Generative AI Search System Using Vector-Based Passage Indexing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional database search systems often return responses with extraneous information, requiring users or agents to sift through irrelevant data, which increases response time and negatively impacts user experience.
Innovation Solution
The system partitions articles into passages, generates token-based and vector-based objects for each passage, and uses these objects to index the data. When a query is submitted, the system converts it into a vector-based object and compares it to the indexed objects to identify relevant passages, returning these passages instead of full articles.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If the system returns full articles in response to queries, then complete information is provided, but users must sift through extraneous information which increases response time
Solution Approach 1:
The patent segments full articles into smaller passage units that can be independently indexed and retrieved. Instead of returning entire articles, the system divides content into manageable passages and retrieves only those relevant to the query, thus maintaining information completeness while reducing extraneous content.
Solution Approach 2:
The system extracts and retrieves only the specific passages containing relevant information rather than returning full articles. This extraction approach removes extraneous information while preserving the essential content needed to answer user queries.
2Loss of time
If the system partitions articles into passages and uses vector-based indexing, then relevant information is retrieved faster, but system complexity increases
Solution Approach 1:
The system performs preliminary actions by pre-partitioning articles into passages and pre-computing vector embeddings for each passage before queries are submitted. This advance preparation creates ready-to-search indexes, enabling fast retrieval without performing complex operations at query time.
Solution Approach 2:
The patent introduces vector embeddings as an intermediary representation between the original text passages and the query matching process. These vector objects serve as mediators that enable efficient similarity search without requiring complex text processing during query execution.
3Ease of operation
If the system returns targeted passages instead of full articles, then user experience is enhanced, but the system requires advanced processing capabilities
Solution Approach 1:
The patent replaces traditional mechanical text processing and keyword matching with vector-based semantic search. This substitution enables the system to understand and retrieve relevant passages based on meaning rather than exact keyword matches, significantly improving user experience through more accurate and context-aware results.
Data Source
AI summary
A method of data processing is described. The method includes converting a plain-text query into a vector-based object by using a text embedding function to process one or more tokens in the plain-text query. The method further includes retrieving a set of passages from a first datastore of the data processing system based on using one or more search indexes stored in a second datastore of the data processing system to compare the vector-based object and the one or more tokens in the plain-text query to vector-based objects and token-based objects associated with the set of passages. The method further includes generating a prompt that includes tokens from the plain-text query, tokens from one or more of the set of passages retrieved from the first datastore, and instructions for creating a response to the plain-text query. The method further includes transmitting the prompt to a large language model (LLM).