Document Classification via Multi-Strategy Hypothetical Entailment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing document classification techniques, such as those using zero-shot classification, face challenges in accuracy and stability due to variations in hypothetical sentence generation for labels, leading to inconsistent classification results for the same label.
Innovation Solution
A document classification apparatus and method that selects multiple generation strategies for generating hypothetical sentences related to candidate classifications, and determines the classification based on entailment between the document and the hypothetical sentences, thereby improving accuracy and stability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a single hypothetical sentence is generated for each label using zero-shot classification, then the classification process is simple and fast, but the determination accuracy and stability are insufficient due to variations in hypothetical sentence generation
Solution Approach 1:
The patent segments the hypothetical sentence generation process into multiple independent generation strategies. Each strategy generates hypothetical sentences from different perspectives or templates, allowing the system to evaluate multiple hypotheses for each label rather than relying on a single generated sentence. This segmentation enables more accurate and stable classification by comparing results across different generation approaches.
Solution Approach 2:
The patent changes the parameter of hypothetical sentence generation by introducing multiple generation strategies with different templates, perspectives, or formulation approaches. By varying how hypothetical sentences are constructed (different parameters of generation), the system achieves more robust classification results that are less sensitive to variations in any single generation approach.
2Reliability
If multiple generation strategies are used to generate hypothetical sentences, then classification accuracy and stability improve, but the complexity of the classification system increases
Solution Approach 1:
The patent merges multiple generation strategies into a unified classification framework. Instead of treating each strategy separately, the system combines their outputs and integrates the results through a coherent evaluation process. This merging approach maintains classification stability while managing system complexity by providing a structured way to handle multiple strategies within a single system architecture.
Solution Approach 2:
The patent creates a universal classification framework that can accommodate multiple generation strategies through a common interface and evaluation mechanism. The system is designed to be multi-functional, handling different generation strategies, entailment evaluations, and result aggregations within a single versatile architecture, thereby improving reliability without proportionally increasing complexity.
3Measurement precision
If multiple hypothetical sentences are generated for each label, then the determination of matching degree becomes more accurate, but the computational time and resources increase
Solution Approach 1:
The patent performs preliminary actions by generating multiple hypothetical sentences in advance using different strategies before the actual classification decision is made. This preliminary generation of diverse hypotheses allows the system to have ready-made candidates for evaluation, improving the accuracy of matching degree determination while managing computational time through structured preprocessing.
Solution Approach 2:
The patent applies partial action by generating a limited but sufficient number of hypothetical sentences through selected generation strategies rather than exhaustively generating all possible variations. This approach achieves adequate accuracy improvement without incurring excessive computational time, balancing the trade-off between precision and time efficiency.
Data Source
AI summary
In order to classify, stably with high accuracy, a document to be classified, a document classification apparatus (1) includes: a strategy selection section (11) that selects at least one generation strategy from among a plurality of generation strategies for generating a hypothetical sentence related to a candidate classification as which a document is to be classified; a hypothetical sentence generation section (12) that generates, in accordance with the at least one generation strategy selected by the strategy selection section (11), the hypothetical sentence, which is a sentence related to the candidate classification; and a classification section (13) that determines, on the basis of entailment between the document and the hypothetical sentence, a classification as which the document is to be classified.