Customer Intent Prediction via Text Normalization and Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automated systems for customer support, such as IVR and chat assistants, face challenges in interpreting natural language communications due to factors like accents, slang, spelling errors, and unrelated questions, leading to frustrating interactions and reduced sales for enterprises.
Innovation Solution
A computer-implemented method and apparatus that converts non-textual portions of natural language communication to text, performs normalization operations like spelling correction and word stemming, and uses classifiers to predict customer intent, enabling provision of relevant responses on various interaction channels.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If automated systems use scripted or menu-based approaches to handle customer interactions, then the system structure is simple and easy to implement, but the system cannot accurately interpret natural language communications containing accents, slang, spelling errors, or unrelated questions
Solution Approach 1:
The patent introduces an intermediary natural language processing system between the customer and the automated support system. This intermediary layer includes components for speech-to-text conversion, text normalization, and intent classification that bridge the gap between informal customer language and structured system responses, thereby improving interpretation accuracy without requiring complete system redesign
Solution Approach 2:
The patent segments the automated system into distinct functional modules: speech-to-text conversion module, text normalization module, intent classification module, and response generation module. Each module handles a specific aspect of natural language processing, allowing the system to manage complexity through modular design while improving overall interpretation reliability
2Reliability
If the system uses comprehensive natural language processing to handle all variations of customer input, then interpretation accuracy improves, but processing time and computational resources increase
Solution Approach 1:
The patent applies preliminary text normalization and preprocessing operations before intent classification. By standardizing text format, correcting common spelling errors, and normalizing speech-to-text conversions in advance, the system reduces the complexity of subsequent intent analysis, thereby improving accuracy without proportionally increasing total processing time
Solution Approach 2:
The patent implements a tiered processing approach where the system applies full natural language processing only when necessary, and uses simplified processing for routine queries. The intent classification system handles common customer intents with minimal processing while applying more comprehensive analysis only for complex or ambiguous cases, balancing accuracy with processing efficiency
3Measurement precision
If the system uses multiple processing operations on textual data including normalization, spelling correction, and word stemming, then interpretation quality improves, but the complexity of text processing increases
Solution Approach 1:
The patent merges multiple text processing operations (normalization, spelling correction, word stemming, and formatting standardization) into a unified text preprocessing module. This consolidated approach handles various text quality issues simultaneously through coordinated processing steps, improving interpretation precision while managing complexity through integrated design rather than separate independent systems
Data Source
AI summary
A computer-implemented method and an apparatus facilitate customer intent prediction. The method includes receiving natural language communication provided by a customer on at least one enterprise related interaction channel. Textual data corresponding to the natural language communication is generated by converting one or more non-textual portions in the natural language communication to a text form. One or more processing operations are performed on the textual data to generate normalized text. The normalized text is configured to facilitate interpretation of the natural language communication. At least one intention of the customer is predicted, at least in part, based on the normalized text and a reply is provisioned to the customer based on the predicted intention. The reply is provisioned to the customer on the at least one enterprise related interaction channel in response to the natural language communication.