Why Recommendation Engines Do Not Work

May 16th, 2017

You finally get to enjoy some downtime, so you sit down to watch some television. Looking through the ‘recommended for you’ list, you still feel like there is nothing that appeals to you. With so much video content available, why does the recommender never seem to work?

Here are the top five reasons that recommendation engines do not work:

  1. Metadata is missing. When there are common gaps in the metadata, the recommendation engine will not return those assets. When there is metadata such as tags, genre, cast, and air date in the system that is inaccurate, missing, or written in a different style that is recognized by your ingestion system, all search and recommendation becomes flawed. Through aioTV’s Metadata Manager, video operators are empowered to provide customers with a viewing experience that is continuously improved by starting with complete, accurate, and standardized metadata.
  2. They restrict your discovery. Both collaborative and content-based recommendation systems are designed to assume that the users want to see content that’s similar to what they (or their peers) have favorited or viewed. This locks users into a similarity cluster without opportunity for new discovery beyond that cluster.
  3. Cold start is an issue. When the recommendations are built on intelligence from data analysis from the customer’s usage, a ‘cold start’ makes this impossible. Relevant data is required to provide accurate recommendations. You want to give your customers a place to start their experience and allow them to make it relevant. aioTV’s Personalization Engine provides a solution to cold start through multi-user profiles.
  4. They lack consistent user data. Inconsistent user data happens when a customer enjoys shows like ‘The Walking Dead’ and ‘Game of Thrones’, but ‘Paw Patrol’ and ‘The Bachelorette’ show up in their recommendations because other family members were watching similar shows earlier that day. With the Personalization Engine by aioTV, user switching can be implemented with multi-user accounts. Alternatively, it can allow personalization through global or session based filters.
  5. Context is king. Serendipitous recommendations are instances when the recommendation might make sense in many cases, but not without taking context into consideration. Yes, you might love watching Christmas movies, but only during the Holidays when your family is in town. Without context, the reasons for the recommendations are unclear. Context can be achieved through the aioTV Personalization Engine with user identified deep filters, flags, and preferences such as moods.

To learn more or schedule a demo for Metadata Manager or the Personalization Engine API, contact aioTV.