Virtual Assistant Server Intent Detection via Dynamic Model Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing virtual assistants struggle to comprehensively understand natural language due to complexities such as multiple intents, tense changes, ambiguous words, and contextual variations, leading to frustration among end users and reduced adoption.
Innovation Solution
A virtual assistant server that receives utterances from end users, identifies a current-node of execution in a conversation definition, and selects a set of language models to interpret the utterance, executing them in order to determine the intent and generate a response.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a linear, predefined conversation path is used for intent detection, then the system structure is simple and easy to implement, but the system cannot adequately support variations from the linear defined paths and fails to meet user expectations
Solution Approach 1:
The patent segments the conversation processing into multiple independent nodes, each representing a different conversation path or intent detection approach. This allows the system to handle both linear predefined paths and variations simultaneously by selecting and executing appropriate nodes based on the current conversation state, resolving the contradiction between implementation simplicity and adaptability.
2Measurement precision
If multiple language models are executed to comprehensively understand natural language, then the intent detection accuracy is improved, but the system complexity and processing time increase
Solution Approach 1:
The patent implements dynamic language model selection where the system determines which language models to execute based on the current conversation node and context. This dynamic approach allows the system to use multiple language models when needed for accurate intent detection while avoiding unnecessary model execution in simpler contexts, thus balancing accuracy with system complexity.
3Measurement precision
If multiple language models are executed to interpret utterances, then the intent detection accuracy is improved, but the processing time increases
Solution Approach 1:
The patent performs preliminary actions by pre-defining conversation nodes and associated language models before actual conversation processing. This allows the system to quickly determine which pre-configured models to execute based on the current node, avoiding the need to dynamically select and load models during processing, thus reducing processing time while maintaining accurate intent detection.
Data Source
AI summary
A virtual assistant server receives an utterance provided by an end user via a channel of a virtual assistant rendered in a client device. The virtual assistant server identifies a current-node of execution from a plurality of nodes of a conversation definition of the virtual assistant and identifies a first set of language models from a group of language models of the virtual assistant to interpret the utterance. Further, the virtual assistant server executes the first set of language models in an order based on the current-node until an intent of the utterance is determined. Subsequently, the virtual assistant server generates a response based on the intent and outputs the response to the client device.