Multi-tenant Conversation Engine Using Pretrained Neural Networks
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine learning-based conversation engines face challenges in supporting multiple languages and require extensive training data and resources, leading to low accuracy and inefficiency, especially in multi-tenant systems with limited data.
Innovation Solution
A pretrained multi-lingual model is used, which is fine-tuned with tenant-specific data, allowing for efficient training across multiple languages without the need for extensive data in each language, reducing computational resources and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If each tenant trains machine learning models independently using their own data, then the model can be customized for specific tenant needs, but the training accuracy is low due to insufficient training data
Solution Approach 1:
The patent combines training data from multiple tenants to create a shared training corpus, enabling the machine learning model to learn from diverse interactions across different tenants. This merging approach resolves the data scarcity problem for individual tenants while maintaining the ability to serve multiple tenants through a single robust model.
Solution Approach 2:
The patent creates a universal machine learning model that can serve multiple tenants with different requirements. The model is trained on aggregated data from various tenants and can adapt to serve each tenant's specific needs, achieving both multi-tenant support and customized performance without requiring separate models for each tenant.
2Measurement precision
If extensive training data is collected and processed for each tenant, then the prediction accuracy of machine learning models improves, but the training time and computational resources increase significantly
Solution Approach 1:
The patent performs preliminary actions by pre-processing and aggregating training data from multiple tenants before model training. This includes collecting interactions, annotating data, and preparing the training corpus in advance, which reduces the actual training time when the model needs to be deployed or updated for specific tenants.
Solution Approach 2:
By merging training data from multiple tenants into a shared corpus, the patent achieves better model accuracy with more diverse training examples, while the shared nature of the training process reduces redundant computational work compared to training separate models for each tenant.
3Productivity
If machine learning models are trained with limited data, then the training resources are conserved, but the quality of automatic conversations deteriorates
Solution Approach 1:
The patent creates a universal training approach where a single machine learning model serves multiple tenants with different conversation requirements. This multi-functional model maintains high conversation quality across various tenants while optimizing resource utilization by avoiding redundant training processes for each individual tenant.
Solution Approach 2:
The patent merges training resources and data across multiple tenants, allowing the system to achieve high conversation quality through a shared, robust model trained on diverse interactions. This approach improves resource efficiency by eliminating duplicate training efforts while maintaining or enhancing conversation quality through exposure to varied training data.
Data Source
AI summary
A system uses conversation engines to process natural language requests and conduct automatic conversations with users. The system generates responses to users in an online conversation. The system ranks generated user responses for the online conversation. The system generates a context vector based on a sequence of utterances of the conversation and generates response vectors for generated user responses. The system ranks the user responses based on a comparison of the context vectors and user response vectors. The system uses a machine learning based model that uses a pretrained neural network that supports multiple languages. The system determines a context of an utterance based on utterances in the conversation. The system generates responses and ranks them based on the context. The ranked responses are used to respond to the user.