Related Experiment Video
Updated: May 9, 2026
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
DS2PT: A Deep Two-Stage Patent Text Segmentation Framework Informed by Low-Latency Neural Network Characteristics
Boting Geng1, Hongxia Wang1, Pengliang Zhang1
1School of Computer Science and Technology, Zhejiang University of Water Resources and Electric Power, Hangzhou, China.
Big Data
|May 8, 2026
Summary
This study introduces a Deep Segmentation Model for Patent Text (DS²PT) to break down complex patent sentences into shorter, meaningful units. This AI-driven approach enhances patent analysis and search accuracy.
Area of Science:
- Computational Linguistics
- Artificial Intelligence
- Data Mining
Background:
- Patent text segmentation is crucial for patent analysis and search.
- Traditional methods are labor-intensive and lack generalizability.
- Complex patent sentences hinder downstream processing.
Purpose of the Study:
- To develop an automated, accurate method for patent text segmentation.
- To improve the efficiency and generalizability of patent data mining.
Main Methods:
- A two-stage framework, Deep Segmentation Model for Patent Text (DS²PT), was proposed.
- Stage 1: Coarse segmentation using a conditional random field model.
- Stage 2: Deep, context-aware segmentation using the ALBERT model.
Main Results:
- DS²PT significantly improves segmentation accuracy without semantic loss.
- The model effectively captures hierarchical contextual information.
- The approach shows potential for real-time patent analysis systems.
Conclusions:
- DS²PT offers a robust solution for patent text segmentation.
- The model's design is inspired by low-latency neural networks for efficiency.
- This work facilitates domain-adaptive patent data processing.
