Utterance Intent Modeling With Significant Word Pairs
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional IVR systems struggle to accurately determine user intent from utterances, particularly in casual speech and multi-lingual contexts, leading to inefficiencies and increased operational costs due to the need for human intervention.
Innovation Solution
A system and method that builds an intent determination model by extracting significant word pairs, computing distinction factors and probabilities, and using these to identify user intent through a server-based model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If conventional IVR systems use natural language intent predictor (BERT model) and rule-based methods for predicting user intent, then the system can process user utterances automatically, but the intent determination accuracy deteriorates in casual speech and multi-lingual contexts
Solution Approach 1:
The patent segments the utterance into individual words and then extracts word pairs from these words. This segmentation approach allows the system to analyze casual speech and multi-lingual content at a granular level, improving intent determination accuracy by focusing on meaningful word combinations rather than treating the entire utterance as a single unit.
Solution Approach 2:
The patent extracts significant word pairs from the segmented words using statistical measures (distinction factor, positive probability, negative probability). This extraction process identifies the most discriminative word pairs that reliably indicate specific intents, filtering out noise from casual speech and multi-lingual variations while maintaining automatic processing capability.
2Device complexity
If conventional systems extract words and identify intent using predefined rules, then the system structure remains simple, but the system fails to determine the right intent from casual and multi-lingual utterances
Solution Approach 1:
The patent changes the parameters used for intent determination from simple predefined rules to statistical parameters including distinction factor, positive probability, and negative probability. These parameter changes enable the system to adapt to casual speech and multi-lingual utterances while maintaining a relatively simple system structure that processes words and word pairs through statistical calculations.
3Reliability
If enterprises allocate sufficient human resources to handle large customer base, then customer service quality improves, but operational cost increases
Solution Approach 1:
The patent implements an automated intent determination system that processes user utterances independently without requiring human intervention. The system uses statistical analysis of word pairs to automatically identify user intent, enabling the IVR system to handle customer queries autonomously. This self-service capability maintains customer service quality while eliminating the need for additional human resources, thereby reducing operational costs.
Data Source
AI summary
Disclosed is a system and method for determining an intent of a user from an utterance of the user. The system initially builds an intent determination model using a plurality of sample utterances, each assigned with at least one intent class. That is, on receiving the sample utterances, the system extracts significant word pairs from each sample utterance, computes a distinction factor of each significant word pairs, computes a Positive Probability and a Negative Probability for each significant word pairs, and generates the intent determination model by storing each significant word pairs and its distinction factor, the Positive Probability and the Negative Probability. Then on receiving any new utterance, the system extracts significant word pairs, identifies one or more matching word pairs in the model and determines the intent of based on the distinction factor, Positive Probability, and the Negative Probability of the one or more matched word pairs.