Text Snippet Merging for Context-Based File Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing file search systems are inefficient in locating specific files due to the large volume of data and the challenge of searching non-text files, as well as the redundancy in text snippets which worsen retrieval algorithms, leading to ineffective content-based searches.
Innovation Solution
A system that captures and indexes text snippets from the user interface level, merging similar snippets using similarity metrics to reduce redundancy, creating a three-level index that maps terms to snippets to documents, enabling task-based classification and retrieval by matching displayed text with files used during specific tasks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If text snippets are captured from user interface to enable task-based search, then search effectiveness is improved, but storage requirements and data redundancy increase
Solution Approach 1:
The patent merges multiple identical or similar text snippets into a single representative snippet. When the same text snippet appears multiple times in the user interface trace, the system consolidates these duplicates into one entry, thereby reducing storage requirements while preserving search effectiveness. This is achieved by comparing snippets for identity or similarity and retaining only unique representations.
Solution Approach 2:
The patent transforms the raw text snippet data by changing its structural parameters. Instead of storing complete redundant snippets, the system extracts and stores only unique snippets with their associated metadata (timestamps, file associations). This parameter change from storing all instances to storing unique instances reduces the quantity of stored data while maintaining search capability.
2Loss of information
If all text snippets are indexed individually, then retrieval completeness is improved, but search performance deteriorates due to redundancy
Solution Approach 1:
The patent merges identical or similar text snippets into single indexed entries. By consolidating duplicate snippets before indexing, the system reduces the total number of indexed items, which improves search performance by reducing the search space. Retrieval completeness is maintained because the unique snippet content is preserved and associated with all relevant files and timestamps.
Solution Approach 2:
The patent extracts only the unique or representative text snippets from the complete set of captured snippets. By taking out and retaining only the essential unique content while discarding redundant duplicates, the system achieves both retrieval completeness (all unique content is preserved) and improved search performance (fewer items to search through).
3Adaptability or versatility
If content-based search is used for non-text files, then file location capability is improved, but search accuracy deteriorates due to lack of indexable data
Solution Approach 1:
The patent uses text snippets captured from the user interface as an intermediary to bridge the gap between non-text files and search functionality. When a user interacts with non-text files (images, videos), the surrounding text in the UI (filenames, descriptions, metadata displayed) is captured as snippets. These intermediary text elements enable search capability for non-text files without requiring the files themselves to be indexable, while maintaining search accuracy through the use of relevant contextual text.
Data Source
AI summary
One embodiment is a method that stores user interface events and file system events to capture text snippets visible on an output device. Similar text snippets are merged into a context graph that is used to locate a document from a query.