Video Recommendation Score Adjustment for Content Diversity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing video content recommendation systems lack diversity and flexibility, often leading to a high frequency of similar video types being recommended, which can reduce user engagement and affect overall user experience.
Innovation Solution
A video content recommendation method that involves selecting a video from a video set, modifying its recommendation score based on the presence and location of similar videos in the recommended sequence, and adjusting the score according to preset modification formulas and user attributes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the recommendation system continuously recommends videos of the same type based on user tapping behavior, then the recommendation accuracy for user preferences is improved, but the diversity of recommended content deteriorates
Solution Approach 1:
The patent applies preliminary anti-action by introducing a diversity constraint mechanism that proactively prevents excessive concentration of same-type videos in recommendations. Before the recommendation list is finalized, the system checks video type distribution and adjusts selection to ensure diversity, counteracting the natural tendency of the algorithm to over-recommend popular or frequently tapped video types.
Solution Approach 2:
The patent changes the recommendation parameters by introducing video type as a new dimension for scoring and selection. Instead of solely relying on user preference metrics, the system incorporates video type distribution parameters and adjusts recommendation scores based on the proportion of different video types already present in the recommendation list, thereby balancing accuracy and diversity.
2Device complexity
If the recommendation system uses a fixed scoring and sorting mechanism, then the implementation complexity is reduced, but the recommendation flexibility deteriorates
Solution Approach 1:
The patent applies dynamics by making the recommendation scoring mechanism adaptive rather than fixed. The system dynamically adjusts video scores based on real-time factors including user behavior, video attributes, video type distribution in the current batch, and diversity constraints. This allows the recommendation list to flexibly adapt to different user preferences and contextual conditions while maintaining a unified scoring framework.
Solution Approach 2:
The patent segments the recommendation process into distinct stages: initial scoring based on user preferences, diversity constraint application, and final sorting. By dividing the process into modular segments, the system achieves flexibility through multiple adjustment points while keeping each individual stage relatively simple and manageable.
3Productivity
If the system recommends a large quantity of videos of a specific type, then user engagement with preferred content is improved, but the overall user experience deteriorates due to lack of variety
Solution Approach 1:
The patent implements feedback mechanisms that monitor video type distribution in the recommendation list and use this information to adjust subsequent selections. The system continuously evaluates the proportion of different video types and provides feedback to the scoring mechanism, which then adjusts scores to maintain appropriate diversity while still prioritizing user-preferred content types within reasonable limits.
Data Source
AI summary
This application provides techniques of improving video recommendation. The techniques comprise temporarily pre-adding a first video to a recommendation video sequence; capturing a sub-sequence including the first video; modifying an initial recommendation score of the first video based on a position sequence number of the first video in the sub-sequence and a position sequence number of another video in the sub-sequence to obtain a modified recommendation score of the first video, the another video sharing a target attribute with the first video; and adding a video with a highest modified recommendation score to the recommendation video sequence.