期刊
IEEE ACCESS
卷 9, 期 -, 页码 152892-152902出版社
IEEE-INST ELECTRICAL ELECTRONICS ENGINEERS INC
DOI: 10.1109/ACCESS.2021.3128178
关键词
Task analysis; Crops; Deep learning; Information retrieval; Semantics; Text recognition; Natural language processing; Agriculture entity recognition; WordNet; semantic class; named entity recognition; deep learning
资金
- SFI Strategic Partnerships Programme [16/SPP/3296]
- Origin Enterprises Plc
Agronomists in digital agriculture must make precise decisions based on knowledge and experience, including identifying agricultural entities in text data. This study proposes a new Agriculture Entity Recognition (AGER) approach, utilizing a two-stage process with deep learning to build an annotated corpus for agricultural entities and demonstrate the efficiency and robustness of the method.
In digital agriculture, agronomists are required to make timely, profitable and more actionable precise decisions based on knowledge and experience. The input can be cultivated and related agricultural data, and one of them is text data, including news articles, business news, policy documents, or farming notes. To process this kind of data, identifying agricultural entities in the text is necessary to update news with agricultural orientation. This task is called Agriculture Entity Recognition (AGER - a kind of Named Entity Recognition task, NER, in the agriculture domain). However, there are very few approaches on AGER because of a lack of the consistent tagset and resources. In this study, we developed a new tagset for AGER to cover popular concepts in agriculture and we also propose a process for this task that consists of two stages: in the first stage, we use semantic-based approaches for detecting agricultural entities and semi-automatically build an annotated corpus of agricultural entities, while in the second stage, we identify the agricultural entities from the plain text using a deep learning approach, train on the annotated corpus. For the evaluation and validation, we build an annotated agriculture corpus and demonstrated the efficiency and robustness of our approach.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据