Machine Learning Topic Model for Privileged Document Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional e-discovery methods for identifying privileged documents are inaccurate, inconsistent, time-consuming, and inefficient due to reliance on keyword and name searches, which may miss relevant information and require recursive reviews of large datasets.
Innovation Solution
A computer-implemented method using a machine learning topic model to identify and rank potentially privileged documents by extracting spans from documents, generating topics, assigning scores based on legal topic probabilities, and adjusting scores for legal entities and communications, thereby prioritizing documents for further analysis.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If keyword and name searches are used to identify privileged documents, then the e-discovery process can be performed using conventional methods, but the accuracy of privilege review is less than optimal and may fail to identify items that include legal discussion without naming an attorney
Solution Approach 1:
The patent replaces conventional mechanical keyword and name search methods with a machine learning topic model system. The system uses natural language processing to generate topics from document spans and automatically identifies legal topics without relying on predefined keywords or attorney names, thereby improving accuracy and preventing missed privileged documents
Solution Approach 2:
The patent changes the search parameters from fixed keywords and names to dynamic topic probabilities generated by the machine learning model. By scoring documents based on the sum of probabilities assigned to legal topics, the system adapts to different document types and contexts, improving the precision of privilege review
2Ease of operation
If multiple reviewers perform conventional e-discovery with different keywords and names, then individual reviews can be conducted, but consistency of review becomes a problem as different reviewers may have different opinions
Solution Approach 1:
The patent implements a self-service system where the machine learning topic model automatically performs the review function that previously required human reviewers. The system consistently applies the same topic modeling algorithm and scoring thresholds to all documents, eliminating reviewer subjectivity while maintaining ease of operation through automated processing
Solution Approach 2:
The patent creates a universal review system that handles all document types through a single machine learning model. The topic model is trained to recognize legal discussions across various contexts and document formats, providing consistent review results regardless of the specific document characteristics or reviewer identity
3Reliability
If quality control processes are applied to identify new keywords or names, then review quality may be improved, but a recursive process is created that is time-consuming and expensive as each unsearched keyword requires a new review of the entirety of electronic data
Solution Approach 1:
The patent performs preliminary action by training the machine learning topic model on a sample set of documents before applying it to the entire corpus. The model learns legal topic patterns in advance, eliminating the need for recursive keyword identification and enabling direct application to all documents without iterative re-review
Solution Approach 2:
The patent extracts the essential function of quality control from the recursive keyword identification process and embeds it within the initial model training phase. By taking out the time-consuming recursive element and replacing it with a pre-trained model, the system maintains high reliability while eliminating repeated full-data reviews
4Reliability
If all documents identified as potentially privileged are reviewed without additional guidance, then comprehensive review can be performed, but the process is inefficient due to the large volume of documents that must be examined
Solution Approach 1:
The patent applies partial action by scoring all documents with the topic model but requiring detailed review only for those exceeding a threshold score. This approach maintains reliability by considering all documents while improving productivity by focusing intensive review resources on the most promising candidates
Solution Approach 2:
The patent segments the review process into two stages: automated scoring of all documents using the topic model, followed by selective human review of high-scoring documents. This segmentation maintains comprehensive coverage while dramatically improving efficiency by reducing the volume of documents requiring manual examination
Data Source
AI summary
A method for identifying and ranking potentially privileged documents using a machine learning topic model may include receiving a set of documents. The method may also include, for each of two or more documents in the set of documents, extracting a set of spans from the document, generating, using a machine learning topic model, a set of topics and a subset of legal topics for the set of spans, generating a vector of probabilities for each span with a probability being assigned to each topic in the set of topics for the span, assigning a score to one or more spans in the set of spans by summing the probabilities in the vector that are assigned to a topic in the subset of legal topics, and assigning a score to the document. The method may further include ranking the two or more documents by their assigned scores.