Scoring NBA Prospects via Social Media

Parker Seegmiller

Project Goal

  • Use social media to develop additional informative inputs for an NBA draft prospect predictive model
    • Determine whether a relationship exists between an NBA player’s social media presence and their on-court play
    • Use that relationship to aid in draft prospect projections

Project Timeline

  • Downloaded draft data (2010-2020)
  • Scraped twitter
  • Used IBM Natural Language Understanding API
  • 3 sets of scores - Sentiment, Usage Rates, Topics
    • XGBoost models predicting BPM
    • BPM projection percentiles
  • Dashboard
    • Projections functionality


  • Trained on 80% of the draft data (2010-2020)
  • XGBoost worked slightly more effectively than multiple regression
  • Filled null BPM values (never debuted) at the 5th percentile
  • Scores were not very predictive on their own
    • Added a projection percentile to my analysis | | RMSE | |

Next Steps

  • Determine helpfulness of scores when used in other draft models
  • Obtain all NBA player data
    • Initially focused on current draft prospects
  • Add similarity portion to projections tab