ML Patent Search System Vector Angle Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current natural language processing techniques are inadequate for accurately comparing and evaluating the novelty of technical concepts across documents, particularly in patent searches, as they rely on manual human deduction and are inefficient in handling large data masses.
Innovation Solution
A machine learning-based system is developed that trains neural networks using pairs of claim and specification blocks from the same patent document to minimize vector angles between relevant concepts and maximize angles between irrelevant ones, enabling more accurate automated novelty evaluations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual human deduction is used for comparing technical concepts, then accuracy and reliability of novelty evaluation is improved, but productivity and efficiency deteriorates due to time-consuming manual work
Solution Approach 1:
The patent replaces manual human deduction with an automated machine learning system that uses neural networks to compare technical concepts. The system embeds claim and specification blocks into vector spaces and calculates cosine similarities to automatically evaluate novelty, substituting the mechanical process of manual reading and comparison with an automated computational approach.
Solution Approach 2:
The system enables self-service by allowing the machine learning model to autonomously perform novelty evaluations without requiring human intervention. The trained model independently processes patent documents, compares technical concepts, and generates search results, making the system self-sufficient for routine patent search tasks.
2Measurement precision
If keyword searches with boolean strategies are used, then search precision for specific keywords is improved, but ease of operation deteriorates due to requirement for expertise and time-consuming strategy creation
Solution Approach 1:
The system eliminates the need for users to manually create boolean search strategies by providing self-service through automated machine learning. The trained model automatically understands technical concepts, extracts relevant information from patent documents, and generates appropriate search queries without requiring user expertise in search strategy formulation.
Solution Approach 2:
The patent introduces an intermediary layer in the form of a trained machine learning model that mediates between user search queries and patent documents. Instead of requiring users to directly construct complex boolean strategies, the intermediary model translates simple user inputs into effective search results by leveraging its pre-trained understanding of technical concepts.
3Ease of operation
If semantic searches with AI technologies are used, then ease of operation is improved by reducing expertise requirements, but measurement precision deteriorates in patent novelty searches due to limited ability to evaluate novelty
Solution Approach 1:
The patent changes the fundamental parameters of semantic search by transitioning from generic similarity matching to domain-specific novelty evaluation. The machine learning model is trained specifically on patent data with labeled novelty information, allowing it to adjust its embedding and comparison parameters to accurately distinguish between relevant prior art and unrelated documents, thereby improving precision for novelty evaluation.
Data Source
AI summary
The invention provides a method and system for training a machine learning-based patent search or novelty evaluation system. The method comprises providing a plurality of patent documents each having a computer-identifiable claim block and specification block, the specification block including at least part of the description of the patent document. The method also comprises providing a machine learning model and training the machine learning model using a training data set comprising data from said patent documents for forming a trained machine learning model. According to the invention, the training comprises using pairs of claim blocks and specification blocks originating from the same patent document as training cases of said training data set.