Search Result Ordering via Bayesian Relevancy Scoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engine technologies fail to accurately consider the position of information in search result lists and click sequences when ranking search results, leading to poor ranking and increased workload on search engine servers, resulting in wasted system resources.
Innovation Solution
A method and apparatus that compute Bayesian posterior probabilities to determine the expected relevance of search results based on their positions and click sequences, storing these values in a data structure and ordering them in descending order to improve ranking accuracy and reduce server workload.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If search engines use CTR (Click-Through-Rate) to rank search results, then the ranking process is simple and fast, but the relevancy scores have large variance and desired information may be placed near the end of the search result list
Solution Approach 1:
The patent transforms the single-parameter CTR metric into a multi-parameter relevancy scoring system that incorporates position information and click sequence data. By changing from using only CTR to using expected relevancy scores that consider multiple factors, the system improves measurement precision while maintaining computational efficiency through the use of pre-computed probabilities and efficient query processing.
2Ease of operation
If search results are poorly ranked, then users need to browse and click on more information to find desired information, but this increases workload of network systems and reduces utilization rate of network systems
Solution Approach 1:
The patent pre-computes expected relevancy scores and stores them in data structures during idle periods or batch processing times. When users submit search queries, the system quickly retrieves pre-computed results rather than performing extensive real-time computations. This preliminary action reduces both user wait time and real-time server workload, effectively addressing the contradiction between user ease of operation and system resource consumption.
3Reliability
If users conduct extensive browsing and clicking to find desired information, then users can obtain their most desired information, but the cost of the search process for the search engine server increases and system resources are wasted
Solution Approach 1:
The patent replaces the mechanical browsing and clicking process with an intelligent ranking system that uses pre-computed expected relevancy scores. Instead of relying on users to manually browse through poorly ranked results, the system substitutes this mechanical process with automated, data-driven ranking that presents relevant information in the optimal order, thereby reducing both user effort and server resource consumption while maintaining high information retrieval success rates.
Data Source
AI summary
Ordering search results may include obtaining an exposed log file from a log system, computing a Bayesian posterior probability for relevancy between the log file and a search request, computing an expected value of the relevancy between the log file and the search request based on the Bayesian posterior probability, storing the search request and an identifier of the log file as a key and the expected value of the relevancy between the log file and the search request as a value into a search data structure, in response to receiving a search request submitted by a user, finding expected values of relevancy between the submitted search request and log files that are relevant to the submitted search request from the search data structure, and ordering the found log files in a descending order of the expected values.