Category Tag Mining via Sequence Labeling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engines face challenges in accurately identifying and mining new category words from user queries, which are essential for understanding user intentions and improving search results, as they often rely on pre-existing category thesauruses that do not account for emerging terms.
Innovation Solution
A method utilizing a pre-trained sequence labeling model and neural network architecture, including semantic representation models, bidirectional GRU layers, and CRF parsing, to label and filter category tags from user queries, removing existing tags and identifying new category tags by analyzing part of speech, fluency, and frequency, ensuring timely updates to the category tag library.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a pre-existing category thesaurus is used for search, then search coverage is limited to known categories, but new emerging category words cannot be identified
Solution Approach 1:
The patent applies preliminary action by pre-training a sequence labeling model on existing category thesauruses before deployment. This pre-training establishes a baseline understanding of known categories, and then the model continuously adapts to new categories through online learning from actual search queries, resolving the contradiction between having solid foundational knowledge and adapting to emerging terms.
Solution Approach 2:
The patent implements dynamics by making the category tag library dynamic rather than static. The system continuously updates the category library by identifying new category words from search queries and incorporating them into the thesaurus. This dynamic update mechanism allows the system to maintain high accuracy on known categories while simultaneously adapting to cover new emerging categories.
2Measurement precision
If manual category tag updates are performed, then category library accuracy can be maintained, but the process is time-consuming and cannot keep pace with emerging terms
Solution Approach 1:
The patent applies self-service by enabling the category library to automatically update itself through the sequence labeling model. The model autonomously identifies new category words from search queries, validates them against existing categories, and updates the thesaurus without human intervention. This self-updating mechanism eliminates the time-consuming manual update process while maintaining high accuracy through the model's learned discrimination capabilities.
Solution Approach 2:
The patent implements feedback by using the sequence labeling model to continuously analyze search queries and provide feedback on potential new category words. The model's predictions are fed back into the category library update process, creating a closed-loop system that automatically refines and expands the category thesaurus based on actual usage patterns, thereby reducing update time while maintaining precision.
3Measurement precision
If a complex neural network model is trained for category tag extraction, then extraction accuracy improves, but model training complexity and computational resources increase
Solution Approach 1:
The patent applies universality by designing a multi-functional sequence labeling model that performs multiple tasks: part-of-speech tagging, named entity recognition, and category tag extraction all within a single unified architecture. This multi-functional approach improves extraction accuracy by leveraging shared representations across different linguistic tasks while avoiding the need for separate complex models for each function.
Solution Approach 2:
The patent merges multiple sequence labeling tasks (POS tagging, NER, and category extraction) into a single unified model. By combining these functions into one architecture with shared layers and parameters, the system achieves high extraction accuracy through joint learning while reducing overall model complexity compared to using separate specialized models for each task.
Data Source
AI summary
Provided by the present disclosure is a new category tag mining method, involving the field of knowledge graph technology, and including: obtaining a plurality of queries during a current preset time period; labeling a category tag on each query of the plurality of queries, by using a pre-trained sequence labeling model, to extract the category tag currently corresponding to the query from the query; and removing a category tag already existing in a preset current category tag library from category tags currently corresponding to all the queries, and determining a remaining category tag as a new category tag. The present disclosure also provides an electronic device and a non-transitory computer-readable storage medium.