Semi-Servised Question Answering Graph Propagation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing customer service user interface systems, such as IVR and chatbots, are rigid and require large amounts of pre-programmed data for supervised learning, limiting their ability to effectively answer natural language questions in real-time.
Innovation Solution
A semi-supervised machine learning method and system that creates a graph based on natural language text, including seen questions and answers, to propagate answer weights and determine the best response to new questions, allowing for learning from small data sets and improving question-answering capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If supervised learning approach is used with large amounts of questions and answers, then the system can provide accurate responses, but the data requirement and system complexity increase significantly
Solution Approach 1:
The patent applies partial action by using only a small subset of training data (seen questions and answers) rather than requiring complete supervised learning datasets. The graph propagation mechanism allows the system to derive answers for unseen questions by propagating information through the graph structure, achieving accurate responses with minimal training data.
Solution Approach 2:
The patent introduces a graph structure as an intermediary between training data and query responses. This graph serves as a mediator that propagates answer weights from seen questions to unseen questions through distance-based relationships, enabling the system to bridge the gap between limited training data and comprehensive question-answering capability.
2Stability of the object's composition
If rule-based systems with pre-programmed matching and response rules are used, then the system can provide consistent responses, but the system rigidity and inability to handle natural language variations increase
Solution Approach 1:
The patent transforms the static rule-based system into a dynamic graph-based system. Instead of fixed matching rules, the system dynamically computes distances between questions in the graph and propagates answer weights based on these distances. This dynamic approach allows the system to adapt to natural language variations while maintaining response consistency through the structured graph propagation mechanism.
Solution Approach 2:
The patent changes the fundamental parameter from discrete rule matching to continuous distance-based weight propagation. By representing questions as nodes in a graph and computing distances between them, the system transitions from rigid categorical matching to flexible continuous parameter-based matching, enabling natural language variation handling while preserving response consistency.
3Productivity
If IVR or chatbot systems are programmed with fixed states and flows, then the system can minimize staffing costs, but the system rigidity and inability to learn from new questions increase
Solution Approach 1:
The patent applies preliminary action by pre-computing the graph structure from a small set of seen questions and answers before deployment. This pre-computed graph serves as a learned knowledge base that enables the system to handle new questions without requiring real-time learning or additional programming, achieving both cost efficiency and learning capability.
Solution Approach 2:
The patent uses copying by propagating answer weights from seen questions to unseen questions through the graph structure. Instead of re-learning or re-programming for each new question, the system copies and adapts answers from similar questions based on graph distances, enabling flexible response generation with minimal additional resources.
Data Source
AI summary
Disclosed herein is a method for performing semi-supervised question answering. The method includes the steps of: (i) creating a graph based on input data, the input data comprising natural language text that includes seen questions, background text, and seen answers, and (ii) querying the graph in response to a new question. Querying the graph in response to a new question includes the steps of: projecting the question into the graph to a projected location; and for each node in a retrieval radius associated with the question, determining a set of answer weights based on the distances between retrieved nodes and the projected location and answer weights associated with retrieved nodes based on seen questions; and determining a best response corresponding to the seen answer associated with the highest answer weight from the set of answer weights determined for each retrieved node within the retrieval radius.