Telemetry-Trained Abbreviation Search for Accurate Term Expansion
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional large language models (LLMs) struggle with accurately predicting expanded terms from abbreviated search queries due to a lack of context, leading to inefficient search results, particularly in domain-specific systems like SKU product search systems.
Innovation Solution
A neural network is trained using telemetry data to generate custom language embeddings that associate abbreviated terms with their most likely expanded terms, leveraging probabilistic tuples and dynamic input data modification during training to improve prediction accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional LLMs are used to predict expanded terms from abbreviated queries, then the system can process search queries, but the prediction accuracy deteriorates due to lack of context and domain-specific knowledge
Solution Approach 1:
The patent changes the training parameters and data distribution of the neural network to focus on abbreviated terms and their expanded forms. By modifying the input data to include probabilistic tuples of abbreviated terms with their expanded versions, the model learns to accurately predict expanded terms from abbreviations without requiring additional context, directly resolving the accuracy-context contradiction.
Solution Approach 2:
The system uses telemetry data from user search behavior as feedback to train the neural network. By analyzing actual user selections of expanded terms from abbreviations, the model continuously improves its prediction accuracy, allowing it to achieve high precision in predicting expanded terms from abbreviated queries.
2Productivity
If LLMs trained on public information are used, then the system can generate responses, but the performance deteriorates for domain-specific abbreviated terms due to inefficient prediction
Solution Approach 1:
The patent performs preliminary training of the neural network on domain-specific telemetry data before the system is deployed for production use. This preliminary action involves pre-processing user search behavior data to create probabilistic tuples that capture domain-specific abbreviation-to-expanded-term mappings, ensuring the model achieves high reliability and relevance for domain-specific queries before actual use.
Solution Approach 2:
The system applies local quality by customizing the neural network training to specific domain requirements. Instead of using generic public information training, the model is specifically trained on telemetry data from the target domain (e.g., SKU product search), allowing it to achieve high prediction reliability and search result relevance for domain-specific abbreviated terms.
3Measurement precision
If simplified NLP techniques are used, then the system achieves comparable search result accuracy, but the capability to handle complex queries deteriorates
Solution Approach 1:
The patent creates a universal neural network model that can handle both simple and complex search queries effectively. By training the model on diverse telemetry data containing various types of queries (abbreviated, expanded, domain-specific), the system achieves multi-functionality where the same model structure and training approach work for different query types, providing both high accuracy for simple queries and adaptability for complex ones.
Data Source
AI summary
A method, computer program product, and computing system for processing telemetry data associated with a target abbreviated term and a plurality of expanded terms. A plurality of probabilistic tuples are generated from the telemetry data associated with the target abbreviated term. A neural network is trained using the plurality of probabilistic tuples by modifying, with each training epoch, input data to the neural network by coupling a target abbreviated term with an expanded term associated with the target abbreviated term that is randomly selected from the plurality of expanded terms.