Multilingual Intent Model with Outlier Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud-based chatbots configured for specific languages struggle to accurately determine the intent of chat utterances in different languages, leading to reduced utility and increased memory and maintenance overhead due to the need for separate models for each language.
Innovation Solution
An intent classification model is configured with an encoder model and an outlier detection model, using triplet semi-hard loss and cross-entropy loss functions, along with a Gaussian mixture model for outlier detection, to support multiple languages and improve intent classification accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If separate machine learning models are configured for each language, then intent classification accuracy for that language is improved, but device complexity and maintenance overhead increase
Solution Approach 1:
The patent implements a single machine learning model that can process and classify intents across multiple languages. The model is trained on multilingual data and can handle different language inputs without requiring separate models for each language, thereby achieving universal functionality while reducing system complexity
Solution Approach 2:
The patent combines multiple language-specific intent classification capabilities into a single unified model. Instead of maintaining separate models for different languages, the system merges these functionalities into one model that processes all languages, reducing the overall number of models and simplifying the system architecture
2Device complexity
If a single machine learning model supports multiple languages, then device complexity is reduced, but intent classification accuracy for each language may deteriorate
Solution Approach 1:
The patent employs parameter tuning and optimization techniques to ensure the single multilingual model achieves high accuracy across all supported languages. By adjusting model parameters, training data distribution, and loss functions, the system maintains precise intent classification performance despite handling multiple languages
3Measurement precision
If separate models are maintained for each language, then intent classification accuracy is improved, but memory usage increases
Solution Approach 1:
The patent merges multiple language-specific models into a single shared model, thereby consolidating the memory resources required to store model parameters. This single model approach significantly reduces the total memory footprint compared to maintaining separate models for each language while still providing accurate intent classification across all languages
Data Source
AI summary
A method that includes receiving an input at an interactive conversation service that uses an intent classification model. The method may further include generating, using an encoder model of the intent classification model, a set of output vectors corresponding to the input, where the encoder model is configured to determine a set of metrics corresponding to intent classifications. The method may further include determining, using an outlier detection model of the intent classification model, whether the input is in-domain or out-of-domain (OOD) based on a first vector of the set of output vectors satisfying a domain threshold relative to one or more of the intent classifications. The method may further include outputting, by the intent classification model, a second vector of the set of output vectors that indicates the set of metrics corresponding to the intent classifications or an indication that the input is OOD.