Sub-query Evaluation for Image Search Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Image search systems face challenges in ranking images for search phrases with multiple terms when a specific image relevance model is not available, as they often rely on models trained for subsets of the terms, requiring evaluation of similarity between user interaction rankings and sub-query model rankings to select the most appropriate model.
Innovation Solution
The system identifies a set of responsive images, determines interaction rankings based on user interactions, creates sub-queries from the search phrase, and evaluates sub-query models to select the one with the highest search phrase score that meets a threshold, allowing for the use of sub-query models to rank images even if a query-specific model is not available.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a query-specific image relevance model is used, then image ranking accuracy is improved, but model availability decreases for diverse search phrases
Solution Approach 1:
The patent segments a complex multi-term search query into multiple sub-queries, each corresponding to a trained image relevance model. For example, the query 'birthday cakes for boys' is divided into sub-queries like 'birthday cakes' and 'boys', allowing the system to leverage pre-trained models for each sub-query even when no model exists for the complete original query.
Solution Approach 2:
The patent makes image relevance models universally applicable across different search phrases by enabling a model trained for a sub-query to serve multiple parent queries. A single model trained for 'birthday cakes' can be reused across various queries containing this sub-query, eliminating the need to train separate models for every possible search phrase combination.
2Adaptability or versatility
If sub-query models are used for search phrases without available models, then model availability is improved, but ranking precision may deteriorate
Solution Approach 1:
The patent implements feedback by comparing the ranking produced by a sub-query model against actual user interaction data (clicks, views, saves). This feedback loop allows the system to evaluate whether a sub-query model's ranking aligns with user preferences, and to adjust or retrain models based on this performance feedback to maintain ranking precision.
Solution Approach 2:
The patent makes the model selection and ranking process dynamic by allowing the system to choose different sub-query models based on the specific search phrase and to adjust rankings based on real-time user interaction feedback. The system can dynamically switch between different sub-query models or combine their outputs depending on which provides the best match for the given query.
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for identifying images responsive to a search phrase are disclosed. In one aspect, a method includes identifying a set of responsive images for a search phrase that includes two or more terms. Interaction rankings are determined for images in the set of responsive images. Two or more sub-queries are created based on the search phrase. Sub-query model rankings are determined for images in the set of responsive images. A search phrase score is determined for the image relevance model. Based on the search phrase scores for the sub-queries, one of the sub-query models is selected as a model for the search phrase.