Scholarlink

<p>Digital technologies offer significant economic, social, and environmental benefits in agriculture and rural areas. However, the level of diffusion of these technologies among farmers remains low. The Operational Groups (OGs) of the European Innovation Partnership for Agricultural Productivity and Sustainability (EIP-AGRI) can promote digitalization in agriculture by disseminating innovative digital solutions. Unfortunately, the existing project information is not classified, representing a major obstacle to the diffusion of innovations. To address this problem, this paper introduces a method that uses multi-label neural network models to automatically classify digital innovations in the projects of the EIP-AGRI OGs. The method tackles challenges such as data limitations and imbalances by adopting a data augmentation technique based on the generation of artificial synonyms and weighted loss functions. The study explores various deep learning models, including state-of-the-art pre-trained language models, and finds that the OpenAI generative pre-trained transformer has the best performance in automatically classifying projects. It also demonstrates that the use of artificial synonyms enhances the ability of neural networks to predict truly positive labels, i.e., to identify digital innovations in projects where they are actually present.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An augmented multi-label neural network-based approach for text classification in small and unbalanced datasets: the case of digital innovation in the EIP-AGRI Operational Groups

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Andrea Bonfiglio

摘要

Digital technologies offer significant economic, social, and environmental benefits in agriculture and rural areas. However, the level of diffusion of these technologies among farmers remains low. The Operational Groups (OGs) of the European Innovation Partnership for Agricultural Productivity and Sustainability (EIP-AGRI) can promote digitalization in agriculture by disseminating innovative digital solutions. Unfortunately, the existing project information is not classified, representing a major obstacle to the diffusion of innovations. To address this problem, this paper introduces a method that uses multi-label neural network models to automatically classify digital innovations in the projects of the EIP-AGRI OGs. The method tackles challenges such as data limitations and imbalances by adopting a data augmentation technique based on the generation of artificial synonyms and weighted loss functions. The study explores various deep learning models, including state-of-the-art pre-trained language models, and finds that the OpenAI generative pre-trained transformer has the best performance in automatically classifying projects. It also demonstrates that the use of artificial synonyms enhances the ability of neural networks to predict truly positive labels, i.e., to identify digital innovations in projects where they are actually present.