Awesome

Topics Over Time

This is an open-source implementation of A Non-Markov Continuous-Time Model of Topical Trends by Xuerui Wang and Andrew McCallum. The paper associated each LDA topic with a beta distribution over timestamps which characterized the evolution of that topic with time.

Instructions

Sanitize main_pnas.py and visualize_pnas.py to ensure all input directories, input files, and output directories are present.
Run python main_pnas.py to execute Topics over Time algorithm.
Run python visualize_pnas.py to visualize the topic-word distributions as well as the beta distributions showing evolution of topics with time.

Dataset

The code is tested on the PNAS titles dataset. The dataset can be found here. The resulting model is pickled and stored in the results folder.

Results

Topic Distributions for PNAS Titles Dataset

Evolution of Topics for PNAS Titles Dataset

License

GNU General Public License