Video Ranking System Using Experimental Data for Revenue and Engagement
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Video sharing websites face challenges in maximizing revenue and user engagement due to the presentation of irrelevant or non-monetizable additional videos, which can lead to user dissatisfaction and reduced monetization opportunities.
Innovation Solution
A system and method for ranking additional videos based on experimental data, using a combination of relevance, monetization, and likelihood rankings sourced from user interaction data, to select and present videos that are both relevant and likely to generate revenue, incorporating machine learning techniques to optimize video selection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If additional videos are selected based solely on relevance to the original video, then user engagement is improved, but revenue generation deteriorates because relevant videos may not be monetizable
Solution Approach 1:
The system changes the selection parameters from single-criteria (relevance only) to multi-criteria optimization, incorporating monetization value and experimental probability data as additional dimensions for video selection
Solution Approach 2:
The video selection system dynamically adjusts its ranking algorithm based on real-time experimental data, adapting the weighting of relevance versus monetization factors according to observed user behavior patterns
2Loss of energy
If videos with high monetization value are prioritized, then revenue generation is improved, but user engagement deteriorates due to presentation of irrelevant content
Solution Approach 1:
The system transforms the selection criteria from monetization-only to a balanced multi-parameter approach, integrating user engagement metrics as a constraint or weighting factor in the optimization function
Solution Approach 2:
The system uses A/B testing and experimental data collection to continuously feedback on user engagement metrics, adjusting the monetization-relevance balance based on observed performance
3Device complexity
If a simple relevance-based selection algorithm is used, then device complexity is reduced, but productivity deteriorates due to suboptimal revenue maximization
Solution Approach 1:
The system implements self-optimization through automated A/B testing and experimental data collection, where the algorithm automatically learns and adapts to optimal video selections without manual intervention
Solution Approach 2:
The system performs preliminary experiments and data collection to establish probability models before full-scale deployment, pre-calculating optimal video selections based on historical performance data
4Measurement precision
If extensive experiments are conducted to gather user interaction data, then measurement precision is improved, but loss of time increases due to data collection requirements
Solution Approach 1:
The system implements a staged experimentation approach, starting with partial data collection from selected user groups and progressively expanding as confidence in the model increases, rather than requiring complete data sets
Solution Approach 2:
The system performs preliminary experiments with small sample sizes to establish initial probability models, then uses these models to make selections while continuing to gather data, rather than waiting for extensive data collection
Data Source
AI summary
A system and method for ranking additional videos based on experiment data. The system includes an additional video request retrieval unit to receive a request for a plurality of additional videos, and a number of the plurality of additional videos based on a video being served; an additional video database to retrieve the plurality of additional videos based on the video, and to retrieve a relevance ranking for each of the plurality of additional videos, a monetization ranking for each of the plurality of additional videos, and a likelihood ranking for each of the plurality of additional videos, the likelihood ranking being sourced from the experimental data; a weighting/ranking unit to assign a score for each of the plurality of videos based on the retrieved rankings from the additional video database; and an additional video selection unit to select the number of the plurality of videos based the score.