Prototype Intent Classifier for Low-Retraining Message Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated communication systems face inefficiencies in determining user intents, requiring significant resources to retrain models for new intents and lacking flexibility in handling unmatched messages, with existing classifiers being computationally intensive and prone to overfitting.
Innovation Solution
The use of prototype vectors and a prototype classifier that allows for efficient addition of new intents without retraining the model, using word and message embeddings to compare messages with precomputed prototype vectors, enabling flexible intent classification and management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing classifiers are used for intent classification, then intent classification can be performed, but computational requirements are high and the system is prone to overfitting
Solution Approach 1:
The patent creates prototype vectors that are simplified representations or copies of the original training data, storing only the essential characteristics needed for classification. These prototype vectors serve as lightweight substitutes for the full training dataset, enabling fast comparison with new messages while maintaining classification accuracy and reducing computational overhead during inference.
2Adaptability or versatility
If existing classifiers are used for intent classification, then intent classification can be performed, but the system lacks flexibility when adding new intents
Solution Approach 1:
The patent segments the intent classification system into independent prototype vectors, where each intent is represented by its own prototype. This segmentation allows individual intents to be added, removed, or modified by simply updating or replacing their corresponding prototype vectors without affecting other intents or requiring complete model retraining, thus enabling flexible adaptation.
Solution Approach 2:
The patent precomputes prototype vectors from training data before deployment. These precomputed prototypes serve as ready-to-use reference points for classification. When new intents need to be added, their prototypes can be precomputed in advance and integrated into the system without disrupting existing functionality, streamlining the adaptation process.
3Adaptability or versatility
If traditional intent classification approaches are used, then messages can be classified into predefined intents, but unmatched messages cannot be handled effectively
Solution Approach 1:
The patent implements a feedback mechanism where messages that do not match any existing prototype vector are identified and fed back into the system. This feedback loop enables the system to learn from unmatched messages, allowing for the creation of new prototypes or refinement of existing ones, thereby continuously improving the system's ability to handle diverse user inputs without losing valuable information.
Data Source
AI summary
In some applications, it may be desired to process a message to determine an intent of the message, where the intent indicates the meaning of the message. An intent classifier may be used to determine the meaning of a message by processing the message to compute a message embedding vector that represents the message in a vector space. Each possible intent may be represented by a prototype vector, and the intent of the message may be determined by comparing the message embedding to one or more prototype vectors, such as by selecting an intent whose prototype vector is closest to the message embedding. An intent classifier may be used, for example, (i) to implement an automated communications system with states where each state is associated with a subset of the possible intents or (ii) for processing usage data of a communications system to update the intents of the communications system.