文本生成方法和对话文本生成方法
By converting object data into graph data and determining the encoder's position parameters and attention mask parameters, the problem of universality and accuracy of existing text generation models on different types of data is solved, achieving highly versatile and low-cost text generation.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- ALIBABA DAMO (HANGZHOU) TECH CO LTD
- Filing Date
- 2023-04-12
- Publication Date
- 2026-07-17
AI Technical Summary
Existing text generation models lack versatility and accuracy when faced with different types of structured data, resulting in high costs and difficulty in capturing the structured characteristics of the data.
The object data of the entity object is converted into graph data. The graph data is then converted into a graph embedding sequence through the embedding layer of the text generation model. The position parameters and attention mask parameters of the encoder are determined based on the graph embedding sequence. Text embedding features are then encoded and the target text is generated through the decoder.
It improves the versatility and accuracy of the text generation model, enabling it to accurately capture the structured characteristics of data when faced with different types of data, and reduces costs.
Smart Images
Figure CN116561269B_ABST