Web Host Clustering for Abnormal Search Visit Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing search engines face challenges in detecting abnormal site visits, which can artificially inflate the ranking of web hosts, leading to user dissatisfaction due to the influence of malicious actors instructing users to submit specific queries and select certain web hosts.
Innovation Solution
A method and system that analyze web search history by generating a graph of web hosts connected by edge weights representing user visits, identifying themes, and clustering nodes to filter out naturally affiliated hosts, then using user interaction data to determine if clusters are affiliated, thereby identifying and reducing the ranking of web hosts with abnormal site visits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If web search history is used to rank search results, then the relevance of search results is improved, but abnormal site visits can artificially inflate rankings leading to user dissatisfaction
Solution Approach 1:
The patent segments web hosts into different clusters based on their affiliation relationships. By dividing the web host space into clusters (e.g., naturally affiliated vs. abnormally affiliated), the system can apply different ranking treatments to different segments, thereby maintaining ranking integrity while preserving useful search history information.
Solution Approach 2:
The patent applies local quality by treating different web hosts differently based on their cluster membership. Web hosts identified as abnormally affiliated receive localized penalty treatments (such as reduced ranking weight or exclusion), while other web hosts continue to benefit from search history-based ranking improvements.
2Reliability
If graph clustering is used to identify affiliated web hosts, then abnormal site visits can be detected, but the system complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-computing the affiliation graph and clusters before the actual search ranking process. Web hosts are pre-segmented into clusters based on their affiliation relationships, and this clustering information is stored for efficient retrieval during ranking, avoiding the need to perform complex graph computations in real-time.
Solution Approach 2:
The patent introduces an intermediary component (the affiliation graph and cluster structure) that mediates between raw search history data and the final ranking process. This intermediary pre-processes and organizes the data, making the subsequent ranking operations simpler and more efficient.
Data Source
AI summary
There is disclosed a method and system for determining web hosts receiving abnormal site visits. The method comprises generating a graph of web search history and clustering nodes in the graph. The method then comprises removing clusters that are affiliated based on user interaction data, and storing indicators that the remaining web hosts are associated with abnormal site visits.