Dialog System Intent Prediction via Concept Graph Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current conversational AI systems, such as IBM Watson Assistant, require extensive manual effort from domain experts to create and assign examples for new intents, which is time-consuming and prone to errors, especially in medium and large-scale dialog systems.
Innovation Solution
The method involves generating predicted intents from an intent concept graph by combining nodes, creating graph and sentence embeddings, and automatically matching these intents with examples from call logs, allowing for the automatic creation and assignment of new intents without manual input, with optional user verification.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If domain experts manually create new intents and assign examples from call logs, then the dialog system can handle new user utterances, but the process is time-consuming and labor-intensive
Solution Approach 1:
The system performs self-service by automatically analyzing call logs, extracting intents, and creating intent definitions without requiring manual intervention from domain experts. The automated intent extraction system processes unstructured call log data, identifies patterns, and generates structured intent representations autonomously.
Solution Approach 2:
The manual mechanical process of domain experts reviewing and creating intents is replaced with an automated computational system that uses natural language processing and machine learning algorithms to extract intents from call logs, significantly reducing the time and effort required.
2Measurement precision
If domain experts manually curate examples for each intent, then the system achieves high accuracy in intent recognition, but the process is prone to errors and requires extensive manual effort
Solution Approach 1:
The system automatically curates examples by analyzing call logs and selecting representative utterances for each detected intent, eliminating the need for manual example selection by domain experts while maintaining high accuracy through automated pattern recognition.
Solution Approach 2:
The system incorporates feedback mechanisms where the automated intent extraction process continuously improves by learning from the structure and patterns identified in call logs, refining its intent detection accuracy over time without requiring manual re-curation of examples.
3Adaptability or versatility
If the dialog system includes thousands of intents to handle diverse user queries, then the system becomes more versatile, but the complexity of maintaining and updating intents increases significantly
Solution Approach 1:
The system segments the large set of thousands of intents by automatically organizing them into hierarchical structures or clusters based on their semantic relationships, making the complex intent space more manageable and easier to maintain while preserving comprehensive coverage of user queries.
Solution Approach 2:
The automated intent extraction system serves multiple functions simultaneously: it discovers new intents, categorizes existing intents, extracts examples, and validates intent definitions, thereby managing the complexity of thousands of intents through a single multi-functional automated process.
Data Source
AI summary
A method of automatically creating new intents and assigning examples to those new intents in a dialog system includes generating at least one predicted intent from an intent concept graph by combining two or more nodes of the intent concept graph. A graph embedding is generated of the at least one predicted intent. A sentence embedding is generated from examples from a call log. The graph embedding of the at least one predicted intent is then matched with the examples from the sentence embedding and the at least one predicted intent, with the examples, is connected into a dialog system.