NLU Processor Entity Resolution for Privacy-Safe Customer Service
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In speech recognition systems, distinguishing between recognition errors due to automatic speech recognition (ASR) and natural language understanding (NLU) processing is challenging, especially when customer service agents need to resolve issues without accessing users' private data, leading to potential privacy concerns and bad customer experiences.
Innovation Solution
Implementing a system where customer service agents interact with a natural language understanding processor using personalized lexicons and gazetteers associated with user-specific Entity IDs, allowing them to submit queries that bypass ASR and directly address NLU for personalized interpretations, while keeping user data private.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If customer service agents access users' private data to resolve speech processing errors, then they can provide accurate recommendations and improve customer service, but user privacy is compromised
Solution Approach 1:
The patent introduces an intermediary system (the speech processing error resolution system) that enables customer service agents to interact with NLU processors and obtain personalized interpretations without directly accessing users' private data. The system acts as a mediator between the agent's need for accurate information and the user's privacy protection requirements, using Entity IDs and personalized lexicons to bridge this gap.
Solution Approach 2:
The patent segments the speech processing system into distinct components: ASR processors, NLU processors, entity resolution systems, and customer service agent interfaces. This segmentation allows the customer service function to operate independently with controlled access to personalized information through Entity IDs, rather than requiring direct access to all user data.
2Measurement precision
If the system uses personalized lexicons and gazetteers for NLU processing, then entity resolution accuracy is enhanced, but system complexity increases
Solution Approach 1:
The patent implements preliminary action by pre-processing and storing personalized lexicons, gazetteers, and entity information in structured formats before they are needed for error resolution. The Entity IDs and associated personalized data are prepared in advance, allowing the NLU processor to quickly retrieve and apply them when analyzing speech errors, rather than creating these resources on-demand during customer service interactions.
3Productivity
If the system bypasses ASR to directly process NLU queries, then speech processing error resolution is improved, but the processing pipeline becomes more complex
Solution Approach 1:
The patent extracts the NLU processing capability from the standard speech processing pipeline and creates a separate, dedicated error resolution pathway. Customer service agents can submit queries that bypass the ASR component and go directly to the NLU processor, which is specifically configured to handle error analysis with personalized lexicons. This extraction allows efficient error resolution without requiring modifications to the main speech processing pipeline.
Data Source
AI summary
A system in which a customer service agent (CSA) is able to assist a customer with obtaining a desired response from a speech-controlled appliance while protecting customer data. The customer service agent submits queries to a natural language understanding (NLU) processor that performs entity resolution using personalized library information stored in an entity library based on the customer identity information and/or an device identifier. The CSA is shielded from the entity library itself, as well as data stored on the speech-controlled appliance. The CSA can instruct the NLU processor to deliver results to multiple endpoints, including both the customer's appliance and the CSA agent's console.