Short Videos Feed

Introduction

A short videos feed (TikTok / Instagram Reels / YouTube Shorts–style) is a continuously personalized stream of bite-sized videos, where each item is selected and ranked in real time to match the user’s immediate interests and engagement patterns.

The feed is consumed one video at a time, with each next recommendation optimized for relevance and engagement.

Sending Interactions

To achieve high-quality real-time personalization, it is essential to continuously send user interactions as they consume the feed.

• Send View Portions for consumed content, with the autoPresented parameter set to true and timeSpent specified.

  • Interactions should be sent periodically (e.g., every few seconds)
  • For short videos, the portion parameter can be set to 0 at the start of playback and to 1 afterwards

• Send likes and dislikes as +1 / -1 Ratings

These signals allow Recombee to quickly adapt the feed to the user’s current preferences.

Scenario Setup

In the Recombee Admin UI, navigate to the Scenarios section and choose Create Scenario.

1. Enter a Scenario ID, such as "short-videos-feed"
2. Choose Recommendation type - Items to User

Logic

The Logic defines the behavior of the recommendation engine.

• Choose the recombee:swiping-feed Logic type from the Video models

This Logic is designed for infinite swiping feed experiences, where users continuously browse short-form content and can express feedback through likes or dislikes.

Content Freshness

The Logic can prioritize newer content using the following parameter:

• creationTimestamp - property representing the creation time of an item, allowing the Logic to evaluate and boost more recent content

Optimization Weights

Recommendations are optimized in real time based on a configurable combination of user engagement signals, such as likes, number of viewed items, purchases, and time spent watching content.

You can control the importance of these signals using the following parameters:

• timeSpentWeight - weight controlling the importance of time spent watching content
• likeWeight - weight controlling the importance of user likes, sent as positive Ratings
• itemsShownWeight - weight controlling the importance of maximizing the number of items consumed by the user
• purchaseWeight - weight controlling the importance of Purchases

Filters

Requesting Recommendations

1. No-Code Widget

   • create an Infinite Feed No-Code Widget using the visual editor in the Admin UI.

2. Widget SDKs

   • use the SDK for a Feed to display the recommendations in the design of your choice.
   • use the Recommend Items to User endpoint when initializing the widget.

3. API SDK

   • request the Recommend Items to User endpoint of the Recombee API using an SDK of your choice.
   • as the user continues browsing, use Recommend Next Items to retrieve the next batch of content

For more information navigate to the Integration tab of the Scenario in the Admin UI.