Abstract Network analysis is an indispensable part of ecological studies. Specifically, networks have played a pivotal role in studying the diversity, dynamics and functionality of pollination systems. Recording plant-pollinator interaction networks is a laborious task, prone to missing or false negative interactions. Several methods enable the assessment of sampling completeness of the network with the use of species accumulation curves or Chao estimators. However, these methods do not provide a way to identify which interactions might be missed in the field. Methods that enable a more directed and focused field sampling are needed. Such methods would greatly benefit plant-pollinator studies and network studies in general. We additively decomposed the information of an interaction matrix into three parts: the difference in importance of the species, the specificity of the interactions, and the generality of the interactions. This information is exploited by a previously proposed linear filtering method to re-score absent interactions, thus pinpointing the likely missing interactions. We evaluated this approach using null models, intensive cross-validation, as well as external validation with the Web of Life database. By means of a case study, we provide insight into the structure of the network using an information-theoretic approach. We show how to use linear filtering to suggest missing interactions. A thorough evaluation shows that these results are both statistically stable and useful to guide the search for missing interactions in real-world networks. The non-uniformity of pollination interactions can be quantified using information theory and extracted using linear filtering. Our work can be valuable as a way to study different interaction networks as well as a tool to help identifying missing interactions.

中文翻译：

---

授粉网络中的信息内容揭示了缺失的相互作用

摘要 网络分析是生态学研究中不可或缺的一部分。具体而言，网络在研究授粉系统的多样性、动力学和功能方面发挥了关键作用。记录植物授粉者交互网络是一项费力的任务，容易丢失或假阴性交互。有几种方法可以使用物种积累曲线或 Chao 估计器来评估网络的采样完整性。然而，这些方法并没有提供一种方法来确定现场可能会遗漏哪些交互。需要能够进行更定向和更集中的现场采样的方法。这些方法将极大地有益于植物传粉者研究和网络研究。我们将交互矩阵的信息加性分解为三部分：物种重要性的差异、相互作用的特异性和相互作用的普遍性。先前提出的线性过滤方法利用此信息对缺失的交互进行重新评分，从而确定可能丢失的交互。我们使用空模型、密集交叉验证以及生命网络数据库的外部验证来评估这种方法。通过案例研究，我们使用信息论方法深入了解网络的结构。我们展示了如何使用线性过滤来建议缺失的交互。彻底的评估表明，这些结果在统计上是稳定的，并且有助于指导搜索现实世界网络中缺失的交互。授粉相互作用的非均匀性可以使用信息论量化并使用线性过滤提取。我们的工作作为研究不同交互网络的一种方式以及帮助识别缺失交互的工具可能很有价值。