This theoretical framework to understand annotations taken through a data visualization was described in two articles (see "links" above) and serve as a base for the Colvis - Interface project.
Abstract of the latest paper:
Annotations are observations made during the exploration of a specific data visualization, which can be recorded as text or visual data selection. This article introduces a classification framework that allows a systematic description of annotations. To create the framework, a real dataset of 302 annotations authored by 16 analysts was collected. Then, three coders independently described the annotations by eliciting categories that emerged from the data. This process was repeated for several iterative phases, until a high inter-coder agreement was reached. The final descriptive attribute-based framework comprises the following dimensions: insight on data, multiple observations, data units, level of interpretation, co-references and detected patterns. This framework has the potential to provide a common ground to assess the expressiveness of different types of visualization over the same data. This potential is further illustrated in a concrete use case.