Skip to content

Recommendation Service Overview#

1. Overview#

The Recommender Service provides passengers with personalized content suggestions during their flight. These recommendations are based on individual preferences, popular trends, and viewing history. The goal is to enhance the in-flight entertainment experience by suggesting relevant and engaging media options.

Note: The recommendation APIs currently exposed through the CA SDK correspond to the “1.0 Recommendation Engine” available on the IFE platform. Any newer recommendation solutions will be delivered via separate REST APIs and are not currently supported in the Companion App SDK.

Passenger data plays a big role in suggesting content during flights. Here's a simplified explanation of how it works:

  1. Passenger Data During the Flight:

    • As passengers watch movies or shows, their viewing history is collected. This helps create lists of what's popular or trending during that specific flight. For example, if a lot of people are watching a particular movie for a long time, it will be considered trending. This list updates as more passengers watch content.
  2. Passenger Data from Previous Flights:

    • After the flight, data from passengers' activities is stored in two ways:

      • By seat number (Paxus)

      • By a unique passenger ID (IPS)

    • Paxus data helps create lists of popular content across all flights by a specific airline. This helps determine what's popular across their fleet.

    • IPS data is used to personalize recommendations. It remembers what passengers watched before and tailors suggestions based on their past preferences. It also avoids suggesting content they've already watched and disliked.

  3. Types of Recommendations and Where They Appear:

    • Recommendations can show up in different parts of the entertainment system based on what passengers are doing. For instance:

      • Similar media might be suggested after someone finishes watching a movie.

      • Trending content might be highlighted during the flight to show what's popular.

      • Recommendations might be personalized based on past behavior when a passenger logs in.

  4. Filtering Recommendations:

    • The system can adjust recommendations based on various factors, like the type of content passengers prefer, what they've disliked in the past, or how many recommendations they want to see.

In essence, passenger data helps create a customized and engaging entertainment experience during flights by suggesting content tailored to individual preferences and what's popular at the moment.

2. Definition and Role of Recommender Systems#

Recommender systems help personalize content for passengers by suggesting media based on their preferences and behaviors. There are different types of recommender systems, including:

  • Collaborative Filtering: This technique is based on user behavior patterns. If two users share similar viewing habits, they are likely to enjoy similar content. For example, if Passenger A watches a movie, Passenger B with similar viewing habits may receive a recommendation for the same or similar movies.

  • Content-Based Filtering: This approach uses item metadata to recommend similar items. Characteristics such as genre, cast, director, themes, and keywords influence the recommendations. Each characteristic has a weight that determines its impact on similarity scoring.

3. How Recommenders Work#

The recommender service generates a ranked list of media recommendations, each assigned a score. A higher score indicates a more relevant suggestion. Various algorithms and available passenger data determine these scores. The recommendations can also be filtered based on parameters like content type, blacklisted media, user preferences, and the number of results needed.

3.1. Collection and Processing of Passenger Data#

Passenger data is essential for generating accurate recommendations. This data can be categorized into:

  1. Passenger Data During the Flight: Viewing history is gathered in real time to track trending and popular content based on what passengers are watching. The system updates the list dynamically, ranking content based on total minutes watched.

  2. Passenger Data from Previous Flights: Passenger usage data is offloaded at the end of a flight through:

    • Paxus (Seat-based data tracking)

    • IPS (Passenger-based tracking via airline ID)

IPS data helps personalize recommendations by ensuring previously watched content contributes to future suggestions while avoiding redundant recommendations. It also considers past playlists, bookmarks, and preferences.

4. Types of Recommendations#

4.1. Similar Media Recommendations#

This feature suggests media that is similar to a specific movie or TV show based on metadata and content similarity.

  • How it works:

    • Passengers receive suggestions based on the media they are currently watching or have watched before.

    • Uses metadata from Rovi(metadata that enhances recommendation systems) to find similarities based on characteristics such as genre, cast, and themes.

  • Example Use Cases:

    • "Recommended For You" based on a specific movie.

    • "Because You Watched" suggestions.

    • "Similar to This Movie" displayed on movie details pages.

4.2. Personalized Recommendations#

This feature suggests content based on a passenger’s viewing history and playlists.

  • How it works:

    • The system collects viewing data from passenger interactions.

    • It analyzes past behaviors to suggest content that aligns with the passenger’s preferences.

    • Uses passenger seat data and playback history to trigger recommendations.

  • Example Use Cases:

    • "Because You Watched" personalized suggestions.

    • "Based on Your Playlist."

    • "Recommended For You."

This feature highlights media that is currently popular among passengers on the flight.

  • How it works:

    • Tracks what other passengers are watching during the flight.

    • Data is reset based on airline-specific conditions (e.g., landing, flight closure, or door opening).

  • Example Use Cases:

    • "What Is Popular" among passengers.

    • "What Is Trending" during the flight.

    • "What Other Passengers Are Watching."

4.4 Combined Recommendations (Ensemble Mode)#

This feature merges similar, personalized, and trending recommendations into a single optimized list.

  • How it works:

    • Prevents duplicate content from appearing in multiple categories.

    • Improves performance by reducing multiple data requests.

    • Uses multiple datasets, including:

      • Popular Data from Rovi (box office revenue-based ranking)

      • Recommendation Data from IPS (based on passenger preferences and seatback activity)

      • Trending Data from Paxus (real-time flight-wide trends)

  • Example Use Cases:

    • A combined list showing personalized, trending, and popular content together.

    • Optimized suggestions balancing different recommendation types.

Recommendation Service Overview

5. Filtering and Customization#

The recommender service allows for customization using filtering logic, which lets clients specify query parameters to refine recommendations. Common filters include:

  • Content type (movies, TV shows, audio, etc.)

  • Blacklist media items (to exclude specific content)

  • Passenger preferences

  • Limit (number of items to return)

6. Data Dependencies#

The recommendation system relies on multiple data sources:

  • Passenger Profile & Viewing History – Used for personalization.

  • Media Metadata – Used for finding similar content.

  • In-Flight Passenger Viewing Trends – Used for identifying popular content.

  • Seat and Playback Data – Required for personalized recommendations.

7. Configuration & Support#

  • The recommendation engine depends on media metadata from the MMA Media Profile.

Version 1.0