Utterance Embedding Clustering for Intent Discovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems for determining customer intent in conversations have low accuracy and maturity levels, leading to frustration for customers and inefficiencies in resource allocation in customer support operations.
Innovation Solution
A system that combines utterance embeddings with latent Dirichlet allocation (LDA) to semi-automatically determine intents by analyzing human dialogues, providing more meaningful descriptions of call types and reducing the need for manual review, using techniques like universal sentence encoders and Siamese bidirectional encoder representations from transformers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing systems use simple keyword-based intent determination, then the system complexity is low, but the accuracy and maturity of intent determination deteriorates
Solution Approach 1:
The system segments the intent determination process into multiple stages: utterance embedding generation, clustering of similar utterances, topic modeling on clustered data, and intent classification. This segmentation allows each component to specialize and improve accuracy while maintaining manageable complexity through modular architecture.
Solution Approach 2:
The patent introduces intermediate representations (utterance embeddings and cluster labels) that bridge the gap between raw text and final intent determination. These intermediaries enable more accurate intent classification by capturing semantic relationships that simple keywords miss, while the automated nature of embedding generation prevents complexity escalation.
2Measurement precision
If manual review of individual dialogs is performed to figure out intents, then the accuracy of intent determination is improved, but the time and resource consumption increases
Solution Approach 1:
The system performs self-service by automatically generating intent labels through the pipeline of embedding generation, clustering, and topic modeling. The algorithm extracts key utterances and determines intents without human intervention, achieving high accuracy while eliminating manual review time. The system serves itself by using the dialogue data to train and refine its own intent determination capabilities.
Solution Approach 2:
The system performs preliminary actions by pre-processing dialogues into embeddings and pre-clustering utterances before intent determination. This preliminary structuring of data enables rapid automated intent classification, replacing time-consuming manual review while maintaining or improving accuracy through systematic analysis of all dialogue features.
3Ease of operation
If chat bots and virtual agents use basic intent recognition, then the ease of operation is maintained, but the customer satisfaction deteriorates due to inaccurate understanding
Solution Approach 1:
The system changes the parameters of intent recognition by transitioning from simple keyword matching to sophisticated embedding-based semantic analysis. This parameter change in the recognition approach dramatically improves understanding accuracy and customer satisfaction, while the automated pipeline maintains operational simplicity by handling complexity internally without requiring changes to user interaction patterns.
Data Source
AI summary
The present disclosure relates to a system, a method, and a product for intent discovery. The system includes a processor in communication with a memory storing instructions. When the processor executes the instructions, the instructions are configured to cause the processor to: obtain documents comprising a set of utterances, extract the set of utterances from the documents, generate a set of utterance embeddings based on the set of utterances, clusterize the set of utterance embeddings to obtain a plurality of clusters, obtain a cluster label for each cluster, encode each document based on a number of times each utterance cluster identifier (ID) appears to obtain an encoded document, perform latent Dirichlet allocation (LDA) on the encoded documents to obtain K topics, and each topic corresponding to a list of key clusters with cluster IDs, and for each topic, replace the cluster IDs with the cluster labels.