SELINA.AI scrubs public, social media sites (text, audio and video) and runs the data through proprietary linguistic filters. The results are assigned dynamic ranges and blended in the Data Model with any other traditional data points.
SELINA.AI then runs a minimum of 1,000,000 simulations on the Model to define a prediction weight to an event. The Model can be back-tested. Data points can be adjusted. Additional simulations can be run.
Results are generated and graded into tiers. Detailed, post-simulation reports can be generated. Post-event, variance analysis is run on the model to identify opportunities.
SELINA continues to learn, grow and evolve over time.
#SELINAMMA is the first, in-production, use case for SELINA.AI. The question we often get asked is, ‘Why are you using this prediction modeler to forecast MMA fights’? Before I answer that, let me start with a little background.
SELINA.AI is different than other AI/ML projects. Other systems create predictions based off hard metrics. These may be trending stats, in-fight probabilities, career totals or a custom combination of all three. But all those metrics are physical-results / quantitative-based. An action happens. That action is measured in some way. That measurement is recorded. Then those recorded metrics are used in different algorithms to create predictions.
SELINA.AI is a bit different.
While we do use physical-results / quantitative-based metrics in our system, it is only a subset of the overall process. Rather, we quantify HQ and CONTXT to understand the human element behind the actions.
For HQ (Human Quotient), we pull aggregated social media data and run that through proprietary linguistic filters. So, in this case it is human communication (words/context/syntax) that is equivalent to the ‘action’ segment of other models, and a weighted value is assigned to that. That is all combined with location, time, usage, and more (where available), and then further woven into the larger model.
For CONTXT, we extract audio tracks out of recorded video from past events, transcribe it and then run that data through the linguistic filters and apply a similar process as we do with HQ. However, while HQ gives a value to the human preparation and activity leading to an event, CONTXT puts past metrics into perspective. An example of this is in a traditional model, you may have that a fighter may have lost by TKO in his last fight. The loss and loss type is a hard metric. But SELINA.CONTXT is able to extract the audio from the fight and uncover if an injury forced a TKO ending in the first 15sec (see, Blaydes v Aspinall, 7.23.2022). So while other models may use the metric of the loss, SELINA.AI puts the metric into perspective… giving it CONTXT.
As a result, our modeling is better suited to forecast events that have direct human involvement. So SELINA.AI will have more success forecasting a sporting event or human decision/action more than weather, a financial market, or anything of the like.
This is why MMA was a great place to start. A 1v1, human event with no weather details and limited outside influences compared to team sports.
We launched the MMA forecasts in May 2022 and have used a public forum to openly tweak the model and track the results. We then publicly validated her core selections by digitally signing them on blockchain ($JUNO) so they could not be altered after the fact. As of the official launch of #SELINAMMA in November 2022, we reached version 6 of the SELINA.AI model. At the start of 2023, we leveled up to version 10.
Once we saw a consistent success rate of over 70% based on our desired segment selection and a community that wanted more of the data, we built the #SELINAMMA app… the first, in-production use case for SELINA.AI. Full of Fighter Profiles, Socials, News, Metrics and more, it is a 1-stop resource that is a must for any MMA fan, reporter/content creator, or sports bettor.