Jointly Trained Chatbot Classifier for Utterance Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Chatbots in provider networks face challenges in accurately understanding user utterances and routing them to the appropriate services, leading to incorrect processing and responses, as they often require users to explicitly specify intent or misinterpret commands as search queries instead of executing them as intended.
Innovation Solution
A jointly trained service classifier and auxiliary classifiers are used to determine if a user utterance is in-scope or out-of-scope and route it to the correct service, employing masked cross-entropy loss and contrastive loss for improved accuracy, allowing the chatbot to infer the user's intent without explicit specification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If a chatbot uses simple routing without joint training, then the device complexity is reduced, but the measurement precision of intent determination deteriorates
Solution Approach 1:
The patent merges the service classifier and auxiliary classifiers into a single jointly trained model. The service classifier determines whether an utterance is in-scope or out-of-scope, while the auxiliary classifiers simultaneously determine the specific service. This unified approach allows shared representations and joint optimization, improving intent determination accuracy without requiring multiple separate systems.
Solution Approach 2:
The service classifier serves multiple functions: it acts as both a scope determination classifier (in-scope vs out-of-scope) and provides features for the auxiliary service classifiers. This multi-functionality reduces overall system complexity while maintaining high precision, as the same classifier contributes to multiple classification objectives through joint training.
2Measurement precision
If a chatbot requires explicit intent specification from users, then the intent determination accuracy is improved, but the ease of operation deteriorates
Solution Approach 1:
The chatbot system performs self-service by automatically inferring user intent through the jointly trained classifiers without requiring explicit intent specification from users. The service classifier and auxiliary classifiers work together to automatically determine both the scope and specific service for each utterance, enabling the system to serve itself in intent interpretation while maintaining high accuracy.
Solution Approach 2:
The joint training process performs preliminary action by pre-training the classifiers on diverse utterance-service pairs before deployment. This preliminary training enables the system to automatically understand and route various types of user inputs (commands, questions, search queries) without requiring users to explicitly specify their intent during actual operation.
3Measurement precision
If the chatbot uses separate classifiers for scope determination and service routing, then the measurement precision is improved, but the device complexity increases
Solution Approach 1:
The patent combines the scope determination classifier and service routing classifiers into a single jointly trained model. The service classifier outputs predictions for both in-scope/out-of-scope determination and service identification, while auxiliary classifiers focus on service routing. This merging reduces the number of separate components while maintaining the precision benefits of specialized classification through joint training and shared representations.
Data Source
AI summary
Techniques for chatbot utterance routing in a provider network include jointly training a service classifier and a plurality of auxiliary classifiers based on a mixed service set of labeled chatbot utterance training examples to yield a trained service classifier. When a particular chatbot user utterance is received, the trained service classifier can be used to determine if the utterance is in-scope or out-of-scope, and if in-scope, to determine which service of a set of services in the provider network to which to route the utterance for further processing. By jointly training the service classifier with the auxiliary classifiers, the accuracy of the in-scope/out-of-scope determination by the trained service classifier is improved as well as its accuracy in routing the utterance to the appropriate service for processing the utterance as intended by the user.