Alpha-9 Machine Learning Project
Faced with the need to grow our community reach, Open Form Games tasked me with finding a way to better "understand, build, and reinforce" our community.
Contributors Stuart Templeton
Created At Open Form Games
Role Lead Designer, Developer
The main challenge was to create a lean, automated mechanism that could be a part of the Open Form Games process. I couldn't engineer something that had a lot of process weight.
The secondary challenge was in identifying potential community members in the proverbial wild and automating an out-reach initiative. The data for this is necessarily huge.
One of the goals was to make a system that mirrored how Open Form Games deeply values its community members and would allow us to reach out to them personally, and not be subject to spam or automated messages.
I chose to use machine learning mechanisms to both construct the extremely large data-set and to understand it. This coded functionality could be used in an automated fashion.
The added benefit of this strategy is that we could leverage hardware to tackle the data in a programatic way, helping to reduce human error and to find potential connections that we hadn't before understood.
We would use this information to provide a conduit for personal interaction, acting as more of a mechanism for direct, personal introduction rather than marketing.
Assembling the training data from social media was relatively easy. We already had a small, but established community. I built the framework of the Alpha-9 system in ruby, leveraging ActiveRecord and a myriad of third party tools.
I iterated over the process of using this data to create a reasonably reliable way to identify traits of other potential community members. The initial prototype leaned heavily on ActiveRecord and SQL to find and evaluate unexpected relationships within the data. This cycle was used as a Litmus test to help weigh and understand incoming data and to empower the Alpha-9 system to come up with new ways to mine additional data and to self-correct.
Output typically was organized for human consumption in weighted trees of potential community members that would be a part of our out-reach initiative, like this:
Top potentials: 7=>0 5=>0 4=>0 3=>62 2=>2184 1=>24590
Alpha-9 would iterate over that data queue to initiate human-to-human contact and begin building our community, with a focus on heavily weighted interests.
I set out to create an automated tool to help us understand, build, and reinforce our community building efforts and succeeded beyond what I thought was possible. In addition to understanding our community, Alpha-9 challenged several misconceptions we held about our business and the market we were entering with Vegas Prime Retrograde. Notably:
The core result was a better understanding of our customer, our product, the market, and our value add. Each of these went on to directly impact Open Form Games' internal process and guiding philosophy, allowing for branding and marketing pivots that improved our position.