III: Small: Statistical Low-Rank Factorization Tools for Ecological Network Link Prediction

在许多种类的网络中，尝试预测哪些链接存在或可能存在是很有趣的，即使它们迄今尚未被观察到。例如，为了进行推荐，Netflix预测用户和电影之间的新链接——用户还没有看过但可能会喜欢的电影，这是基于类似用户对类似电影的评分。链接预测在电影和产品推荐方面已经得到了很好的研究，但生态系统中物种之间的链接预测却很少受到关注。了解和预测生态网络中未观察到的联系对自然资源管理、物种监测和作物生产具有重要意义。生态网络中的链接预测与推荐的链接预测既有相似之处，也有鲜明的对比。特别是，生态网络数据比用户-电影或用户-产品网络包含更多的错误，因为现场采样不能捕获网络中所有真正的链接。生态网络还受到物种丰富度和对有限资源的竞争等因素的影响，这些网络随着时间和空间的变化而变化。该项目由一个跨学科的研究小组领导，将彻底解决生态网络中出现的独特挑战。它将提出一个统一的框架，包括建模工具和计算基础设施，用于从时间和空间的不完整数据中分析生态网络。在理论和方法方面，生态网络中链路预测的许多关键方面，如统计模型、可扩展算法和多模态集成，仍然知之甚少。该项目将提供一套分析和计算工具来应对这些挑战。研究将遵循三个协同的方向。首先，基本框架的发展将产生统计模型和优化算法，以解释生态网络的独特特征。二是针对有限资源和物种竞争条件下的链接预测等具有挑战性的问题提出解决方案。第三，团队将对感兴趣的问题提供系统的评估方案。为生态网络设计有效的统计模型和健壮的、可扩展的算法是现代生态学研究和计算机科学的良好动机。该项目将利用现代数据科学工具，如低秩矩阵/张量分解、图形模型和神经网络，为生态网络分析奠定基础。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

In many kinds of networks, it is interesting to try to predict which links exist or could exist even though they have not been observed so far. For example, in order to make recommendations, Netflix predicts new links between users and movies - movies that the user has not watched yet but is likely to enjoy based on the ratings that similar users have given to similar movies. Link prediction has been well studied for movie and product recommendation, but link prediction between species in an ecosystem has received less attention. Understanding and predicting unobserved links in ecological networks is important for natural resource management, species monitoring, and crop production. Link prediction in ecological networks has some similarities to link prediction for recommendation but also some sharp contrasts. In particular, ecological network data contain more errors than user-movie or user-product networks because field sampling cannot capture all of the true links in the network. Ecological networks are also influenced by factors like the abundance of species and competition for limited resources, and these networks change over time and space. Led by a cross-disciplinary team of investigators, this project will thoroughly address the unique challenges that arise in ecological networks. It will put forth a unified framework including modeling tools and computational infrastructure for analyzing ecological networks from incomplete data over time and space.In terms of theory and methods, many key aspects of link prediction in ecological networks such as statistical models, scalable algorithms, and multimodality integration are still poorly understood. This project will provide a suite of analytical and computational tools addressing these challenges. The research will follow three synergistic directions. First, basic framework development will produce statistical models and optimization algorithms that account for the unique traits of ecological networks. Second, the researchers will put forth solutions for highly challenging issues like link prediction under limited resources and species competition. Third, the team will provide a systematic evaluation plan for the problems of interest. Designing effective statistical models and robust, scalable algorithms for ecological networks is well-motivated for both modern ecology research and computer science. This project will lay out the foundations for ecological network analytics by leveraging modern data science tools such as low-rank matrix/tensor factorization, graphical models, and neural networks.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.