Skip to content

Propose Experiment: Which human-labeled topics differ #54

@pmhauck

Description

@pmhauck

We would like to understand how using abridged versus full articles causes certain qualitative features in topics to emerge or not.

We would like an interesting result such as: using abbridged API version down weights "Politics" topic in the resulting topic model.

We know this is done when we have proposed a measurement between matched topics.

-Proposed metric: Jenson-Shannon divergence.
-Once we have human-labeled topics, we are interested in which ones maximize or minimize the JS statistic.

Metadata

Metadata

Assignees

Labels

No labels
No labels

Type

No type

Projects

No projects

Milestone

No milestone

Relationships

None yet

Development

No branches or pull requests

Issue actions