Knowledge-Driven Dialog Workflow for Automated Agents
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current automated virtual agents, such as chatbots, often fail to meet user expectations in customer support due to the complexity of possible questions and interactions, relying heavily on manual dialogue scripts and semantic machine learning models that are not robust, leading to inefficiencies and incomplete resolutions.
Innovation Solution
Implementing a knowledge-driven dialog (KDD) workflow that uses conversation models to dynamically decide questions and solutions through iterative interactions, minimizing the need for pre-authored scripts and leveraging AI-assisted processes to analyze user inputs and provide appropriate responses based on intent and entity properties.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If dialogue scripts are used to conduct agent-to-human interactions, then the conversation flow can be controlled, but the system cannot handle unexpected questions or answers from users
Solution Approach 1:
The patent transitions from static dialogue scripts to dynamic conversation models that can adapt to unexpected user inputs. The system uses machine learning to dynamically generate appropriate responses based on the current conversation state, allowing the agent to handle unforeseen questions while maintaining coherent conversation flow.
Solution Approach 2:
The system employs self-service mechanisms where the conversation model automatically learns from and adapts to user interactions without requiring manual script updates. The model serves itself by continuously improving its response generation capabilities through exposure to diverse user inputs, reducing the need for extensive pre-authoring of dialogue scripts.
2Adaptability or versatility
If semantic machine learning models are created to learn conversation pathways, then the system can handle more variations, but the results are not robust due to heavy dependence on dialogue scripts and training inputs
Solution Approach 1:
The patent applies preliminary action by pre-training the conversation model on extensive diverse data before deployment. The model is prepared in advance with broad knowledge and conversation patterns, enabling it to handle variations robustly without being overly dependent on specific dialogue scripts during actual interactions.
Solution Approach 2:
The system changes the parameters of the machine learning model during training and deployment, adjusting hyperparameters, training data composition, and model architecture to optimize both adaptability and robustness. This includes modifying the balance between script-based training and free-form conversation training to achieve reliable performance across diverse scenarios.
3Adaptability or versatility
If manual work is expended to author and maintain support knowledge for dialogue scripts, then the conversation coverage can be improved, but significant time and effort are required
Solution Approach 1:
The conversation model serves itself by automatically learning conversation patterns and generating appropriate responses without requiring extensive manual authoring. The system reduces the need for human editors to maintain dialogue scripts by autonomously adapting to new conversation scenarios through continuous learning from user interactions and diverse training data.
Solution Approach 2:
The system uses copying mechanisms where the conversation model learns from existing conversation patterns and replicates successful interaction templates. Instead of manually creating unique responses for every scenario, the model copies and adapts proven conversation patterns from training data, significantly reducing the time required to achieve comprehensive conversation coverage.
Data Source
AI summary
Systems and devices to perform knowledge-driven dynamic conversations and select content within automated agents such as chatbots and virtual assistants are disclosed. In an example, operations to facilitate a knowledge-based conversation session with a human user using an automated agent include: receiving a conversational input regarding a support issue; analyzing the conversational input to determine an intent and applicable entity properties associated with the intent; performing a multi-turn conversation to identify a solution using the intent and the applicable entity properties, by exchanging iterative questions and answers between the automated agent and the user to dynamically recalculate applicability of the solution to the support issue; and outputting information associated with the identified solution. In further examples, the operations include a dynamic application of a solution policy and a diagnosis policy in the multi-turn conversation, to determine whether to deliver a solution or ask diagnosis questions.