Video Recommendation Vector Index with Multi-Target Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The dual-tower structure in video recommendation systems lacks cross-learning between user and content features, leading to inefficient calculations and reduced accuracy due to excessive parameter scales and inability to handle multiple target vectors simultaneously.
Innovation Solution
A method that involves obtaining an object feature vector, performing vectorization and multi-target feature learning to generate an object multi-target vector, which is used to recommend videos by concatenating historical playback sequences and video indices, reducing calculation complexity and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the double-enhanced dual-tower structure is used to generate enhanced vectors for fitting information of the other tower, then cross-learning between user and content features is enabled, but the parameter scale becomes excessively large resulting in high calculation delay
Solution Approach 1:
The patent extracts only the necessary cross-learning information by generating enhanced vectors only for the content tower using user tower information, rather than generating enhanced vectors for both towers. This selective extraction reduces the parameter scale while maintaining the essential cross-learning capability between user and content features.
Solution Approach 2:
The patent applies partial action by implementing enhanced vector generation only where necessary (content tower) rather than universally for both towers. This partial implementation achieves sufficient cross-learning effects without the excessive computational cost of full double-enhanced structure.
2Reliability
If the enhanced vector is continuously updated using target vector of the other tower, then cross-features are improved, but the tower structure does not support multiple targets and multiple target vectors cannot be fitted at the same time
Solution Approach 1:
The patent implements multi-functionality by designing the content tower to serve multiple purposes: it processes content features, generates enhanced vectors using user tower information for cross-learning, and simultaneously supports multiple target vectors for different recommendation objectives. This universal design enables the tower to handle both cross-features and multi-target fitting without requiring separate structures.
3Ease of manufacture
If independent dual towers are used for offline training and fast online retrieval, then training convenience and retrieval speed are improved, but model effect is limited due to lack of cross-learning opportunity
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing enhanced vectors for the content tower during offline training using user tower information. This preliminary computation enables fast online retrieval while incorporating cross-learning effects, as the enhanced vectors are ready in advance without requiring complex real-time calculations during service.
Data Source
AI summary
A video recommendation method includes: obtaining an object feature vector of a target object, a historical playback sequence of the target object in a preset historical time period, and a video multi-target vector index of each video in a video library; and performing vectorization processing on the historical playback sequence to obtain an object enhanced vector of the target object; sequentially performing vector concatenation processing and multi-target feature learning on the object feature vector and the object enhanced vector to obtain an object multi-target vector of the target object; and determining, from the video library based on the object multi-target vector and the video multi-target vector index of each video, a target recommended video corresponding to the target object.