Bidirectional Query Encoding for Missing Knowledge Graph Entities
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 entire query context and often result in inferior performance when decomposing queries into subqueries.
Innovation Solution
A method and system that encode query graphs into sequence representations using a bidirectional transformer network, incorporating a novel positional encoding scheme to handle queries with multiple missing entities, allowing the transformer to reason over the entire query context.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If existing AI methods decompose complex queries into subqueries, then the query processing becomes manageable, but the performance deteriorates due to loss of overall query context
Solution Approach 1:
The query graph is segmented into multiple path queries from root nodes to leaf nodes, which are then processed independently and aggregated. This segmentation allows complex queries to be broken down into manageable components while the bidirectional transformer ensures context is preserved through attention mechanisms that consider all paths simultaneously.
Solution Approach 2:
The solution nests multiple path queries within a unified bidirectional transformer framework. Each path query is encoded independently but then integrated through the bidirectional attention mechanism, creating a nested structure where individual path analyses are contained within the broader context of the entire query graph.
2Loss of information
If bidirectional transformer encoding is used for all path queries, then query context understanding improves, but computational complexity increases
Solution Approach 1:
The computational burden is segmented by processing each path query independently through the bidirectional transformer, then aggregating results. This avoids the need to process the entire query graph as a single unit, reducing computational complexity while maintaining context through the attention mechanism.
Solution Approach 2:
The bidirectional transformer is applied to each individual path query rather than the entire query graph at once. This partial application of the complex encoding mechanism reduces overall computational requirements while still capturing essential contextual information from each path.
3Manufacturing precision
If positional encoding is applied to each path query independently, then query structure is preserved, but processing time increases
Solution Approach 1:
Positional encoding is applied independently to each segmented path query, preserving the structural information of individual paths. The segmentation allows parallel processing of multiple paths, which offsets the time cost of applying positional encoding to each path separately.
Solution Approach 2:
Positional encoding is applied as a preliminary step to each path query before the main bidirectional transformer processing. This preliminary structuring of positional information enables more efficient subsequent processing by pre-organizing the input data in a format that the attention mechanism can process effectively.
Data Source
AI summary
A method for encoding a query graph into a sequence representation includes receiving m×n path queries representing a query graph having m root nodes and n leaf nodes, each path query beginning with a root node and ending with a leaf node, and encoding positions of each node and each edge between nodes within each path query independently, wherein the encoded positions include a positional order within each path query. The positional encodings may include no positional ordering between different path queries. The query graph may include one or more missing entities and each missing entity may be masked to produce a masked sequential query, which may be fed to a transformer encoder. The invention 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, and significantly outperforms the optimized model.