SMILE - Semantic Modelling of Intent through Large-language Evaluations

SMILE - 通过大语言评估进行意图语义建模

• 批准号: 10097766
• 金额: $ 14.11万
• 依托单位: OVERTONE TECHNOLOGY LIMITED
• 依托单位国家: 英国
• 项目类别: Collaborative R&D
• 财政年份: 2024
• 资助国家: 英国
• 起止时间: 2024 至 无数据
• 项目状态: 未结题
• 来源: https://gtr.ukri.org/projects?ref=10097766
• 关键词: SMILE; Semantic; Modelling; Intent; through

The SMILE project aims to take advances in Natural Language Processing on which Overtone has been working for years and take them even further to create a new way for content businesses and those who rely on content to understand why a piece performs as it does. It will create a prototype of a tone classifier that takes large language models and fine-tunes them to create nuanced understanding of whether a piece is humorous, serious, or positive, as well as other qualitative aspects that can impact how an article resonates with an audience. This project will help add to Overtone's offering by addressing one of the most pressing needs for publishers, news avoidance. News avoidance has been growing as audiences turn away from the repeated coverage of traumatic events. Overtone's models will allow publishers to optimise the delivery of their news to include both coverage of important, serious events while ensuring that audiences also are served lighter, more personal news that fulfils their needs to be inspired or entertained.

Smile Project旨在在Postrone多年来一直在运作的自然语言处理中进步，并将其进一步发展，以为内容业务和依靠内容来理解为什么作品表现出色的人创造一种新的方式。它将创建一个音调分类器的原型，该原型采用大型语言模型并微调它们，以对作品是幽默，严肃还是积极的作品以及其他可以影响文章如何引起观众共鸣的定性方面的细微理解。 该项目将通过满足发布商最紧迫的需求之一，避免新闻来帮助加入Optrone的产品。随着观众的反复报道，新闻避免的消息一直在增长。 Pollone的模型将允许发行商优化其新闻的交付，以包括对重要的，严重的事件的报道，同时确保观众也更轻，更个人的新闻，以满足他们的启发或娱乐的需求。