Deep Neural Network Query Classification for Search Appropriateness
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for detecting inappropriate queries in search engines, such as blocklists and SVM models, face limitations including manual maintenance, lack of generalization, and low recall, especially with novel or grammatically incorrect queries.
Innovation Solution
A deep neural network (DNN) model, specifically a deep structured semantic model (DSSM), is trained to classify query appropriateness, generating vector representations that enable the identification of patterns in queries, including previously unseen ones, and filtering auto-complete suggestions or managing responses accordingly.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If blocklist techniques are used to detect inappropriate queries, then precision is improved, but recall deteriorates and the system cannot generalize to novel query forms
Solution Approach 1:
The patent replaces mechanical blocklist techniques with a neural network-based semantic analysis system. The neural network model learns semantic representations of queries and automatically identifies inappropriate content through pattern recognition, eliminating the need for manual blocklist curation and enabling generalization to novel query forms while maintaining both precision and recall
Solution Approach 2:
The patent transforms the detection approach from exact string matching to semantic similarity measurement. By changing the parameter from lexical overlap to semantic distance in a vector space, the system can identify inappropriate queries even when they use different wording or grammatical structures, thereby improving recall without sacrificing precision
2Measurement precision
If blocklist techniques are used to detect inappropriate queries, then precision is improved, but device complexity increases due to manual curation and maintenance
Solution Approach 1:
The patent implements a self-training neural network system that automatically learns from query data without requiring manual blocklist curation. The model autonomously identifies patterns of inappropriate content through supervised learning, eliminating the need for continuous human maintenance and reducing operational complexity while maintaining high precision
Solution Approach 2:
The patent replaces the manual mechanical process of blocklist creation and maintenance with an automated neural network system. The model learns semantic representations and classification rules automatically through training, substituting human labor with intelligent algorithms that reduce long-term system complexity
3Extent of automation
If SVM models are used to classify query appropriateness, then automation is improved, but precision deteriorates compared to blocklists
Solution Approach 1:
The patent replaces SVM-based automated classification with a neural network system that learns hierarchical semantic representations. The neural network's ability to capture complex non-linear relationships and contextual meanings enables it to achieve both high automation and superior precision compared to traditional SVM approaches
4Reliability
If deep neural network models are used to classify queries, then recall is improved and generalization to novel queries is enabled, but processing time increases
Solution Approach 1:
The patent pre-trains the neural network model on large datasets to learn robust semantic representations before deployment. This preliminary training enables the model to quickly and accurately classify novel queries during operation, reducing inference time while maintaining high recall and generalization capabilities
Data Source
AI summary
Non-limiting examples of the present disclosure describe query classification to identify appropriateness of a query. A query may be received by at least one processing device. A deep neural network (DNN) model may be applied to evaluate the query. A vector representation may be generated for query based on application of the DNN model, where the DNN model is trained to classify queries according to a plurality of categories of appropriateness. The DNN model may be utilized to classify the query in a category of appropriateness based on analysis of the vector representation. In one example, auto-complete suggestions for the query may be filtered based on the classification of the category of appropriateness. In another example, classification of the query may be provided to an entry point. In yet another example, a response to the query is managed based on the classification of the query. Other examples are also described.