Semantic Embedding Routing for Multi-Domain Chatbots
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing chatbots in the enterprise space are limited to interoperating with a single type of back-end software system, requiring multiple chatbots to be deployed and used when multiple different systems are employed, increasing complexity and cost for users and IT infrastructure.
Innovation Solution
An AI routing framework for a multi-domain chatbot that processes user queries by calculating semantic embeddings, matching them to stored embeddings in a repository, and routing them to appropriate back-end software systems, allowing a single chatbot interface to handle queries across multiple domains.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple separate chatbots are deployed to support multiple back-end systems, then each chatbot can be optimized for its specific domain, but the device complexity and number of components increase
Solution Approach 1:
The patent combines multiple domain-specific chatbots into a single multi-domain chatbot that can handle queries across multiple back-end systems (Oracle Fusion, SalesForce, SAP) through unified semantic routing. The system merges the functionality of separate chatbots while maintaining domain-specific processing capabilities through embedding-based routing to appropriate handlers.
Solution Approach 2:
The chatbot is designed with universal functionality to interact with multiple different back-end software systems through a single interface. The semantic routing framework enables the same chatbot to universally handle diverse domains by calculating embeddings and routing queries to domain-specific handlers without requiring separate specialized chatbots.
2Reliability
If users must learn and use multiple different chatbots for different systems, then each chatbot can provide specialized functionality, but the ease of operation decreases
Solution Approach 1:
Users interact with a single multi-domain chatbot that provides universal access to multiple back-end systems. The chatbot maintains specialized functionality for each domain through semantic routing while presenting a unified interface, eliminating the need for users to learn and switch between multiple different chatbots.
Solution Approach 2:
The semantic routing framework acts as an intermediary between the user's natural language query and the domain-specific handlers. It calculates embeddings, matches them to stored embeddings in the repository, and routes queries to appropriate handlers, allowing users to interact naturally without needing to know which specific system handles which query.
3Device complexity
If a single chatbot handles multiple domains, then the device complexity is reduced, but the difficulty of detecting and measuring the appropriate routing increases
Solution Approach 1:
The patent replaces traditional mechanical or rule-based routing mechanisms with semantic embedding-based routing. Instead of using keyword matching or predefined rules, the system calculates semantic embeddings of queries and matches them to embeddings in a repository, enabling accurate routing based on meaning rather than surface-level text patterns.
Solution Approach 2:
The system changes the parameter space for routing from simple keyword or category matching to high-dimensional semantic embeddings. By representing queries and handlers in a shared embedding space, the system can accurately detect and measure semantic similarity, improving routing accuracy while handling multiple domains within a single chatbot.
Data Source
AI summary
An AI routing framework for a multi-domain chatbot is provided. According to one set of embodiments, a computer system implementing the AI routing framework can receive a user query/request submitted to the multi-domain chatbot, determine an input statement corresponding to the user query/request, and calculate an embedding of the input statement reflecting the input statement's semantic content. The computer system can further identify, from among a plurality of statement embeddings maintained in a repository, a statement embedding that most closely matches the embedding of the input statement, and determine a domain associated with the identified statement embedding in the repository. The computer system can then forward the input statement to a handler of a back-end software system configured to process user queries/requests belonging to the determined domain.