I want to predict the topic of a project tracker narrative entry to identify stages of the project. This entry is written by employees describing what they did in the project.

The projects are actually huge and they have thousands of entries. I want to group the tasks employees do in project phases by using topic modelling techniques. Fairly simple.. but what I want to include is something that considers the time where that entry happened because the entries may be similar and look like they are in the same group but if they happened with a lot of difference in time according to the project start date it is more probable that the entry belongs to another group of tasks.

I was thinking of using LDA without considering the entry date to get a distribution of topics per entry and then create a training dataset that includes days from the project start date to the entry date and join it with the topic probabilities as a feature vector and classify the real tasks considering those variables.

Are there any models that consider this time variable? Or have you heard of sequential topic modelling applications? That do this without the multi-step procedure? Do you have any examples?

Do you have any ideas for a methodology to do this?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.