar review sentence; the mean cosine similarity is 0.78, and an entailment–contradiction NLI score between assessments and reviews is 0.79.
Consistency with reviewer scores. Averaged per submission, ECNA labels correlate with the reviewers' technical novelty score (Spearman ρ = 0.63, ICLR 2022–2023, 5,851 submissions) and contribution score (ρ = 0.52, ICLR 2024–2025 and NeurIPS 2022–2024, 26,588 submissions). The aggregation model reads the full reviews, which contain these scores, so this is a consistency check rather than independent validation.

Personal and sensitive information

The dataset is built from public OpenReview data, and identifying information was removed from all text fields:

Full paper texts are excluded, since camera-ready versions list authors, affiliations and e-mail addresses.
Each text field (titles, abstracts, reviews, extractions, assessments, baseline reviews) was matched against the author list of its submission. Author names, including reversed and initialized forms, are replaced by [REDACTED]; e-mail addresses are removed.
All links in abstracts, and links in other fields that contain an author's name, are replaced by [URL].
7,297 redactions were made in 6,132 submissions. Reviewer identities are not part of the data.

Citations by surname (e.g. "Chen et al.") are kept. Two submissions are no longer visible on OpenReview and could not be checked against their author lists.

Data quality

The extraction model returns JSON, from which novelty_excerpts is parsed. Outcomes per submission:

Outcome	Submissions	Meaning
Valid	30,655	Parsed directly
Repaired	1,345	Parsed after escaping LaTeX backslashes or closing brackets
Repaired, truncated	200	Output was cut off; the last excerpt may be incomplete
Recovered	118	Excerpts recovered line by line
Failed	121	Not parsable; `novelty_excerpts` is empty and the output is in `novelty_extraction_raw`
Not extracted	5,460	NeurIPS 2025; extraction to be added in a later version

625 assessments do not follow the justification/detailed-assessment format; their full text is in novelty_summary. Nine NeurIPS 2025 assessments contain an out-of-scale score and have novelty_score set to null.

Considerations for using the data

Intended uses. Training and evaluating models that assess or explain research novelty; novelty-aware review generation; related-work retrieval, using reviewer comparisons with prior work as signals; meta-research on peer review, such as reviewer agreement and the evolution of novelty judgments.

Out-of-scope uses. Automated accept/reject decisions; evaluating or ranking individual researchers; attempts to re-identify anonymous reviewers or redacted authors.

Limitations and biases.

ECNA labels encode reviewer opinion, not ground-truth novelty, and inherit the subjectivity of peer review.
Labels are generated by an LLM. They are validated on a sample, not verified for every submission.
The class distribution is skewed toward moderate novelty.
The data covers two English-language machine learning venues and may not transfer to other fields.
Submissions include accepted and rejected papers as available on OpenReview; withdrawn or non-public submissions are missing, so the data is not a complete record of each venue.

License

The dataset is released under CC BY-NC 4.0. Reviews and abstracts come from OpenReview, under licenses that vary by venue and paper (mostly CC BY 4.0, some CC BY-NC 4.0); CC BY-NC 4.0 covers all of them. Please credit OpenReview and the original reviewers and authors. The extractions and assessments were generated with Llama 3.1 (Llama 3.1 Community License).

Citation

@misc{mostafa2026novelty,
  title         = {Are We Truly Innovating? A Qualitative and Quantitative Study of Originality in AI Research Papers},
  author        = {Mostafa, Abeer and Nguyen, Thi Huyen and Ahmadi, Zahra},
  year          = {2026},
  eprint        = {2602.06054},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CL},
  url           = {https://arxiv.org/abs/2602.06054}
}