Masked Intent Model for Support Ticket Routing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing intent detection techniques face challenges in accurately predicting relevant intents for end users, leading to incorrect routing of support tickets and inefficient resource utilization due to the need for multiple models and complex industry assignments.
Innovation Solution
A masked intent model that combines the benefits of generic and organization-specific models by masking irrelevant intents, allowing only relevant predictions, thereby reducing the number of models required and improving prediction performance.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If organization-specific models are trained for each organization, then prediction accuracy is improved, but the number of models to train, deploy, and maintain increases significantly
Solution Approach 1:
The patent segments the intent prediction task by introducing a mask that divides the full intent taxonomy into organization-specific relevant intents and irrelevant intents. This allows a single generic model to function effectively for multiple organizations by selectively enabling only relevant intent predictions through the mask mechanism.
Solution Approach 2:
The patent changes the parameters of the prediction task by introducing a mask vector that dynamically adjusts which intents are considered valid predictions. Instead of training separate models with different architectures, the same model parameters are reused with different mask configurations for different organizations.
2Ease of manufacture
If a single generic model is used for all organizations, then model maintenance is simplified, but the model predicts invalid intents that are not relevant to specific organizations
Solution Approach 1:
The patent extracts the organization-specific intent filtering requirement from the model training process and implements it as a separate mask component. This allows the generic model to maintain its simplicity while the mask extracts and enforces organization-specific intent relevance rules.
Solution Approach 2:
The mask serves as an intermediary between the generic model and the organization-specific intent requirements. It mediates the prediction process by filtering out irrelevant intents without requiring modifications to the underlying model architecture or training data.
3Device complexity
If industry-specific models are used, then the number of models is reduced, but accurate assignment of organizations to industries is required which is difficult in many cases
Solution Approach 1:
The patent creates a universal intent mask that can serve multiple organizations across different industries with a single configuration. The mask is designed to be adaptable to any organization by simply specifying which intents are relevant, eliminating the need for industry-specific model variants or complex classification systems.
Data Source
AI summary
The present disclosure relates to predicting an intent associated with a support ticket. In one example, a method for predicting intents includes providing, as inputs to an intent model: the support ticket; and a set of one or more relevant intents associated with an organization associated with the support ticket, wherein the set of one or more relevant intents is a subset of a set of possible intents the intent model is configured to predict; and receiving, as output from the intent model, one or more predicted intents associated with the support ticket, wherein the one or more predicted intents are in the set of one or more relevant intents associated with the organization.