DPP
A standard protocol for document parsing quality comprising 140 rules, used to evaluate LLM document parsing results. It aims to unify the criteria for judging parsing correctness and to reduce subjective disagreement in evaluation.
Timeline (last 90 days)
The maintenance cost of the rule set, the degree of adaptation to different document types, and possible conflicts among rules are issues DPP needs to discuss going forward.
If DPP is accepted by the community, the most direct change would be improved comparability of evaluations, reducing ranking distortion caused by style preferences.
DPP's 140 rules can in turn guide the construction of training data, incorporating the edge cases covered by the rules into the annotation system, which helps models handle ambiguous structures.
DPP (Document Parsing Protocol) is a set of 140-rule document parsing evaluation rubrics introduced by its publisher on Twitter.
DPP's 140 rules cover multiple aspects of document parsing such as tables, forms, and charts, and are used to turn vague parsing judgments into checkable, reproducible items.
The publisher explicitly states that DPP is not yet a complete evaluation benchmark, but rather an initial evaluation guide.
DPP can serve as a checklist for human or model-assisted evaluation and provides developers with a list of improvement directions to check against.
DiCE使用DPP（行列式点过程）建模多样性，通过半正定核矩阵L参数化，子集S被选中的概率正比于L_S的行列式
DiCE使用DPP（行列式点过程）作为多样性建模工具，并支持对特征施加可行性约束，支持模型无关模式
All Facts (9)
The maintenance cost of the rule set, the degree of adaptation to different document types, and possible conflicts among rules are issues DPP needs to discuss going forward.
50%UnverifiedThe publisher explicitly states that DPP is not yet a complete evaluation benchmark, but rather an initial evaluation guide.
50%UnverifiedDPP can serve as a checklist for human or model-assisted evaluation and provides developers with a list of improvement directions to check against.
50%UnverifiedDPP's 140 rules can in turn guide the construction of training data, incorporating the edge cases covered by the rules into the annotation system, which helps models handle ambiguous structures.
50%UnverifiedDPP (Document Parsing Protocol) is a set of 140-rule document parsing evaluation rubrics introduced by its publisher on Twitter.
50%UnverifiedDPP's 140 rules cover multiple aspects of document parsing such as tables, forms, and charts, and are used to turn vague parsing judgments into checkable, reproducible items.
50%UnverifiedDiCE使用DPP（行列式点过程）建模多样性，通过半正定核矩阵L参数化，子集S被选中的概率正比于L_S的行列式
50%UnverifiedDiCE使用DPP（行列式点过程）作为多样性建模工具，并支持对特征施加可行性约束，支持模型无关模式
50%UnverifiedIf DPP is accepted by the community, the most direct change would be improved comparability of evaluations, reducing ranking distortion caused by style preferences.
50%