Multi-Task Recommender Training via User-Item Graph Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing recommender systems often fail to provide accurate recommendations as they primarily utilize user activity data from a single domain, which may not fully depict user preferences or intentions.
Innovation Solution
A personalized multi-task training framework that generates comprehensive user/item embeddings by aggregating user activities from multiple domains, using an encoder to encode user-item interactions into embeddings, and applying these embeddings to various recommendation tasks with different loss functions and gradients.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If user activity data from a single domain is used to build preference profile, then the system complexity is low, but the recommendation accuracy is insufficient
Solution Approach 1:
The patent merges user activity data from multiple domains (e.g., e-commerce, social media, entertainment) into a unified preference profile. This combining of data sources enriches the user representation, enabling more accurate recommendations while managing complexity through integrated processing architecture.
Solution Approach 2:
The system implements a universal preference profile that serves multiple recommendation tasks across different domains. This multi-functional approach allows the same user profile to improve accuracy for various types of recommendations (products, content, services) without requiring separate single-domain profiles for each task.
2Measurement precision
If user activity data from multiple domains is aggregated, then the recommendation accuracy is improved, but the data processing complexity increases
Solution Approach 1:
The patent segments the multi-domain data processing into distinct modules: data collection from various domains, preprocessing-specific-to-each-domain, integration layer for unified representation, and recommendation generation. This segmentation manages complexity by organizing processing steps while maintaining comprehensive data aggregation for accurate preference profiles.
3Adaptability or versatility
If single-domain data is used for training, then the training process is simple, but the model generalization capability is limited
Solution Approach 1:
The training framework is designed to be universal across multiple domains, enabling the model to learn generalized user preferences that transfer across different recommendation tasks. The same training architecture processes data from various domains, improving model adaptability and versatility without requiring domain-specific training pipelines for each application.
Data Source
AI summary
Embodiments described herein provide a method for training a recommendation neural network model using multiple data sources. The method may include: receiving, via a data interface, time series data indicating a user-item interaction history; transforming the time series data into a user-item graph; encoding, by a neural network encoder, the user-item graph into user embeddings and item embeddings; generating a plurality of losses according to a plurality of training tasks performed based on the user embeddings and, item embeddings; training the recommendation neural network model by updating the user embeddings and the item embeddings via backpropagation based on a weighted sum of gradients of the plurality of losses; and generating, by a neural network decoder, one or more recommended items for a given user based on the updated user embeddings and the updated item embeddings.