ESTEEM: A Novel Framework for Qualitatively Evaluating and Visualizing Spatiotemporal Embeddings in Social Media
ESTEEM is an interactive web-based prototype visualization tool to allow you to explore dynamic word embeddings across regions using siple keyword queries.
To run the demo, install python and the following dependencies:
- Python 3.5
We developed and tested our tool in the Anaconda python distribution (download: https://www.continuum.io/downloads), which has all of the above dependencies.
Make sure you put some embedding data in the appropraite folder. By default, ESTEEM will look for data in esteem/data following the examples from the paper:
esteem server.py package.json src ... data Brussels <- name of dataset that appears in esteem Belgium.pkl <- name of region that appears in esteem France.pkl United Kingdom.pkl News unverified.pkl verified.pkl
The names of the files and folders determine how the regions and datasets appear in the tool. Each region gets its own pickle file.
The embeddings are stored as pickles of a multi-indexed pandas DataFrames with the following format:
- The first index is the keyword
- The second index is the timestamp (pandas DatetimeIndex)
- The columns are the embedding dimensions. Column names are ignored.
Once the data is in place and the depencencies are installed, you can start up the python server as follows:
and then go to:
The development server can be started with the command