Event Sequence Classification

I have a list of sequences.
Each sequence is a series of events.
Each event has a number of tags.

Every sequence is either of type A or type B. I’m trying to build a model to predict the type.

This looks like a classification of whole graphs, but because it’s sequential, it’s not your typical graph problem.

I was wondering if StellarGraph would be appropriate for this type of problem. If so, are there an examples of this type of problem? If not, where should I look?

Thank you!

Hi Jonathanng,
Thanks for your interest in StellarGraph. At the moment, we have a graph convolution and lstm based method for forecasting using spatio-temporal data: gcn-lstm.
However, it cannot be used for predicting sequence type - a kind of sequence classification problem, directly.
If I understand the problem you are trying to solve, you are trying to predict the label of the sequence, each of which is composed of a set of events and those events are the graphs (with tags as nodes and some relationship between the tags?). Basically you have a sequence of graphs and in the end you want to label the entire sequence as of type A or B?
We do not have any methods in the StellarGraph library to handle this type of problem.
We might add some methods in future that could help you with your problem.
In the meanwhile, here are a resources from the literature in this research space to give you some ideas. Specifically, these works are about predicting human actions based on sequence of movements.

Hope that helps.

1 Like