Family-Based Document Review for Active Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional document review technologies in eDiscovery processes fail to efficiently present families of documents together, leading to duplicative and time-consuming reviews, and lack robust techniques for determining the stopping point in active learning processes, resulting in inefficient use of computational resources and increased costs.
Innovation Solution
Implementing a family-based review method that selects relevant documents using machine learning models, identifies relationally-linked documents, and generates batches for display, allowing for streamlined review and improved elusion testing by determining an appropriate stopping point based on error rate estimation algorithms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If conventional document review technologies rank and present individual documents in isolation, then reviewers can analyze documents separately, but reviewers must conduct duplicative reviews and cannot efficiently determine when to stop reviewing
Solution Approach 1:
The patent merges related documents into 'document families' that are reviewed together in a single pass. Instead of reviewing individual documents separately, the system groups documents by relationship (e.g., email threads, contracts with exhibits) and presents them as unified review units, eliminating duplicative reviews and improving overall review efficiency
Solution Approach 2:
The system performs preliminary actions by pre-identifying and grouping related documents into families before the review process begins. This preliminary organization allows reviewers to efficiently determine when to stop reviewing by evaluating complete document families rather than individual documents, reducing the time needed to reach confidence levels
2Reliability
If active learning processes review documents individually to achieve confidence levels, then computational resources can be managed, but the process becomes time-consuming and requires multiple processing cycles
Solution Approach 1:
The patent combines multiple documents into document families and reviews them together in a single active learning cycle. This approach maintains the confidence level guarantee by evaluating complete familial units while reducing the number of review rounds needed, as reviewers assess all related documents simultaneously rather than in separate passes
Solution Approach 2:
The system enables continuous review of document families without interruption or repeated processing cycles. By presenting complete familial units in each active learning iteration, the process maintains continuous useful action and achieves confidence levels more efficiently, avoiding the time loss associated with stopping and restarting review processes
3Measurement precision
If reviewers analyze each individual document separately, then document-specific analysis is thorough, but related documents must be re-reviewed multiple times
Solution Approach 1:
The patent merges related documents into document families while preserving the ability to analyze each document's specific content thoroughly. Reviewers examine complete familial units in context, maintaining analysis accuracy for each document while improving overall review throughput by eliminating the need to re-review related documents in subsequent passes
4Device complexity
If conventional systems present documents in isolation, then document presentation is simple, but compliance with discovery rules requiring contextual production is difficult
Solution Approach 1:
The patent merges related documents into document families that are presented together, naturally complying with discovery rules that require contextual production (e.g., Rule 106 and Rule 34). This approach maintains system simplicity by using straightforward grouping logic while significantly improving adaptability to legal compliance requirements, as complete document families are produced in a single review cycle
Data Source
AI summary
A active learning family-based review method includes selecting a document ranked as relevant by a machine learning model, identifying family documents relationally-linked to the ranked relevant document, generating a batch including the ranked relevant document adjacent to the family documents, and displaying the batch in a computing device. An active learning family-based review computing system includes a processor and a memory storing instructions that, when executed, cause the computing system to select a relevant document using machine learning, identify family documents, generate a batch including the relevant document adjacent to the family documents, and display the batch. A non-transitory computer readable medium stores program instructions that when executed, cause a computer system to select a relevant document using machine learning, identify family documents, generate a batch including the relevant document adjacent to the family documents, and display the batch.