Recommendation engine in the Telebreeze end-user application

The development of Information technologies has legitimized the process of constant changes in user behaviour and gave rise to new navigation mechanisms and methods of searching the information. Manual navigation was replaced by special tools of search and recommendation systems. In continuous flow of information it is too hard for the user to manually separate helpful content from information noise and bad quality content. Thus during last decades recommendation systems became an integral tool of OTT platforms and a required condition for customer loyalty.

Recommender aggregators are a subfamily of content filtering systems which provide the user with such content that he will be interested in, considering his behavior and preferences. The system predicts the reaction for one or another content part, suggests or excludes it.

The recommendation engine in the Telebreeze end-user app is a part of the integrated system that accumulates statistical information about viewed content. Based on that, the system offers another videos and audios that fit for the client.

The goal for such engine or system is a recommendation of content parts, which the user didn't know about before, but relevant in the existing conditions. It optimizes searching of new content and increases the time of user interaction with the service.

Before the integration of recommendation engine in our app, operators displayed content in the recommendations by themselves, depending on library updates, personal preferences or wish to pay users' attention on certain content.

The application constantly keeps statistics of the viewed content. Based on general statistics for users and for each individual client (preferred genres, number and duration of viewing a particular movie, rating), the aggregator offers new elements of library: video and audio which may interest the user.

For example, if the user viewed several movies with Jackie Chan the statistics will collect the information that this user prefers "action movies", and "movies with Jackie Chan". The system will offer similar films in the recommendations.

The magic of prediction is based on software which calculates recommendations by several methods at one time, and produces consolidated result.

There are several filtering methods used in our system: collaborative filtering, content filtering, and a hybrid method.

• Content recommendation method: the content offered to the user is similar to the content he viewed before;
• Collaborative recommendation method: the content offered to the user is similar to the content that was chosen by another user with similar preferences;
• Hybrid recommendation method: combines collaborative and content methods.

The results of implementation of the recommendation service:

• customer retention,
  
• attracting new users,
  
• ARPU growth.
  

The recommendation system offers TV-shows and movies to the viewer, similar to the content he viewed before. Thus more content is consumed, that leads to an increase of the number of advertisements showed (in advertising business model) or an increase in sales of paid content.

It is important for the end-user not to think about what to watch next. The recommendation engine optimizes the process of content consuming, allows viewer to save time on searching for interesting content without switching to third-party resources with movie ratings and descriptions.