Bidirectional Sequence Encoders for Missing-Entity Knowledge Graph Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing AI methods struggle to effectively answer complex queries in knowledge graphs, particularly those with multiple missing entities, as they fail to consider the post-intersection target query and do not account for the entire query context.
Innovation Solution
A novel method using bidirectional transformer networks to encode query graphs into sequences, incorporating a positional encoding scheme that allows transformers to handle conjunctive queries with missing entities by decomposing the query graph into path queries and masking missing elements, enabling self-attention mechanisms to predict entities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If feed forward neural network performs set intersection in vector space, then simple path queries can be answered, but complex queries with multiple missing entities cannot be answered accurately
Solution Approach 1:
The query graph is decomposed into multiple path queries between root nodes and leaf nodes. Each path query is processed independently to generate intermediate results, which are then combined to answer the complex query. This segmentation allows the system to handle complex queries with multiple missing entities by breaking them down into manageable components.
Solution Approach 2:
Positional encodings are introduced as intermediary elements that carry position information through the transformer network. These encodings enable the model to understand the structural relationships and directional information in the query graph, allowing accurate handling of complex queries that require understanding of query context and path relationships.
2Loss of information
If positional encodings include counter values reset at start position of each path query, then path-specific position information is preserved, but computational complexity increases
Solution Approach 1:
Positional encodings are pre-computed and embedded into the query representation before being processed by the transformer network. The counter values are reset at the start position of each path query in advance, allowing the model to receive position-aware inputs without performing complex runtime calculations. This preliminary preparation reduces computational complexity during query processing.
3Measurement precision
If bidirectional transformer network is used to encode entire query context, then complex queries with multiple missing entities can be answered accurately, but computational resources and processing time increase
Solution Approach 1:
The bidirectional transformer network processes segmented path queries rather than the entire complex query at once. By dividing the query into multiple smaller path components and processing them separately, the computational burden is reduced while still capturing the full query context through the combination of path-level representations. This segmentation approach maintains accuracy while reducing processing time.
Solution Approach 2:
The system processes only the necessary path queries that are relevant to answering the complex query, rather than processing all possible queries in the knowledge graph. This partial action approach focuses computational resources on the specific paths needed to answer the query, reducing overall processing time while maintaining answer accuracy.
Data Source
AI summary
A memory for storing data for access by an application program includes an encoded sequential query representing m×n path queries of a query graph having elements including m root nodes and n leaf nodes, and a missing element, wherein m and n are integer values. Each of the path queries includes a root node as a start position and a leaf node as an end position, wherein positional encodings of the elements include a positional order within each path query, wherein the positional encodings of the elements include counter values that are reset at a start position of each of the path queries, and wherein the missing element is masked. The memory can be used for, but is not limited to, medical uses, such as for predicting protein-drug interactions or medical conditions, as well as other applications, to support decision making or optimizing processes or predictions.