Search Query Category Classification Model
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing online shopping systems struggle to accurately reflect product category scores in search results for infrequent queries, leading to mismatched user intentions and reduced click-through rates.
Innovation Solution
A method and system that calculate category probabilities using a classification model trained on product category distribution data, incorporating word distribution information and importance weights to improve category prediction for both frequent and infrequent queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a classification model is trained based on user selections from frequent queries, then product category scores are well reflected in search results for frequent queries, but product category scores cannot be properly reflected in search results for infrequent queries
Solution Approach 1:
The system performs preliminary action by pre-calculating and storing category appearance frequency data for words in the product database before user queries are received. This pre-computed frequency information is then retrieved and used by the classification model to handle infrequent queries, eliminating the need to wait for user feedback to learn category relationships.
Solution Approach 2:
The patent introduces an intermediary mechanism that combines multiple data sources: user selection data from frequent queries, category appearance frequency data from the product database, and query word frequency data. This intermediary composite data structure enables the classification model to make accurate predictions for infrequent queries by leveraging patterns from both frequent queries and pre-computed frequency information.
2Ease of operation
If simple query-comparison extraction is used, then the system is easy to operate and fast, but it cannot derive product search results that match user's intended categories
Solution Approach 1:
The patent replaces the simple mechanical query-comparison extraction method with an AI-based classification model that uses machine learning. The model processes query words, retrieves category appearance frequency data, and calculates category probabilities through neural network computations, substituting basic string matching with intelligent pattern recognition and learning.
Solution Approach 2:
The system changes the fundamental parameters of search processing by introducing multiple new dimensions: category appearance frequency as a weighting parameter, query word frequency normalization, and probability threshold adjustments. These parameter changes enable the system to distinguish between frequently and infrequently queried terms, improving category match accuracy while maintaining operational simplicity.
3Reliability
If category probability is calculated using only user selection data, then the model learns from actual user behavior, but the system cannot handle infrequent queries with limited training data
Solution Approach 1:
The patent merges multiple data sources into a unified training approach: user selection data from frequent queries, category appearance frequency data from the product database, and query word frequency data. This combination creates a richer, more comprehensive training dataset that provides sufficient examples even for infrequent queries, solving the data quantity limitation.
Solution Approach 2:
The classification model is designed with multi-functionality to handle both frequent and infrequent queries using the same unified architecture. The model universally applies the same processing steps for all queries while adapting to different data availability patterns, making it effective across the entire query distribution spectrum without requiring separate models for different query frequencies.
Data Source
AI summary
A method for determining categories for a search query includes obtaining distribution information on each of a plurality of categories of one or more words included in the query; calculating features of the one or more words based on the distribution information on each of the plurality of categories; and calculating, by a classification model, information on at least one category related to the query based on the features of the one or more words and the query.