User Portrait Vector Mapping for Cold-Start Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The accuracy of user portraits for 'cold-start' users is low due to limited user behavior data, leading to inaccurate content recommendations.
Innovation Solution
A method that determines a user feature vector based on attribute information and historical behavior data, and uses level embedding processing to match alternative tags from multimedia content, improving the representation of user preferences and expanding the user portrait's generalization capability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If simple statistical processing is performed on portrait tags based on user behavior data, then the user portrait construction process is simple and fast, but the accuracy of user portrait is low for cold-start users due to limited behavior data
Solution Approach 1:
The patent introduces a pre-trained user portrait model as an intermediary that maps user features to portrait tags. This model serves as a mediator between limited user behavior data and accurate portrait construction, enabling cold-start users to obtain accurate portraits through feature vector mapping without requiring extensive behavior data
Solution Approach 2:
The patent performs preliminary embedding processing on user features and portrait tags before actual portrait construction. By pre-processing features into embedding vectors and pre-training the model on abundant data, the system prepares the necessary structures in advance, allowing accurate portrait generation even for cold-start users with limited data
2Measurement precision
If portrait tags are scored according to frequency from user behavior data, then the method is simple to implement, but the precision of content recommendation is low due to insufficient user preference representation
Solution Approach 1:
The patent transforms discrete portrait tags into continuous embedding vectors and uses similarity calculations between feature vectors and tag vectors. This parameter transformation from categorical to continuous space enables more nuanced preference representation and more precise recommendation matching
Solution Approach 2:
The patent adds a new dimension by introducing embedding vector space for portrait tags. Instead of simple frequency counting in one dimension, the system operates in multi-dimensional vector space where similarity can be calculated, providing richer preference representation and more accurate recommendation matching
Data Source
AI summary
A method for obtaining a user portrait includes: obtaining a user feature vector of a target user and tag feature vectors of content tags of multimedia content in a target application, and determining an alternative tag of the target user according to similarities between the user feature vector and the tag feature vectors, to further determine a user portrait of the target user according to the alternative tag.