Bidirectional Query Graph Encoding for Missing Entity Reasoning
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 adequately handle conjunctive queries with free variables.
Innovation Solution
A method using bidirectional transformer networks to encode query graphs into sequences, incorporating positional information and masking missing entities, allowing the transformer to answer conjunctive queries by decomposing the graph into path queries and using self-attention mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If feed forward neural network is used for set intersection in vector space, then simple path queries can be answered, but complex queries with multiple missing entities cannot be effectively handled
Solution Approach 1:
The query graph is decomposed into multiple path queries, each representing a specific traversal path from root nodes to leaf nodes. This segmentation allows the transformer encoder to process each path independently while maintaining the ability to handle complex queries with multiple missing entities through the collective analysis of all paths.
Solution Approach 2:
Positional encoding is introduced to add dimensional information to the node representations. By encoding positions of nodes and edges with positional order and counter values, the system gains the ability to distinguish between different structural contexts in the query graph, enabling accurate handling of complex queries that previous methods could not address.
2Measurement precision
If positional encoding is incorporated into transformer encoder, then accurate positional information is captured, but computational complexity increases
Solution Approach 1:
Positional encoding is applied locally to each node and edge within the query graph context. Each node receives positional information relative to its specific position in the path query, rather than applying a global complex transformation. This localized approach captures precise positional information while keeping the overall system complexity manageable.
3Extent of automation
If masking is applied to missing entities, then the transformer can learn to predict missing values, but training complexity increases
Solution Approach 1:
Masking of missing entities is performed as a preliminary step before feeding the query to the transformer encoder. By pre-processing the query graph to identify and mask missing nodes and edges, the system enables the transformer to focus on predicting these masked values during training, automating the missing entity prediction process while structuring the training workflow in a manageable sequence.
Data Source
AI summary
A method of encoding a query graph having m root nodes and n leaf nodes includes receiving m×n path queries representing the query graph, where each path query starts with a root node and ends with a leaf node, and where the query graph includes one or more missing nodes or edges between nodes. Positions of each node and edge within each path query are encoded independently, where the encoded positions include a positional order, and positional counter values that are reset at a start position of each path query. The missing nodes or edges are masked to produce a masked query that is fed to a transformer encoder. The method 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.