ThePodcastRecommender app is live at www.ThePodcastRecommender.com
A novel recommender for Podcasts.
A recommender system to find similar podcasts. Currently, users have to sift through data on iTunes to attempt to find similar podcasts. Here I leverage podcast similarity along with 'user' data to create a podcast recommender. This recommender will take one or more podcasts and use that information to recommend similar podcasts. Using this system will enable the user to find a similar podcast in a much easier manner.
Podcast Data was collected with the following tools:
- BeautifulSoup
- Itunes Website: Podcast Metadata
- Bing: Podcast Twitter Handles
- Twitter: Twitter API
- MongoDB: Used for Data Storage
Scripts: - Podcast_Meta.py: scraping website and placing data into mongodb - ParrallelMEta.py: running Podcast_Meta.py in parallel - ObtainHandles.py: using data from itunes to grab twitter handles - GetTwitterFollowers.py: Obtaining Twitter followers
I used SVD, K-means, and several NLP methods including TF-IDF and NMF to generate features for the podcast recommender.
clean_podcast_csv.py: obtains the saved data frame and cleans it for prediction. PodcastFeatureEngineering.py: is a child of clean_podcast_csv.py and allows for calling additional functions for feature engineering.
I used scikit-learn and plotly to perform and visualize the exploratory data analysis.
Using Flask, I created ThePodcastRecommender.com. run.py is the flask app I created to generate the website
