Knowledge-Aware Recommendations With Objective Natural-Language Explanations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional recommendation systems rely on user purchase history and item reviews, which often provide subjective, broad, and irrelevant explanations, fail to account for new items, and cannot justify recommendations effectively.
Innovation Solution
A knowledge-aware explainable recommendation system using natural language processing (NLP) and knowledge graphs (KG) to generate concise, objective explanations by representing queries and items as a KG, converting it into a narrative to explain why an item is recommended.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional collaborative filtering algorithms are used to generate recommendations based on purchase history, then recommendations can be provided efficiently, but the explanations become subjective, broad, and irrelevant
Solution Approach 1:
The patent introduces a knowledge graph as an intermediary structure between user purchase history and recommendation explanations. The knowledge graph serves as a mediator that organizes item attributes, relationships, and contextual information, enabling the system to generate both efficient recommendations and detailed, objective explanations by querying structured knowledge rather than relying solely on collaborative filtering patterns.
Solution Approach 2:
The patent segments the explanation generation process into multiple components: identifying key item attributes from the knowledge graph, selecting relevant relationships, and formulating explanations based on specific user preferences and item characteristics. This segmentation allows the system to provide targeted, relevant explanations for each recommendation rather than generating generic broad explanations.
2Productivity
If recommendations are based on similar user purchase history, then user engagement can be maximized, but the system fails to account for new items without sufficient explanation
Solution Approach 1:
The patent implements preliminary action by pre-building and maintaining a comprehensive knowledge graph that includes new items before they are recommended. The knowledge graph is continuously updated with item attributes and relationships, allowing the system to immediately provide explanations for new items when they appear, without requiring retraining or adaptation of the collaborative filtering model.
Solution Approach 2:
The knowledge graph acts as an intermediary that bridges traditional collaborative filtering with new item handling. It allows the system to leverage existing user preferences and purchase patterns while simultaneously incorporating new items with their structured attributes, enabling the system to explain recommendations for new items based on their knowledge graph representations rather than requiring historical purchase data.
3Device complexity
If traditional recommendation systems are used, then the system structure remains simple, but the explanations lack objectivity and are difficult to justify
Solution Approach 1:
The patent segments the recommendation system into distinct functional modules: a collaborative filtering component for generating recommendations based on user behavior, and a knowledge graph component for providing objective explanations. This segmentation allows each module to specialize in its function, with the knowledge graph module independently maintaining structured, objective information about items and their relationships, thereby improving explanation reliability without requiring complete system redesign.
Data Source
AI summary
A method includes receiving query data, receiving item data, initializing the query data as at least one natural language query token, and initializing the item data as at least one natural language item token. The method also includes generating a knowledge graph for the item based on the at least one natural language item token, flattening the knowledge graph for the item to generate a knowledge graph string, mapping at least one token associated with the knowledge graph string and the at least one natural language query token to an embedding vector using a matrix of parameters, and providing, to a machine learning model, the embedding vector. The method also includes receiving, from the machine learning model, a recommendation and a natural language explanation of the recommendation, and providing, to a user at a display, the recommendation and the natural language explanation of the recommendation.