Last.FM RecSys with Sentiment Analysis

The goal of this project is to test sentiment-aware recommender systems by use of the Last.FM API, sentiment analysis tools, and the NRC-VAD Lexicon (NRC Valence, Arousal, and Dominance Lexicon). It consists of various recommendation techniques (collaborative filtering and content-based) and sentiment analysis of textual content from different contexts (e.g Album or artist description, tracks, comments, tags, titles...) applied to recommendation.

The tools and recommendation models are showcased in LastMood, a web application that can be set up with the following section's steps.

Setup

1. To set up the required environment modules and libraries:
   • Install Conda for environment management.
   • Run ./setup.sh.
2. To set up the database (~760 MB):
   • Install PostgreSQL for database management.
   • Execute sudo -u postgres psql -c "CREATE USER alumnodb WITH PASSWORD 'alumnodb' createdb".
     • This creates user alumnodb with database creation permissions.
   • Go to /lastfm_RS and run make restore_db.
     • This creates lastfm_db's tables and loads all the necessary data.
3. To set up the recommendation data (~500 MB):
   • Go to /lastfm_RS and run make recsys_data.
     • This extracts the prepared dataset into /lastfm_RS/backend/research/recbole_research/saved/.
     • Otherwise, the data needs to be preprocessed, which takes considerable time and memory.
4. To start the web application on localhost, go to /lastfm_RS and run python3 manage.py runserver. It will be hosted at http://localhost:8000.

Note: The only available recommenders are Random, Pop and Cosine Similarities, due to size constraints.

References

• Obtaining Reliable Human Ratings of Valence, Arousal, and Dominance for 20,000 English Words (Mohammad, ACL 2018)
• https://github.com/dwzhou/SentimentAnalysis
• https://github.com/cjhutto/vaderSentiment
• https://github.com/pylast/pylast
• https://github.com/goldsmith/Wikipedia