Natural Language Querying of Data Graphs via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Creating large data graphs is challenging due to the manual process of adding entities and relationships, leading to missing data that reduces the graph's usefulness for querying.
Innovation Solution
A machine learning algorithm is trained to extract syntactic and semantic knowledge from text and combine it with data graph information to predict additional tuples, expanding the graph and allowing for natural language queries by mapping features to query results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If manual processes are used to add entities and relationships to a data graph, then data accuracy can be maintained, but the productivity of creating large data graphs is severely limited and many facts remain missing
Solution Approach 1:
The patent replaces the manual mechanical process of adding entities and relationships with an automated machine learning system. The ML algorithm automatically extracts entities, relationships, and facts from unstructured text data and populates the knowledge graph without human intervention, dramatically increasing both productivity and the quantity of data that can be added to the graph
Solution Approach 2:
The system enables the data graph to self-populate by automatically ingesting data from external sources such as Wikipedia and other text corpora. The machine learning model autonomously processes incoming data, identifies relevant entities and relationships, and adds them to the graph without requiring manual curation, allowing the graph to grow and update itself continuously
2Adaptability or versatility
If more entities and relationships are added to expand the data graph, then the usefulness for querying improves, but the device complexity of managing and maintaining the graph increases
Solution Approach 1:
The patent implements a universal machine learning framework that handles multiple tasks simultaneously: entity extraction, relationship identification, fact verification, and graph population. This multi-functional system manages the complexity of expanding the graph by consolidating multiple processing functions into a single automated pipeline, making it easier to maintain and update the graph as it grows
Solution Approach 2:
The machine learning model acts as an intermediary layer between raw text data and the knowledge graph structure. It translates unstructured text into structured graph elements (entities, relationships, tuples), managing the complexity of data integration and ensuring consistent formatting and validation as the graph expands
3Reliability
If manual verification is performed on predicted tuples before adding to the data graph, then data reliability improves, but the time required for graph creation increases significantly
Solution Approach 1:
The system applies partial verification by automatically adding only those predicted tuples that meet a confidence threshold to the knowledge graph, while flagging lower-confidence predictions for manual review. This selective approach maintains high reliability for automatically added data while minimizing the time investment required for manual verification of only the uncertain cases
Solution Approach 2:
The system implements feedback mechanisms where manually verified tuples are used to retrain and improve the machine learning model's accuracy over time. This continuous feedback loop increases the reliability of automated predictions, reducing the proportion of tuples requiring manual verification and thereby decreasing the time loss associated with verification
Data Source
AI summary
Implementations include systems and methods for querying a data graph. An example method includes receiving a machine learning module trained to produce a model with multiple features for a query, each feature representing a path in a data graph. The method also includes receiving a search query that includes a first search term, mapping the search query to the query, and mapping the first search term to a first entity in the data graph. The method may also include identifying a second entity in the data graph using the first entity and at least one of the multiple weighted features, and providing information relating to the second entity in a response to the search query. Some implementations may also include training the machine learning module by, for example, generating positive and negative training examples from an answer to a query.