With nearly 1,700 players in 272 games, the quantity of information generated through the NFL football season is gigantic. Fantasy football team owners face complex decisions and a flood of data. It is usually a daunting task deciding who to start out, who to bench, and who to trade each week. It can be numerous fun – which is why the ESPN Fantasy app delights 12 million fantasy football users yearly.
Over the past eight years, IBM has worked closely with ESPN to counterpoint its fantasy football experience with insights that help fantasy owners of all skill levels make more informed decisions. These insights are available in the shape of player rankings that help end users find the very best players to trade or pick up off the waiver wire. And this 12 months the team goes even deeper, adding a brand new feature that uncovers the explanations behind the AI-generated grades. When a user taps on a player to accumulate or trade, an inventory of “Top Contributing Factors” is now displayed next to the numeric grade, providing team managers with a customized natural language explanation powered by the IBM® Granite™ Large Language Model (LLM) is generated.
As in real-world organizations, managers of fantasy football teams need clarity on the “why” behind the AI-generated output. “Explainability – the reasoning behind the output – is becoming almost as essential because the output itself,” says Aaron Baughman, IBM Fellow, Master Inventor and IBM Quantum™ Ambassador.
The key contributing aspects provide explanations based not only on a player's pure performance, but in addition on the particular way he complements your fantasy roster. This is how it really works.
Create the notes
As players grow to be available in your fantasy league's weekly waiver report, they may receive a customized waiver tier that takes under consideration the strengths and weaknesses of your roster, in addition to the rosters of other teams in your league. The grades even consider each league's custom settings. The rating is predicated on the worth they’d add to your team in comparison with the typical grade of players in the identical position in your current squad. Trade grades work similarly and are based on the relative profit that a player from an opponent's roster would bring to your team.
This process begins with calculating raw grades using a rules-based system combined with quite a lot of machine learning models. “Working with our developers and soccer experts at ESPN, we determine grades based on quite a lot of aspects,” said Aaron Baughman. “How many leagues own this player, what percentage of leagues start the player, what are their projected season stats, who they played against, who they may play against – a lot of these predictions are rolled up right into a single result.”
The scoring system is written in Node.js and Python and is powered by a scaled workflow that analyzes the billions of information points generated over the course of the NFL season. The results are then saved within the cloud.
Personalized analytics at scale
Closer to the buyer, a Node.js Team Needs application personalizes these grades specific to a user's team every 10 minutes. “The Team Needs application takes under consideration your squad, league and their specific settings based on an algorithm we developed that has been in development for over a 12 months,” says Baughman.
Why did it take so long? In a word: scale. There are 12 million or more Fantasy users each week, and a few days – typically Tuesdays and Thursdays – can see significant usage. This may end up in the app receiving 1000’s of hits per second, scaled across pods in a Red Hat® OpenShift® cluster.
When a team owner taps on a player on waivers in the appliance, this system runs and provides a grade tailored to their team. “Although points per reception (PPR) is essentially the most common scoring system, leagues can have an infinite variety of custom settings. So we couldn’t pre-calculate grades, that they had to be personalized,” says Baughman.
Beyond fantasy football, one can imagine many business use cases that may benefit from this mix of wealthy data evaluation and personalization.
Personalized explainability at scale
As in any leadership situation, it can be crucial for busy decision makers to succinctly summarize analytical insights. For this reason, the system summarizes these findings in a summary of the three most vital aspects.
“Working at this scale was difficult, so we invented a brand new algorithm,” says Baughman. “In watsonx™, a generative Granite AI model outputs cloze sentences. These incomplete sentences are then personalized based on value percentiles.” The possible adjectives and phrases change depending on the numerical value, for instance from “no help in any respect” to “significant improvement” (your position).
It is a multidimensional model with 12 varieties of contributing aspects and a whole bunch of permutations. Some of essentially the most common aspects are the proportion of teams that own a player, the proportion of teams that use a player, and the end result forecast. Inferences are conducted roughly every two hours and linked to raw grade results to fill the gap.
“When you click on an opponent's team in Trade Grades, you'll see three positions your team needs to deal with, resembling: B. Tight end, wide receiver or defense – the circle will turn green if that’s one in every of your needs,” says Baughman. “I also can discover my opponent’s needs, which promotes fair trade.”
Subsequent tapping on a player card displays the trade's overall rating, which is decided by combining raw evaluation and team requirements with the three most vital aspects.
Drive business insights and explainability
Whether you run a fantasy football team, an organization, or a business function, IBM watsonx™ AI platform can allow you to make higher decisions. It helps you collect, store and analyze enterprise data relevant to your use case, then leverage quite a lot of machine learning and traditional AI models to evaluate strengths, weaknesses and opportunities – and supply timely, Deliver current insights with fine-tuned contextual detail. Enterprise-focused Granite LLMs support natural language explainability and supply quite a lot of customer support features. Add the scalability of Red Hat® OpenShift® within the hybrid cloud and stay up for your enterprise' strongest season yet.
Discover the watsonx AI platform. Redefine the best way you’re employed with IBM AI consulting services
