Corporate Knowledge Base Vector Search Refinement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional data search methods, including exact matching and fuzzy matching, face challenges in accurately determining similar data as the number of characters increases, leading to low search accuracy. Additionally, vector search technology struggles with accuracy when dealing with large amounts of data.
Innovation Solution
The proposed solution involves an extraction system for a corporate knowledge base that includes a company knowledge base and a server-end device. The system vectorizes key words into key vectors, compares them with math vectors in the knowledge base, and integrates search results. The search results are then labelled, vectorized again, and stored as new math vectors, allowing for iterative vector searches to enhance accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If conventional exact matching or fuzzy matching is used for data search, then the search process is simple and fast, but the search accuracy deteriorates as the number of characters increases
Solution Approach 1:
The patent transforms text data into vector representations, changing the parameter space from discrete character matching to continuous vector space comparison. This allows the system to capture semantic meaning beyond exact character matches, improving search accuracy while maintaining computational efficiency through vector operations.
Solution Approach 2:
The patent replaces traditional mechanical string matching algorithms with vector-based similarity computation. Instead of character-by-character comparison, the system uses vector operations (dot product, cosine similarity) to measure semantic proximity, achieving both accuracy and speed.
2Measurement precision
If vector search technology is used to improve search accuracy, then the search accuracy improves, but the computational complexity and processing time increase significantly when dealing with large amounts of data
Solution Approach 1:
The patent pre-computes and stores vector representations of all patent data in advance, transforming raw text into searchable vector form before actual search queries. This preliminary action eliminates the need for real-time text processing during search, significantly reducing computational complexity while maintaining high accuracy.
Solution Approach 2:
The system maintains a continuously updated knowledge base of patent vectors, where new patents are automatically vectorized and added to the search index. This continuous preparation ensures that the search system always has pre-processed data ready, avoiding repeated computation for each query.
3Quantity of substance
If all patent data is directly vectorized for vector search, then the search coverage is comprehensive, but the search accuracy deteriorates due to the large amount of data
Solution Approach 1:
The patent segments patent data into meaningful units (titles, abstracts, claims, keywords) and creates separate vector representations for each segment. During search, the system can selectively combine or weight different segments, improving accuracy by focusing on the most relevant portions while maintaining comprehensive coverage of all patent data.
Solution Approach 2:
The system applies different vectorization strategies or weighting schemes to different parts of patent documents based on their relevance. For example, claims may be weighted more heavily than abstracts, or recent patents may receive different treatment than older ones, optimizing search accuracy for each local context while maintaining overall comprehensive coverage.
Data Source
AI summary
An extraction system for corporate knowledge base and a method thereof are disclosed. In the extraction system, a server-end device receives a key word, the key word is vectorized to perform a vector search in a company knowledge base, and a search result is labelled to generate a new vector which is then stored in the company knowledge base. The vector search is again performed on the company knowledge base based on the keyword vector.