WatchMojo’s immersion into Deep and Machine Learning

Big data holds big promises for WatchMojo. Back in Q1 of 2017, in my quest to create proprietary products and innovations to link past video views and engagement to future video projects, I had heard about algorithms that can learn from data without relying on rules-based programming. To create value for our viewers, I asked if we had a dataset, then set our objective to discover which videos would be highly engaging and shared. We researched online, hired a data scientist and after a few attempts analyzing our records, got to develop statistical techniques and build interactive visualizations. In our first attempt, we envisioned gaining a better understanding of our data through machine learning, but truth be told, I grew more confused as the model evolved.

Our data scientist was busy using natural language processing to extract insights from our publishing data, but I asked him to work with the lead AI engineer from Erudite AI. Lucky for me, he agreed, and in that second initiative, Clemente Cuevas stepped in and taught us about the deep learning industry — and that’s when I was blown away.

We set a goal to train our big-data at a single point: predicting if our audience will view and engage with our content. Answering this makes video publishing more efficient by improving targeting and by identifying and eliminating the production budget that is wasted when bad videos get published. WatchMojo is a digitally networked enterprise — whether YouTube, Facebook, our website analytics, Google Doc spreadsheets — compounding data aggregation. Stakeholders cannot instantly access, and share, actionable information. We chose to build tools (Unity) and work with technology (supervised learning) to facilitate greater transparency and visibility throughout our distribution ecosystems.

In pursuit of this, we strived to paint a detailed portrait of each legacy video project: Memorizing titles, metadata, scrutinizing talent input, and cataloging audience interests, aspirations and desires. The result is a detailed, high-resolution close-up of each video project that reveals our next move.

To combat algorithmic bias, we constructed summaries and visualizations that examined why the model works (data science). This was an important way to discover flaws in our model. We then put our new models through multiple rounds of validation and real-world testing before we were confident enough to deploy them.

Fast-forward to today, and WatchMojo’s content plan is a proprietary software called Unity, a predictive analytics platform meant to inform WatchMojo’s content team which videos will be engaging and highly shared. Unity’s dashboard compares what’s going viral with historical data and trends. The Machine Learning Modules are used to make content appealing for the social web.

I know from managing products in a number of digitization revolution initiatives that garnering company-wide buy-in is very hard, so I credit both Ashkan Karbasfrooshan for thinking big and allowing me to make a strong case — and David for making the bet on Eroms, Julian, and the Erudite AI team to make WatchMojo’s first investment in Machine Learning.

This is a personal post. The opinions expressed here represent my own and not those of WatchMojo.

—

I thrive in middle market and scrappy startup environments, where everyone does a little of everything. I work closely with CSuite leadership teams towards product-market fit and operating at scale. Check me out on Reflektions.com and view my LinkedIn profile.

Thanks to David Massé, Hannah Cowen for reading drafts of this. Photo by Ant Rozetsky on Unsplash. Also, if you have any feedback or criticism about this article then shoot me an email.