Inferring Unknown Linguistic Intents via Click Log Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional machine learning techniques for annotating training data are labor-intensive and require manual labeling, which is cumbersome and inefficient for assigning intents to linguistic items.
Innovation Solution
A computer system that determines intents of linguistic items using a model, leveraging known intent labels and selection log data, such as click log data, to infer intents for items without labels, employing a generative model like a Bayesian hierarchical graphical model to represent user actions and influence subsequent actions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If manual annotation techniques are used to assign intent labels to linguistic items, then labeling accuracy can be maintained, but the process becomes labor-intensive and time-consuming
Solution Approach 1:
The system performs preliminary automated labeling using a generative model before manual review, pre-processing the data to reduce the workload of manual annotation while maintaining accuracy through subsequent verification steps
Solution Approach 2:
A semi-automated intermediary system is introduced that combines automated intent classification with manual verification, acting as a mediator between fully manual and fully automated approaches to balance accuracy and efficiency
2Productivity
If fully automated techniques are used to generate intent labels, then productivity increases, but labeling accuracy and reliability deteriorate
Solution Approach 1:
The system implements feedback mechanisms where automated labels are evaluated and refined through iterative processes, using performance metrics to adjust and improve the generative model's accuracy over time
Solution Approach 2:
The labeling system transitions from static manual processes to dynamic automated processes that can adapt and learn from data, allowing the system to improve its accuracy automatically while maintaining high productivity
3Adaptability or versatility
If a comprehensive set of intent classes is maintained to cover all possible user intentions, then the system's adaptability improves, but the complexity of intent classification increases
Solution Approach 1:
The comprehensive intent classification space is segmented into hierarchical levels and modular categories, organizing the complex set of intent classes into manageable groups that reduce classification difficulty while maintaining complete coverage
Solution Approach 2:
The generative model is designed with universal capabilities to handle diverse intent types through a unified framework, allowing a single system to perform multiple classification functions without requiring separate specialized classifiers for each intent type
Data Source
AI summary
Functionality is described herein for determining the intents of linguistic items (such as queries), to produce intent output information. For some linguistic items, the functionality deterministically assigns intents to the linguistic items based on known intent labels, which, in turn, may be obtained or derived from a knowledge graph or other type of knowledge resource. For other linguistic items, the functionality infers the intents of the linguistic items based on selection log data (such as click log data provided by a search system). In some instances, the intent output information may reveal new intents that are not represented by the known intent labels. In one implementation, the functionality can use the intent output information to train a language understanding model.