Testing out IBM Watson’s new Fantasy Football analysis
Recently, ESPN and IBM have partnered up to test out AI in fantasy football. As an avid fantasy football player myself, I was intrigued when I saw a new “Fantasy Insights with [IBM] Watson” feature for each of my players. The feature charted out fantasy experts’ overall feeling about each player over the past seven days by analyzing sentiments towards the player across thousands of articles. For instance, for one of my players Melvin Gordon, the chart portrays that over the past week, Gordon has hovered around a “neutral” consensus feeling. The feature also lists a few of the articles from which it drew these sentiments, which I found quite useful to explore articles at length and draw my own conclusions.
The feature also has a second section called “Week Projections,” which isn’t available until the start of the season next week. The section looks to be extremely similar to the weekly player projections which ESPN already provides. However, I hope that Watson can compile useful stats every week like defensive strength and the player’s historical performance against that defense, which can aid an owner’s decision-making. For instance, Watson already includes a convenient ‘compare players’ feature which puts their weekly highs and lows projections head to head. Optimally, AI in fantasy football would be able to group together relevant data to make fantasy owners' and analysts’ jobs easier.
IBM Watson’s fantasy football analysis system is centered around machine learning technology. Built by IBM iX and IBM Cloud, the system has been trained to read and understand articles and information, allowing it to ‘learn’ the game of football. Over the season, the machine can consume more than 3,000 sources per hour and produce accurate summaries of the authors’ sentiments, rankings, and conclusions. Human scientists used dozens of machine learning techniques to provide a model for the machine through supervised training. In a four-step procedure including the Watson Knowledge Studio, a vector model, a Deep Player state classifier, and finally a Multiple Regression for player projection, scientists feed the machine basic knowledge and keywords so that the system knows what to look for when reading sources.
According to the IBM Watson website, the machine was given more than 90 gigs of raw fantasy football text from past seasons to ingest into its vector model. After, scientists quizzed the machine on various topics, testing its accuracy on which players would bust, breakout, play meaningful minutes, and more. Overall, accuracy in specific tests ranged from 57% to 82%, percentages which far outpace most human analysts. Ultimately, in the fourth and final step, Watson inputs ESPN projections, expert sentiments, and other calculations in order to put out a combined ESPN/Watson projection and predictions for good and bad performances.
IBM Watson’s new ventures into fantasy football demonstrate AI’s growing impact in the sports industry. Whether fantasy owners agree with Watson’s conclusions or not, people can use features such as listed articles as a convenient tool to form their own analysis. Because I follow both artificial intelligence and fantasy football closely, I’m excited to see AI expand its capabilities in this popular branch of the sports world. With unlimited access to data and statistics and revolutionary machine learning technology, I believe AI can outperform human analysts very soon in the world of fantasy sports. However, because injuries and other game-changing factors are inevitable throughout the season, it is extremely hard to be accurate with all of your decisions. Although ESPN fantasy owners are excited to have the AI at their disposal, it concerns me that one day, with sufficient technology, such a machine could be 99% accurate, ruining what I feel is the appeal of fantasy football. For example, in 2016, running back David Johnson had a breakout year, clinching the championship for a lot of owners. And yet, just the following year, people who drafted him with a top-5 pick had to face a season-ending wrist injury. These surprises define fantasy football, and a machine which can predict them with extreme accuracy- while also being available to the public- might tarnish owners’ experiences. That being said, I am curious to see how the machine considers players' injury histories in their projections, and how good a Watson-led fantasy team would be.
Read more from RebellionResearch.com:
Why a Machine Learning Investment?
Interview with Astronaut Scott Kelly: An American Hero
Are Derek Jeter & the Miami Marlins Going to Face a Cash Crunch?
AQUAPONICS: How Advanced Technology Grows Vegetables In The Desert
A New Breed of Airline Looks to Take the World by Storm!
$91 Million for a Pair of Gloves: Chelsea Leads Global Sports' Spending Spree
Shedding Light on Dark Matter: Using Machine Learning to Unravel Physics’ Hardest Questions
On Black Holes: Gateway to Another Dimension, or Ghosts of Stars’ Pasts?
The World Cup Does Not Have a Lasting Positive Impact on Hosting Countries…
Dating Wine Using Nuclear Signatures
Written by Michael Ding & Edited by Rachel Weissman & Alexander Fleiss