Document Reuse Prediction via ML Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Users of document authoring applications like Microsoft Word often face irrelevant lists of recent files, requiring manual searching to find relevant content from previously generated documents, which is inefficient and resource-intensive.
Innovation Solution
A system and method using machine learning models to identify and display candidate documents relevant to the current document by computing reuse scores and similarity scores based on document attributes and user behavior, presenting these in a user interface element for easy access.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a list of recent files is provided to users, then users can access previously generated documents, but the list may not be relevant to the user's current needs requiring manual searching
Solution Approach 1:
The system performs preliminary analysis of document content and user needs before the user requests documents. By pre-computing document embeddings and analyzing current document context, the system prepares candidate documents in advance, so when the user needs related documents, they are already identified and ready for immediate display, eliminating manual search time
Solution Approach 2:
The patent replaces the mechanical manual searching process with an automated machine learning system. Instead of users manually browsing through folders and recent files, the system uses embedding-based similarity computation and ranking algorithms to automatically identify and present relevant documents, substituting human cognitive effort with automated computational processes
2Loss of information
If manual searching through folders is performed to find relevant content, then users can locate specific documents, but the process is inefficient and resource-intensive
Solution Approach 1:
The system replaces inefficient manual folder browsing with automated machine learning-based document retrieval. By computing embeddings for all documents and using similarity search algorithms, the system automatically identifies relevant documents based on content semantics rather than manual navigation, significantly improving retrieval efficiency and reducing user effort
Solution Approach 2:
The patent transforms the document retrieval problem from a manual navigation task to a computational similarity search task by changing the parameter space. Instead of organizing documents by folder hierarchy, the system represents documents in a continuous embedding space where similarity is measured by distance metrics, enabling efficient retrieval based on content relevance rather than structural organization
3Measurement precision
If a machine learning model computes reuse scores for all documents, then relevant documents can be identified accurately, but computing resources are consumed
Solution Approach 1:
Instead of computing reuse scores for all documents in the repository, the system applies partial action by limiting computations to a subset of candidate documents. The two-stage approach first identifies a smaller set of promising candidates using efficient filtering, then applies comprehensive ML scoring only to these candidates, reducing overall computational resource consumption while maintaining accuracy
Solution Approach 2:
The patent segments the document retrieval process into distinct stages: initial filtering to identify candidate documents, followed by detailed scoring of only those candidates. This segmentation allows the system to apply different computational strategies to different subsets of documents, reducing the overall computational burden while maintaining high accuracy for the final results
Data Source
AI summary
A system and method for identifying relevant content from other documents and presenting candidate documents in a document authoring application is described. The system identifies a pattern of events from the document authoring application. The system receives a request to identify candidate documents pertinent to a document present in the document authoring application. A set of candidate documents is identified from a document library based on the pattern of events, a reuse score for each candidate document, a similarity score for each candidate document, and content in the document present in the document authoring application. The set of candidate documents is provided in a user interface element of the document authoring application.