Language Model Segmentation for NLU Classification Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional natural language understanding (NLU) systems face challenges in accurately classifying user requests due to the complexity of handling multiple actions and nuances in language inputs, requiring extensive manual effort and skill to develop and configure monolithic statistical language models, which can lead to misinterpretation and reduced accuracy.
Innovation Solution
A system and method for automatically generating a set of language models configured into an optimal combination using a hierarchical tree structure, allowing for recursive building and partitioning of language models to maximize interpretation performance, reducing the need for manual effort and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a single monolithic statistical language model is used to classify user requests, then the system can handle a broad range of language inputs, but the complexity of training and tuning increases significantly as the number of actions increases
Solution Approach 1:
The patent divides the monolithic language model into multiple specialized language models, each trained on a specific subset of training data corresponding to particular actions or categories. This segmentation reduces the complexity of training and tuning each individual model while maintaining the ability to handle a broad range of language inputs through the collective ensemble of models.
2Adaptability or versatility
If more training data is provided to enable sophisticated tasks, then the language model can perform more complex actions, but the training and tuning process becomes more complex
Solution Approach 1:
The patent segments the large training data set into multiple smaller subsets, each dedicated to specific actions or categories. Multiple language models are trained on these subsets independently, which simplifies the training process for each model while collectively enabling sophisticated task capabilities through the combined knowledge of all models.
3Device complexity
If a single monolithic language model is used, then the system structure is simpler, but the amount of required data increases and classification accuracy decreases due to overlap between actions
Solution Approach 1:
The patent segments the classification task into multiple specialized models, each focusing on specific actions or categories. This reduces overlap and confusion between actions by dedicating specific models to specific tasks, thereby improving classification accuracy while maintaining manageable system structure through modular organization.
Solution Approach 2:
The patent applies local quality by making each language model specialized in specific domains or actions rather than general-purpose. Each model develops expertise in its specific area, leading to more accurate classifications for those particular actions while the overall system maintains comprehensive coverage through the collection of specialized models.
4Ease of operation
If manually training language models is performed, then the developer can control the classification process, but the time and skill required to identify and classify training data increases
Solution Approach 1:
The patent implements self-service by enabling the system to automatically identify, classify, and organize training data into appropriate subsets for different language models. This automation reduces the manual time and skill required while maintaining developer control through the ability to configure and supervise the automated classification process.
Data Source
AI summary
The invention disclosed herein concerns a system (100) and method (600) for building a language model representation of an NLU application. The method 500 can include categorizing an NLU application domain (602), classifying a corpus in view of the categorization (604), and training at least one language model in view of the classification (606). The categorization produces a hierarchical tree of categories, sub-categories and end targets across one or more features for interpreting one or more natural language input requests. During development of an NLU application, a developer assigns sentences of the NLU application to categories, sub-categories or end targets across one or more features for associating each sentence with desire interpretations. A language model builder (140) iteratively builds multiple language models for this sentence data, and iteratively evaluating them against a test corpus, partitioning the data based on the categorization and rebuilding models, so as to produce an optimal configuration of language models to interpret and respond to language input requests for the NLU application.