TV Recommendation Engine

Improved content discovery through intelligent algorithms

Our recommendation technology employs a range of patented, self-learning, artificial intelligence algorithms that have been proven to more accurately match content offerings to individual viewer tastes and preferences in the course of extensive, multi-year consumer testing in four-person households in three European markets.

The online TV service Zattoo uses watchmi recommendations in its programming guide.

source: zattoo

The online TV service Zattoo uses watchmi recommendations in its programming guide.


Personalized recommendations are on the threshold of becoming ubiquitous in the living room and on connected devices as a trustworthy and highly relevant tool for individual content discovery. This is not only due to the emancipation of media consumers who demand instant gratification according to their needs, but also with regard to the ever more complex offerings of content producers and providers. The watchmi recommendation engine ensures that your customers will always receive the best recommendations available to them.


Developed by our partner XroadMedia, the TV recommendation engine of watchmi provides appealing results through a complete set of recommendation algorithms and engines:

  • Similarity: content-to-content recommendations, based on the similarity of assets (TV shows, movies, etc.)
  • Preference: personal profile recommendations, based on behavioral, implicit and explicit information of users or groups of users 
  • Collaborative: user-to-user recommendations, based on “people who liked this, also liked that” paradigm
  • Statistical: trend and hit list recommendations, e.g. “most watched”, “best rated”
  • Social: social profile and user-to-user recommendations, based on activity and connections in social networks
  • Editorial: Hand-picked recommendations from expert curators

Similarity Recommendations

Using the available program metadata, the similarity engine uses sophisticated algorithms to calculate the similarity between reference assets (e.g. those titles that have been watched before) and the available library of content. It then recommends items based on the similarity it finds ‐ a user watching an action movie for instance, would be recommended a list of similar titles with the same lead actors or similar storyline. 

Preference-based Recommendations

The preference engine learns the user’s taste by collecting explicit (e.g. like/dislike ratings) and implicit (e.g. VOD purchases) inputs and compiles those into individual profiles. Profiles are dynamic and can be associated with a single user, with a user group or with one or more personal content categories. The sophisticated algorithms result in preference learning up to six times faster than competitive solutions. Presenting recommendations in the context of “personal channels” provides the traditional lean‐back experience users are familiar with in a multi‐source (linear TV, VOD, Online Video) and multi‐screen (TV, computer, smartphone, tablet) environment. 

Collaborative Recommendations

The collaborative recommendation engine in watchmi analyzes aggregated viewing patterns and preferences. Based on its analyses, it promotes content discovery following the paradigm of "viewers who liked show A also liked show B." Collaborative filters introduce a social element to the provision of great television and video content.

Statistical Recommendations

Statistical recommendations draw on big data to visualize the most sought-after trends and topics. Deep metadata ensures high-quality content discovery across a variety of criteria, e.g. consumption, ratings, times, genres, etc. Popular use cases of the statistical recommender include „The Top 10 Most Viewed Assets“, „The Top 10 Best Rated Movies“, etc.

Social Recommendations

The social engine parses the viewer's social network profile (e.g. Facebook, YouTube or Google+) and recent postings (e.g. Facebook, Twitter) of the user and generates recommendations based on information derived from them. Sophisticated algorithms are used to select and prioritize profile information relevant to TV and entertainment, e.g. ratings of and references to actors, music, movies, TV shows, etc...   The social engine then recommends assets matching the user's social network profile. Moreover, recommendations can also be derived from the preferences (ratings) of the user’s trusted group of friends, resulting in truly personalized recommendations based on the social graph of the user. 

Editorial Recommendations

Editorial recommendations are provided by trustworthy and knowledgable curators. Being experts in their area of expertise, the hand-selected content discovery choices are highly relevant. Editorial recommendations assure that viewers receive recommendations that go beyond their own selection of criteria and get to discover something new once in a while.

see also: