Natural Language Processing System Parallel Domain Sorting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech processing systems face challenges in efficiently incorporating new supplemental applications and domains due to restrictive rule-based approaches, which limit flexibility and resource utilization, especially when handling complex queries that may overlap with multiple applications.
Innovation Solution
A natural language processing system that performs early sorting of incoming queries into potential supplemental intent categories, allowing parallel processing across built-in and supplemental applications, and delays the selection of the top result until later in the NLU pipeline, thereby improving system performance and scalability.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If rule-based approaches are used to determine application domains, then system structure is simple and easy to implement, but system flexibility and adaptability deteriorate when handling complex queries
Solution Approach 1:
The system segments the domain determination process into two distinct stages: an initial rule-based filtering stage that quickly eliminates unrelated domains, followed by a machine learning-based classification stage that handles complex queries with flexibility. This segmentation allows the system to maintain simplicity where rules suffice while introducing adaptability where needed.
Solution Approach 2:
The system performs preliminary rule-based filtering before applying more complex machine learning models. By pre-processing queries with simple rules to eliminate obviously unrelated domains first, the system reduces the computational burden on the ML models and improves overall efficiency while maintaining adaptability for complex cases.
2Productivity
If all applications are processed in parallel to handle complex queries, then query handling capability improves, but resource overload increases
Solution Approach 1:
The system segments the application processing into two phases: initial parallel processing of all applications using rule-based filtering, followed by focused processing of only the top-scoring supplemental applications using machine learning. This segmentation maintains high query handling capability while reducing resource consumption by limiting full parallel processing to only necessary cases.
Solution Approach 2:
The system applies partial parallel processing by initially evaluating all applications in parallel through rule-based filtering, then selectively continuing parallel processing only for the top-scoring applications through machine learning classification. This partial action approach balances productivity with resource efficiency.
3Device complexity
If restrictive rule-based approaches are used, then system complexity is low, but ability to handle overlapping queries across multiple applications deteriorates
Solution Approach 1:
The system segments the query processing into rule-based initial filtering and ML-based refined classification. The rule-based segment handles simple, non-overlapping queries with low complexity, while the ML segment specifically addresses complex overlapping queries, allowing the system to manage complexity while improving adaptability for edge cases.
Solution Approach 2:
The machine learning classification system acts as an intermediary between the simple rule-based approach and the final domain selection. It mediates by taking the top-scoring applications from rule-based filtering and performing refined classification on overlapping queries, bridging the gap between simplicity and adaptability.
4Measurement precision
If machine learning models are trained for every supplemental application, then classification accuracy improves, but training time and computational resources increase
Solution Approach 1:
The system segments the machine learning application into two groups: built-in applications that receive dedicated trained models for high accuracy, and supplemental applications that use a general-purpose classification model. This segmentation achieves good classification accuracy for critical built-in apps while avoiding the time cost of training separate models for every supplemental application.
Solution Approach 2:
The system employs a universal machine learning model that can handle classification for multiple supplemental applications simultaneously. This single multi-functional model replaces the need for individual trained models for each supplemental application, significantly reducing training time and computational resources while maintaining acceptable accuracy.
Data Source
AI summary
A system capable of performing natural language understanding (NLU) using different application domains in parallel. A model takes incoming query text and determines a list of potential supplemental intent categories corresponding to the text. Supplemental applications within those categories are then identified as likely candidates for responding to the query. Application specific domains, including NLU components for the particular supplemental applications, are then activated and process the query text in parallel. Further, certain system default domains may also process incoming queries substantially in parallel with the supplemental applications. The different results are scored and ranked to determine highest scoring NLU results.