CAREER:Ontology-powered Named Entity Recognition and Robust Semantic Similarity Metrics

该项目的目标是开发新的方法（即自然语言处理方法）来识别已发表的生物学研究中的术语组（即本体概念）。本体是一个组织信息的系统（例如生物学术语和这些术语之间的关系）。本体论可以很容易地被计算机分析，使大规模的生物分析成为可能。目前，从科学出版物到本体驱动格式的知识翻译仍然主要依赖于文献的人工管理；这个过程是不可扩展且乏味的。这个新项目所带来的进步将使人们能够从科学文献中发现知识，并将使计算分析能够回答复杂的生物学问题。该项目的教育影响将涉及开发有关前沿自然语言处理方法的新课程，以培养下一代跨学科计算机科学家。这项研究将推动机器学习模型在科学文献自动管理和知识发现领域的发展。该研究旨在推进和建立生物文献深度学习方法/技术领域的新范式，实现两个具体的研究成果：1)开发基于本体的命名实体识别的新型深度学习模型，以及2)开发用于评估上述深度学习模型的鲁棒语义相似度度量。稳健的语义相似性度量对于准确评估这里将要开发的深度学习模型的性能至关重要。这些相似性度量将广泛应用于生物学的几个领域，如假设进化过渡的候选基因，罕见的人类疾病诊断等。欲了解更多信息，请访问该项目的网站https://prashanti.github.io/deeplearningNER/.This。该奖项反映了美国国家科学基金会的法定使命，并通过基金会的智力价值和更广泛的影响审查标准进行评估，认为值得支持。

The goal of this project is to develop novel methods (i.e. natural language processing methods) for recognizing groups of terms (i.e. ontology concepts) in published biology research. Ontologies are a system to organize information (e.g. biology terms and relationships of those terms). Ontologies can be easily analyzed by computers, making large scale biological analyses possible. Currently, the translation of knowledge from scientific publications to an ontology-powered format still relies largely on manual curation of literature; this process is unscalable and tedious. Advances pioneered by this new project will enable knowledge discovery from scientific literature and will power computational analyses to answer complex biological questions. Educational impacts of this project will involve the development of new courses on cutting edge natural language processing methods to train the next generation of interdisciplinary computer scientists. The research will advance the state of art in development of machine learning models in the domain of automated curation and knowledge discovery in scientific literature. The research aims to advance and establish new paradigms in the area of deep learning methods/techniques for biological literature, towards two specific research outcomes- 1) development of novel deep learning models for ontology-based named entity recognition, and 2) development of robust semantic similarity metrics for evaluating the above deep learning models. Robust semantic similarity metrics are essential to accurately evaluate the performance of the deep learning models that will be developed here. These similarity metrics will be broadly applicable in several areas of biology such as hypothesizing candidate genes for evolutionary transitions, rare human disease diagnosis, etc. For further information visit the project website at https://prashanti.github.io/deeplearningNER/.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.