Activity Tracing Through Event Correlation Across Multiple Software Applications

An example embodiment may involve identifying local traces of related events within a plurality of event data repositories, wherein each of the event data repositories is respectively associated with a software application; using a clustering model, assigning the local traces into clusters; determining positive rules that define when pairs of the local traces are linked to a common global trace, and negative rules that define when the pairs are linked to different global traces; linking the pairs into global traces; iteratively training a similarity model to project the local traces into a vector space such that the pairs that are linked to common global traces exhibit a greater similarity with one another than the pairs that are linked to different global traces; and based on the similarity model as trained, linking further local traces to the global traces.