Transductive Learning for Patent Industry Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic industry classification methods for patent texts are either labor-intensive due to high manual labeling requirements or lack accuracy due to reliance on computationally expensive natural language processing that overlooks important information, particularly word order relations and hierarchical vectors.
Innovation Solution
The proposed method employs a transductive learning approach to maximize the utilization of small annotation quantities, incorporates International Patent Classification (IPC) information to enrich dimensionality and reduce computational load, and utilizes hierarchical vectors generated from abstracts, claims, and descriptions to preserve word order relations and mine deep patent text information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual labeling method is used, then classification accuracy is high, but labeling speed is slow and cost is high
Solution Approach 1:
The patent segments the classification task into two stages: rough classification using IPC marks for initial grouping, and fine classification using hierarchical vectors for precise categorization. This segmentation allows manual labeling to be applied only where needed (fine classification) while automation handles the bulk (rough classification), resolving the contradiction between accuracy and productivity.
Solution Approach 2:
The patent introduces hierarchical vectors as an intermediary representation that bridges manual labeling and automatic classification. These vectors capture semantic information from patent texts and serve as a mediator between the raw text and the final classification, enabling automatic systems to achieve accuracy comparable to manual labeling without the time cost.
2Extent of automation
If natural language processing methods are used, then automatic classification is achieved, but computational cost is high and word order information is lost
Solution Approach 1:
The patent extracts hierarchical vectors from patent texts (abstract, claims, description) that encode semantic meaning while preserving word order relationships. By taking out these pre-computed vectors as features, the system avoids re-processing the entire text through computationally expensive NLP models during classification, thus achieving automation with reduced computational cost.
Solution Approach 2:
The patent performs preliminary action by pre-extracting and pre-processing text information to generate hierarchical vectors before the actual classification task. This preliminary processing captures all necessary semantic and structural information in advance, so that during classification only the vector matching is needed, dramatically reducing real-time computational requirements.
3Productivity
If existing NLP methods are used, then patent text analysis is performed, but deep information from hierarchical structures is omitted
Solution Approach 1:
The patent transitions from traditional flat NLP analysis to a hierarchical dimensional representation. Instead of analyzing text at a single level, the system extracts hierarchical vectors that operate across multiple levels (abstract, claims, description) and integrate them into a unified representation, capturing deep structural information that flat methods miss while maintaining processing efficiency.
Data Source
AI summary
An automatic industry classification method comprises: determining a scope of target patents, defining a target industry tree; generating marks on the target industry tree; performing a rough classification for the target patents by using the marks; performing a fine classification for the target patents according to a result of the rough classification. The automatic industry classification method and system provided by the present invention uses a transductive learning method, so that full mining of small annotation quantity information is realized. The automatic industry classification method and system uses information of IPC, so that information dimension is enriched, and calculation amount needed in the classification is reduced. The automatic industry classification method and system further uses the hierarchical vectors generated by the abstract, the claims and the description, so that the information of word order relation is reserved, and the patent text is deeply mined.