Encrypted File Search Using LLM Keyword Vectors
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current encryption technologies compromise data security and efficiency when searching encrypted files, as conventional search methods require decryption, which is computationally intensive and prone to errors, especially with large volumes of data.
Innovation Solution
A method involving a large language model to identify keywords from files, generate keyword vectors, and index encrypted files with these vectors and metadata in a protected index, allowing secure and efficient searching without decryption.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional search methods are used on encrypted files, then data security is maintained, but search efficiency and accessibility deteriorate
Solution Approach 1:
The patent segments the file data into two distinct parts: encrypted content (maintaining security) and extracted keywords/metadata (enabling search). The large language model extracts keywords from the file content before encryption, creating separate searchable indices that do not compromise the encrypted data itself. This segmentation allows search operations to work on the keyword layer while the actual file content remains securely encrypted.
Solution Approach 2:
The patent introduces keywords and metadata as an intermediary layer between the search query and the encrypted file content. Instead of searching the encrypted data directly or decrypting it for search, the system uses extracted keywords as a mediator that bridges the gap between search functionality and security requirements. The search operates on these intermediary keywords rather than the sensitive encrypted content.
2Ease of operation
If data is decrypted for searching, then search accessibility improves, but computational resources and time increase substantially
Solution Approach 1:
The patent performs keyword extraction and metadata preparation in advance, before the actual search operation. The large language model processes the file content once to extract relevant keywords and generate metadata, which are then stored alongside the encrypted file. During search operations, the system only needs to query these pre-extracted keywords rather than decrypting and processing the entire file content, significantly reducing computational overhead during search.
3Productivity
If keywords are extracted before encryption, then search capability is maintained, but data confidentiality during processing is compromised
Solution Approach 1:
The patent extracts only the essential search-related information (keywords and metadata) from the file content, separating it from the sensitive data. The large language model processes the file to identify and extract only the keywords necessary for search functionality, leaving the rest of the content encrypted and protected. This selective extraction minimizes the amount of unencrypted data handled during processing.
Data Source
AI summary
A method, computer program product, and system are provided for managing encrypted files. A file is received, wherein the file comprises file metadata. A large language model identifies keywords from the file, and generates keyword vector(s) from the identified keywords. The file is encrypted and indexed with the keyword vector(s) and the file metadata in a protected index.