Multimedia Recommendation Using Relationship Graph Features
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing multimedia resource recommendation algorithms lack accuracy and have a low matching degree, failing to effectively recommend multimedia content based on relationship types between target objects and their adjacent objects.
Innovation Solution
A multimedia resource recommendation method that utilizes an association information network graph to acquire representation vectors and relationship feature information, determining multimedia resources with a high matching degree based on interaction behaviors and relationship types between target and adjacent objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional recommendation algorithms are used, then the system is simple to implement, but the recommendation accuracy and matching degree are low
Solution Approach 1:
The patent segments the recommendation system into multiple components: constructing an association information network graph from social relationship data, extracting relationship feature information, building a graph convolutional network model with specific layers (embedding layer, graph convolutional layers, pooling layer, fully connected layer), and separately processing different relationship types. This segmentation allows the complex task of accurate recommendation to be broken down into manageable processing stages, improving accuracy while organizing system complexity systematically
Solution Approach 2:
The patent introduces an association information network graph as an intermediary structure that connects target objects with their adjacent objects through defined relationship types. This graph structure serves as a mediator between raw social data and recommendation outputs, enabling the system to capture and process relationship features systematically. The graph convolutional network then processes this intermediary representation to generate accurate recommendations
2Reliability
If relationship feature information is extracted using graph convolutional networks, then the matching degree improves, but the computational complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-construcing the association information network graph and pre-extracting relationship feature information before the actual recommendation process. The graph convolutional network model is pre-trained with specific configurations (number of layers, learning rate, batch size) to establish efficient processing patterns. This preliminary setup enables faster, more energy-efficient inference during actual recommendation operations while maintaining high matching degree
Solution Approach 2:
The patent implements dynamic adjustments in the graph convolutional network processing by adaptively selecting which relationship types to process based on their relevance to the target object. The system dynamically adjusts the scope of relationship feature extraction and processing intensity, allowing it to maintain high matching degree while optimizing computational energy consumption by focusing resources on the most relevant relationship features
Data Source
AI summary
A multimedia resource recommendation method and apparatus includes: acquiring a representation vector of a target object and an adjacent object representation vector set; determining, based on the representation vector of the target object and the adjacent object representation vector set, representation feature information of the target object, the representation feature information being determined based on relationship feature information corresponding to each first relationship type of K first relationship types of the target object and adjacent objects; and acquiring a multimedia resource set, determining, based on the representation feature information of the target object, a first multimedia resource from the multimedia resource set, and transmitting the first multimedia resource to the target object, the first multimedia resource being one or more multimedia resources having a matching degree with the representation feature information of the target object higher than a matching degree threshold.