Search Result Ranking via Query Embedding Similarity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engine ranking algorithms struggle to effectively rank search results for new queries based on similarity to past queries, leading to suboptimal relevance and user satisfaction.
Innovation Solution
A method and system utilizing a Deep Structured Semantic Model (DSSM) to generate features and training data, which computes the closeness of new search queries to past queries, allowing for the generation of additional ranking features based on past user interactions, and using these features in algorithms like MatrixNet to rank search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional ranking algorithms are used to rank search results, then the system is simpler to implement, but the relevance and user satisfaction of search results deteriorates
Solution Approach 1:
The system pre-computes and stores query embeddings and similarity metrics in advance during training phases. When a new query arrives, the pre-trained models immediately generate relevance scores without requiring complex real-time computations, thus improving response accuracy while maintaining system efficiency
Solution Approach 2:
The patent introduces query embeddings as an intermediary representation that bridges the gap between raw search queries and search results. These embeddings capture semantic meaning and enable the ranking system to compute relevance without directly comparing complex query-result pairs, simplifying the ranking process while improving accuracy
2Adaptability or versatility
If query similarity based on past queries is not used, then the ranking process is faster, but the ability to handle new queries effectively deteriorates
Solution Approach 1:
The system pre-trains models on historical query data to learn query similarity patterns and relationships. This preliminary training enables the system to quickly adapt to new queries using the learned patterns, avoiding the need for slow real-time analysis of past queries while maintaining high adaptability
Solution Approach 2:
The patent creates embedding representations that capture the essential semantic features of queries. These embeddings serve as compressed copies of query meanings, allowing the system to efficiently compare new queries against historical patterns without processing the full complexity of original query texts, thus maintaining speed while improving adaptability
Data Source
AI summary
A method and a system for ranking search results using a machine learning algorithm (MLA), the MLA having been trained to determine query similarity based on textual content thereof, comprising: receiving a new query, retrieving a plurality of search queries, computing a respective similarity parameter between the new query and each query of the plurality of search queries, selecting a given past query associated with a highest respective similarity parameter, retrieving a set of search results associated with the given past query, each of the set of search results being associated with an annotation including respective search queries, and computing for each respective one of the plurality of respective search queries having been used to access the respective search result, a respective second similarity parameter, and using the respective second similarity parameter as a ranking feature for ranking the associated search results as search results for the new query.