Multipart Neural Network for Question Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional question answering systems for autonomous machines, such as humanoid robots, rely heavily on manually labeled examples and manual coding of question classification rules, which are scarce, expensive, and inefficient, and are primarily designed for factual questions, failing to effectively handle a variety of question types including situations, observations, and commands.
Innovation Solution
A system utilizing a multipart artificial neural network (ANN) that includes an auxiliary ANN trained unsupervisedly with unlabeled data to generate labeled data and a main ANN trained with labeled data, allowing for efficient classification and mapping of questions to appropriate knowledge sources without requiring extensive manual labeling, using inductive transfer to improve training efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manually labeled examples and manual coding of question classification rules are used, then question classification accuracy can be achieved, but the system becomes expensive and inefficient due to the scarcity and high cost of manual labeling
Solution Approach 1:
The system performs preliminary unsupervised learning on unlabeled data to extract features and patterns before the main supervised classification task. This preliminary action prepares the system to handle diverse question types more effectively, reducing the need for extensive manual labeling while maintaining classification accuracy.
Solution Approach 2:
The patent introduces an intermediary component that bridges unlabeled and labeled data. The system uses unlabeled data as an intermediary resource to pre-train or fine-tune classification models, which then process labeled data more efficiently. This intermediary step leverages the abundance of unlabeled data to improve overall system performance without requiring proportional increases in manual labeling effort.
2Adaptability or versatility
If conventional question answering systems are designed for factual questions, then they can handle simple query types, but they fail to effectively handle a variety of question types including situations, observations, and commands
Solution Approach 1:
The patent implements a universal question classification system that handles multiple question types (factual, situational, observational, command-based) through a single unified framework. The system uses general-purpose natural language processing techniques and learned representations that adapt to different question types without requiring separate specialized modules, thereby achieving versatility without proportionally increasing complexity.
Solution Approach 2:
The system dynamically adjusts classification parameters and thresholds based on the detected question type. By changing parameters such as classification thresholds, feature weights, and decision boundaries according to the specific question category, the system effectively handles diverse question types while maintaining a relatively simple underlying architecture.
3Measurement precision
If extensive manual labeling is performed to train question classification systems, then training accuracy can be improved, but training time and resource requirements increase significantly
Solution Approach 1:
The system applies partial supervised learning by using a smaller, strategically selected subset of labeled data combined with extensive unsupervised learning on unlabeled data. This approach achieves comparable or superior classification accuracy while significantly reducing the time and resources required for manual labeling and training, as the unlabeled data provides additional learning signals that compensate for the reduced labeled data volume.
Data Source
AI summary
A system and a method are disclosed for automatic question classification and answering. A multipart artificial neural network (ANN) comprising a main ANN and an auxiliary ANN classifies a received question according to one of a plurality of defined categories. Unlabeled data is received from a source, such as a plurality of human volunteers. The unlabeled data comprises additional questions that might be asked of an autonomous machine such as a humanoid robot, and is used to train the auxiliary ANN in an unsupervised mode. The unsupervised training can comprise multiple auxiliary tasks that generate labeled data from the unlabeled data, thereby learning an underlying structure. Once the auxiliary ANN has trained, the weights are frozen and transferred to the main ANN. The main ANN can then be trained using labeled questions. The original question to be answered is applied to the trained main ANN, which assigns one of the defined categories. The assigned category is used to map the original question to a database that most likely contains the appropriate answer. An object and/or a property within the original question can be identified and used to formulate a query, using, for example, system query language (SQL), to search for the answer within the chosen database. The invention makes efficient use of available information, and improves training time and error rate relative to use of single part ANNs.