Temporal fact extraction of fruit cultivation technologies based on deep learning

Area of Science:

• Agricultural Science
• Natural Language Processing
• Information Extraction

Background:

• Fruit planting techniques vary regionally, leading to diverse and complex online information.
• Existing information extraction methods struggle with the temporal nuances crucial for cultivation.
• A need exists for specialized models to handle time-sensitive agricultural data.

Purpose of the Study:

• To develop an advanced model for joint extraction of temporal facts from fruit cultivation texts.
• To address the limitations of traditional fact extraction frameworks that ignore temporal information.
• To improve the usability of agricultural knowledge by incorporating time-based data.

Main Methods:

• Proposed Basic Fact Extraction and Multi-layer CRFs (BFE-MCRFs), an end-to-end neural network.
• Introduced an improved schema to incorporate a time dimension for temporal knowledge representation.
• Utilized multi-layer Conditional Random Fields to detect objects under predefined temporal relationships.

Main Results:

• The BFE-MCRFs model demonstrated superior performance in extracting temporal facts.
• Achieved state-of-the-art results on both public and self-constructed datasets.
• Significantly outperformed existing baseline models in temporal fact extraction accuracy.

Conclusions:

• BFE-MCRFs effectively extracts temporal facts from unstructured fruit cultivation data.
• The model's temporal dimension enhancement is vital for downstream agricultural applications.
• This approach offers a robust solution for managing and utilizing time-sensitive agricultural information.