FlowingFlowing
← 返回全部开发案例
开发阶段案例 · 模型盲评尚待独立人工复核，不代表正式 Benchmark 结论。
自动补全无 Evidence 更优15 / 162 · 2efbfb8491689b52

A Unified View of Deep Learning for Reaction and Retrosynthesis Prediction: Current Status and Future Challenges

定量生物学 · 2306.15890v1

FLOWING EVIDENCE BENCHMARK

怎么判断 Evidence 真的有帮助？

核心问题是：在同一写作位置、使用同一模型和任务时，提供检索文献片段会怎样改变首次输出？我们成对比较两种条件，保留持平、不可用和评审未完成的结果。

同一段原稿 · 比较是否提供 Evidence

01 · 固定写作位置

论文原稿

同一处写作点位 ▌

同一原稿位置的自动补全成对比较

02 · 构造两组输入

两组共用

原稿上下文、模型、任务和提示词

A · 提供 Evidence

额外提供检索文献片段

B · 不提供 Evidence

不提供检索文献片段

LLM

同一模型、同一版本

A → 首次输出

B → 首次输出

盲评 Agent

匿名标记两份首次输出为 X、Y

按续写质量判断：准确性、任务贴合度和可用性
输出：X 更优 / 持平 / Y 更优 / 两组均不可用

换序复评：X / Y → Y / X

盲评具体怎么判？

① 匿名两份输出
评审看到同一原稿和两份首次输出，但不知道哪份用了 Evidence。

② 比较并换序复评
根据当前任务的标准，按 X/Y、再按 Y/X 的顺序各评一次。

③ 复核分歧
两次结论不一致时再做第三次判定；未完成的评审也留在分母。

自动补全点位怎样分层？

在查看生成结果前，先核查检索片段是否含有能直接支撑下一步续写的具体命题；有则放在左侧机会组，否则放在右侧普通组。每个学科从可准入论文中均衡选取两组点位。50/50 是实验设计，不代表真实写作中两类点位各占一半。

AUTOCOMPLETE · 110

定量生物学、统计学、天体物理学

左侧 55 个、右侧 55 个点位；统计学采用 10 篇论文的复测结果。

AUTOCOMPLETE · 52

心理学与气候科学

左侧 26 个、右侧 26 个点位；心理学计入 9 篇，气候科学计入 4 篇。

表格里的百分比怎么算？

五学科共有 81 个左侧点位，其中 51 个点位的 Evidence 版本获评更优。持平、两组均不可用和评审未完成的点位仍计入分母。

51Evidence 版本更优
÷
81该层全部点位
=
63%该层 Evidence 获评更优的比例

来源贡献是另一项复核：已进入复核的 22 个 Evidence 获胜点位中，16 个确认直接使用了检索论文；另有 29 个胜出点位尚待复核。

当前是开发阶段的模型评审结果，尚未完成独立人工复核；这些数字不代表正式 Benchmark 结论，也不能单独证明因果关系。

原稿写作位置

原文摘录 · 非 PDF 页面

研究论文 · 原文片段

A Unified View of Deep Learning for Reaction and Retrosynthesis Prediction: Current Status and Future Challenges

4 Limitations and Challenges > 4.3 Limitations in Evaluation

…ugh, which only contains 50k backward reactions. Considering many recent approaches only have slight numerical differences in top-k accuracies, the current small-scale dataset is not adequate for testing model capability. Second, the current dataset will bias edit predictions and leaving group selections. Most backward reactions only have one single edit while very few have multiple edits, which results in poor prediction accuracy in multiple-edit cases. Additionally, leaving group distribution is very imbalanced, which makes graph-based models tend to select a few frequently occurring atoms.

…aluations since most of the reaction types in the testing set are already covered in the training set. Therefore, harder dataset split like scaffold splits and time splits should be taken into account for future evaluations of reaction predictions. Scaffold split is testing whether the model can generalize well under out-of-distribution settings, in which the training data distribution is very different from testing data distribution. Time split is splitting reactions in the order of discovery time. This split aims at testing whether the model can truly discover new chemical reactions or not.

↓ 此处生成下一段续写

黄色标出写作点位附近的原文；本页根据评测输入重排文本，不是原始 PDF 截图。

同一写作位置的成对对照盲评结果：无 Evidence 更优

两组使用同一模型、同一写作任务和原稿上下文，仅比较是否提供检索文献片段。下方展示评测记录中的首次输出。

Evidence

同一模型与写作位置 · 提供检索文献片段 · 首次输出

本次未获偏好

These limitations are not unique to retrosynthesis prediction; benchmarks of temporal distribution shift in other domains likewise report substantial performance drops when evaluation moves from in-distribution to out-of-distribution data, underscoring the need for evaluation protocols that reflect realistic deployment conditions.

提供给 Evidence 版本的文献片段

Wild-Time: A Benchmark of in-the-Wild Distribution Shift over Time

To address this gap, we curate Wild-Time, a benchmark of 5 datasets that reflect temporal distribution shifts arising in a variety of real-world applications, including patient prognosis and news classification. On these datasets, we systematically benchmark 13 prior approaches, including methods in…

展开完整文献摘录

To address this gap, we curate Wild-Time, a benchmark of 5 datasets that reflect temporal distribution shifts arising in a variety of real-world applications, including patient prognosis and news classification. On these datasets, we systematically benchmark 13 prior approaches, including methods in domain generalization, continual learning, self-supervised learning, and ensemble learning. We use two evaluation strategies: evaluation with a fixed time split (Eval-Fix) and evaluation with a data stream (Eval-Stream). Eval-Fix, our primary evaluation strategy, aims to provide a simple evaluation protocol, while Eval-Stream is more realistic for certain real-world applications. Under both evaluation strategies, we observe an average performance drop of 20% from in-distribution to out-of-distribution data.

无 Evidence

同一模型与写作位置 · 不提供检索文献片段 · 首次输出

盲评更优

Third, evaluation protocols often rely on random splits, which can overestimate performance because test reactions may share scaffolds or reaction types with the training set. Adopting distribution-shift-aware splits, such as scaffold or time splits, would provide a more realistic assessment of generalization to unseen chemistry.

模型盲评结论

模型盲评判定“无 Evidence 更优”。下方保留评审原始理由（英文），供核对判断依据。

查看模型评审原始理由（英文）

Output A continues the manuscript's argument coherently: it introduces a third limitation (evaluation protocols with random splits) that parallels the draft's 'First...Second...' structure, then proposes scaffold/time splits as solutions—directly flowing from the draft's final sentence about 'harder dataset split like scaffold splits and time splits.' The content is grounded in the draft's own discussion of distribution shifts and evaluation methodology. Output B introduces a cross-domain comparison to 'benchmarks of temporal distribution shift in other domains' that report 'substantial performance drops.' The source (Wild-Time) does discuss temporal distribution shifts and 20% performance drops, but the source is about patient prognosis and news classification—not chemistry or retrosynthesis. The claim that these limitations are 'not unique to retrosynthesis prediction' and that other domains 'likewise report substantial performance drops' constitutes an unsupported cross-domain generalization. The source does not establish that retrosynthesis prediction shares these patterns, nor does the draft mention other domains. The 'substantial performance drops' phrasing also loosely echoes the source's 20% figure without proper grounding. This is an unsupported claim that misleads by implying the source validates a chemistry-domain parallel.

文献片段是输入材料；出现于此不代表输出使用了它，也不代表它能够支持全部主张。原文与检索片段经过截取；页面没有展示模拟分数或模拟 PDF。