Knowledge Graph Extrapolation via Multi-Layer Perception
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Temporal knowledge graphs are limited by the closed world assumption and incomplete historical data, leading to inaccurate predictions, especially when dealing with future events or never-appeared entities and relations, resulting in reduced reasoning accuracy and efficiency.
Innovation Solution
A knowledge-graph extrapolating method based on multi-layer perception, specifically the TMP-Net model, which learns continuous representations of entities, relations, and timestamps to capture long-term dependencies, and classifies prediction tasks into different reasoning scenes to handle never-appeared entities and relations, using relational graph convolutional networks and emerging task processing units to achieve accurate predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If temporal knowledge graphs use closed world assumption and existing facts only, then reasoning accuracy for historical events is improved, but adaptability to future events and never-appeared entities is reduced
Solution Approach 1:
The model dynamically adapts its reasoning approach based on the prediction task type. It identifies whether entities or relations are new (never-appeared) and switches between different reasoning modes: using historical pattern matching for known entities and relational path mining for new entities, thereby achieving both accuracy for historical events and adaptability for future events
Solution Approach 2:
The model changes the parameters of entity representation and reasoning depth based on whether entities are new or existing. For never-appeared entities, it uses relational path information and multi-hop reasoning; for existing entities, it leverages historical temporal patterns. This parameter adaptation resolves the contradiction between accuracy and adaptability
2Measurement precision
If temporal knowledge graphs encode historical records separately for each search, then measurement precision of individual facts is improved, but productivity of the system is reduced
Solution Approach 1:
The model pre-computes and stores relational path information, entity embeddings, and temporal patterns during an offline preprocessing phase. This preliminary action enables the system to quickly retrieve and combine pre-computed information during online prediction, avoiding separate encoding of historical records for each search while maintaining high accuracy
Solution Approach 2:
The model transitions from encoding entire historical records sequentially to encoding relational paths and temporal patterns in parallel dimensions. By representing knowledge in terms of relational graphs and temporal embeddings, it enables efficient parallel computation and retrieval, improving productivity without sacrificing precision
3Adaptability or versatility
If temporal knowledge graphs consider never-appeared entities and relations, then adaptability to future events is improved, but device complexity increases
Solution Approach 1:
The model segments the reasoning process into distinct modules: entity identification module (detecting new vs. existing entities), relational path mining module (for new entities), and temporal pattern matching module (for existing entities). This segmentation allows the system to handle never-appeared entities without overwhelming complexity, as each module has a specific, simplified function
Solution Approach 2:
The model introduces relational path information as an intermediary representation that bridges the gap between known and unknown entities. By using relational paths as intermediate structures, the system can reason about never-appeared entities through their relationships with known entities, reducing the complexity of directly modeling entirely new concepts
Data Source
AI summary
The present invention relates to a knowledge-graph extrapolating method and system based on multi-layer perception, the method comprising: using relational graph convolutional network encoders to learn embedding representations, and capturing dynamic evolution of a fact; designing emerging task processing units to construct multiple layers of entity sets, and assigning a matching historical relevance; classifying prediction tasks into different reasoning scenes, and connecting them to the corresponding processing unit for partition of entity sets; and using a multi-class task solving method to acquire predicted probability distributions of target entities, and taking the highest one as a prediction answer, so as to accomplish extrapolation of a temporal knowledge graph, wherein the prediction tasks are classified into different reasoning scenes according to whether it contains any entity or relation that has never appeared historically. The knowledge-graph extrapolating system comprises a processor that can run program code information of the disclosed method.