Multimodal Knowledge Graph Zero-Shot Product Recommendations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Standard recommendation systems require training on new products before making recommendations, limiting their ability to suggest products not included in the training dataset and failing to provide recommendations for new entities.
Innovation Solution
A multi-modal knowledge graph representation system that uses zero-shot learning and logical reasoning to generate recommendations, allowing for the inclusion of new entities and attributes without retraining, by combining various data modalities such as images, text, and user profiles to infer relationships and make product suggestions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If standard recommendation systems use collaborative filtering or matrix factorization algorithms trained on existing product ratings, then recommendation accuracy for trained products is improved, but the system cannot provide recommendations for new products or entities that were not included during training
Solution Approach 1:
The patent introduces a knowledge graph as an intermediary structure that connects products, users, and attributes through semantic relationships. This knowledge graph enables the system to reason about new products by leveraging relationships with existing products and attributes, without requiring direct training data for the new products. The knowledge graph acts as a mediator between the trained model and new entities, allowing zero-shot recommendations.
Solution Approach 2:
The system performs preliminary construction of a comprehensive knowledge graph that includes product attributes, relationships, and semantic information before new products arrive. By pre-establishing the knowledge structure and relationships, the system is prepared to immediately reason about and recommend new products as soon as they are added to the knowledge graph, without requiring retraining or preliminary training data collection.
2Ease of operation
If standard recommendation systems profile users based on purchase history and product ratings, then personalized recommendations are improved, but the system requires extensive training data for each user
Solution Approach 1:
The knowledge graph serves as an intermediary that stores structured user profiles, product attributes, and relationships. Instead of requiring extensive training data for each user, the system leverages the pre-built knowledge graph to infer user preferences and product characteristics through logical reasoning over relationships, reducing the dependency on large quantities of user-specific training data.
Solution Approach 2:
The knowledge graph provides a universal structure that can represent any user, product, or attribute without requiring user-specific training. The same knowledge graph infrastructure serves all users and products, enabling personalized recommendations through reasoning over shared knowledge rather than through user-specific training data collection.
3Reliability
If standard recommendation systems require training on new products before making recommendations, then recommendation reliability is improved, but system productivity and response time deteriorate due to retraining requirements
Solution Approach 1:
The system performs preliminary construction of the knowledge graph with comprehensive product attributes, relationships, and semantic information before new products arrive. This pre-established knowledge structure allows the system to immediately reason about new products upon their addition, eliminating the need for time-consuming retraining processes while maintaining reliable recommendations through logical inference.
Solution Approach 2:
The patent replaces the mechanical retraining process with a knowledge-based reasoning system. Instead of mechanically retraining models on new product data, the system uses logical reasoning over the knowledge graph to infer recommendations for new products, significantly reducing processing time while maintaining reliability through semantic relationships and zero-shot learning capabilities.
Data Source
AI summary
A method for generating a product recommendation in a retail system includes collecting a dataset containing a plurality of entities and attributes for the entities. Relationships between the plurality of entities are generated. The plurality of entities, attributes and relationships are stored in a knowledge graph. A representation of the plurality of entities, attributes and relationships stored in the knowledge graph is learned. Zero-shot learning is performed for a new entity and attributes for the new entity. The new entity and attributes for the new entity are stored in the knowledge graph. A recommendation for a user is generated based on the knowledge graph.