Procedural Instruction Flow Graphs for Entity Substitution
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Determining viable substitutions in procedural instructions, such as recipes, is challenging due to the need for comprehensive understanding of entities and actions involved, which is difficult to achieve with unstructured natural language text.
Innovation Solution
A method involving natural language processing to generate a directed flow graph from procedural instructions, using ontologies for domain-specific knowledge, and applying knowledge graph embeddings to identify substitutions by calculating embeddings for entities and actions, allowing for plausible modifications.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If natural language text is used to describe procedural instructions, then the instructions are easy to write and understand, but it becomes difficult to extract entities and determine substitutions
Solution Approach 1:
The patent introduces an intermediary representation layer (structured data format with entities, actions, and parameters) between the natural language instructions and the substitution determination process. This intermediary structure captures the semantic meaning of instructions while organizing information in a machine-processable format, enabling entity extraction and substitution analysis without losing the natural language's readability benefits
Solution Approach 2:
The patent replaces manual entity extraction and substitution determination (mechanical process) with automated natural language processing and machine learning models. The system automatically parses instructions, identifies entities and their roles, and determines viable substitutions using trained algorithms, eliminating the need for manual text analysis
2Measurement precision
If comprehensive understanding of entities and actions is achieved, then substitution accuracy improves, but the complexity of processing increases
Solution Approach 1:
The patent segments the complex task of substitution determination into distinct processing stages: instruction parsing, entity identification, action recognition, parameter extraction, and substitution evaluation. Each stage handles a specific aspect of the problem, reducing overall complexity while maintaining comprehensive understanding through systematic breakdown of the processing pipeline
Solution Approach 2:
The patent develops a universal processing framework that handles multiple types of entities (ingredients, equipment, tools), actions (cooking methods, preparation steps), and parameters (temperature, time, quantity) within a single integrated system. This multi-functional approach achieves comprehensive understanding without proportionally increasing complexity by reusing common processing components across different entity and action types
Data Source
AI summary
A computer-implemented method for identifying substitutions for procedural instructions include parsing a procedural instruction to identify object-action tuples pertaining to the steps of the instruction, and generating a flow graph for the procedural instruction using the identified object-action tuples. The flow graph is then processed using a knowledge graph embedding model to calculate embeddings for the intermediate nodes and the result node, respectively, of the flow graph based on triples that include a head entity, a relation and a tail entity, wherein the head entity for a given intermediate nodes or the result node comprises an aggregation of the input nodes connected to the given intermediate node or the result node. A substitution for nodes in the flow graph may then be identified by calculating new embeddings for the result node based on the substitution and determining a similar between the original embeddings and the new embeddings.