Weighting Dictionary Entities for Spoken Language Understanding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Spoken Language Understanding (SLU) systems face challenges in improving accuracy due to noisy dictionaries, which are either manually collected and time-consuming or automatically collected with noise, leading to inadequate performance, especially when test data differ significantly from training data.
Innovation Solution
A classifier model is built using a seed list of entities belonging to a particular entity class and a background list of entities not belonging to that class, trained with features from search click logs, search result URLs, and entity graph attributes to weight entities in a candidate dictionary, providing a cleaned and improved dictionary for SLU systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If dictionaries are manually collected, then entity quality is improved, but collection time and labor cost increase
Solution Approach 1:
The system uses automatically collected entities to train a classifier model that automatically filters and weights entities, enabling the dictionary collection process to self-improve without manual intervention. The classifier learns from the automatically collected data and produces cleaned dictionaries autonomously.
Solution Approach 2:
The patent replaces manual dictionary collection and cleaning processes with an automated machine learning-based system. The classifier model automatically identifies and weights relevant entities, substituting the mechanical manual process with an automated computational approach.
2Productivity
If automatically collected dictionaries are used, then collection speed is improved, but noise and accuracy deteriorate
Solution Approach 1:
The system uses search click logs and user interaction data as feedback to train the classifier model. The classifier continuously learns from actual user behavior patterns to improve its ability to identify relevant entities, enabling automatic cleaning of noisy dictionaries while maintaining high accuracy.
Solution Approach 2:
The patent transforms automatically collected entities into weighted entities by changing the parameter of entity relevance through machine learning. The classifier model assigns weight scores to entities based on their likelihood of belonging to the target entity class, converting raw automatic collection data into refined, high-accuracy dictionaries.
3Reliability
If large dictionaries are used to increase coverage, then model performance is improved, but dictionary size and processing complexity increase
Solution Approach 1:
The system extracts and removes noisy, irrelevant entities from large automatically collected dictionaries using the trained classifier model. By taking out only the problematic entries rather than manually curating entire dictionaries, the system maintains comprehensive coverage while significantly reducing noise and improving processing efficiency.
Data Source
AI summary
A dictionary used by a spoken language understanding (SLU) system is improved by providing weightings for entities in the dictionary that represent the likelihood each entity belongs to an entity class represented by the dictionary. A classifier model may be trained using a seed list containing sample entities that belong in the entity class and a background entity list containing samples that do not belong in the entity class. Clicked URLs from search logs, search result URLs, and attributes from an entity graph may be used as features of the sample entities to train the classifier model. The classifier model may be used to weight entities from a candidate dictionary. The entity weightings are used to generate an improved dictionary for use in the SLU system.