Probabilistic Intent Prediction for Unclear Queries in Conversation Systems

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional conversation systems fail to effectively process unclear intent queries due to difficulties in interpreting user intents, particularly in cases of voice recognition failures or noise-induced false recognition, without providing adequate solutions for handling such scenarios.

Innovation Solution

A deep learning-based probabilistic model is employed to predict the intent of a user for unclear intent queries, utilizing a query and associated context, and providing responses such as guides or confirmations based on natural language understanding partial analysis results.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional NLU-based intent analysis is used, then the system can process clear intent queries, but it fails to handle unclear intent queries effectively

Engineering Contradiction:

Improveintent analysis reliabilityVSAvoidquery type coverage

Core Design Contradiction:

ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent applies universality by making the intent analysis system handle both clear and unclear intent queries through a unified architecture. The probabilistic model serves multiple functions: it processes traditional NLU queries and also handles unclear queries by predicting intent probabilities, making the system versatile across different query types without requiring separate processing paths

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Measurement precision

If the system requests user re-utterance for unclear queries, then it maintains accuracy, but it increases interaction time and user burden

Engineering Contradiction:

Improveintent recognition accuracyVSAvoidinteraction time

Core Design Contradiction:

Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by providing useful responses even when intent recognition is not complete. Instead of requiring full clarification before responding, the system generates probabilistic intent predictions and provides partial responses or guidance based on the most likely intents, reducing interaction time while maintaining acceptable accuracy levels

Inventive Principle:

Principle #16Partial or excessive action

3Adaptability or versatility

If deep learning probabilistic model is added, then unclear intent queries can be processed, but system complexity increases

Engineering Contradiction:

Improveunclear query handling capabilityVSAvoidsystem architecture complexity

Core Design Contradiction:

Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies merging by integrating the probabilistic model with the existing NLU pipeline. Rather than creating a completely separate system for handling unclear queries, the probabilistic model is combined with the traditional NLU components, sharing resources and processing paths, which reduces overall system complexity while extending capability

Inventive Principle:

Principle #5Merging (Combining)

Data Source

PatentUS11403345B2Method and system for processing unclear intent query in conversation system

Publication Date: 2022.08.02 NAVER CORP

AI summary

A query processing method may include receiving a query based on an utterance of a user and a context associated with the user; analyzing an intent of the user for the query based on a natural language understanding (NLU) for the query; and in response to the intent of the user being undetermined through the analyzing, predicting the intent of the user using a deep learning based probabilistic model having the query and the context as an input.