Research On Network Entity Landmark Mining Technology

Network entity geolocation has a wide application prospect in network security and national security,and can be used for protecting social network privacy,tracing network attack and intrusion detection.Reliable network entity geolocation requires the support of a large number of network entity landmarks with high location accuracy,stable performance and wide distribution.However,the current network entity landmark mining technology has a series of problems,e.g.,the limited number of candidate landmarks,the low accuracy of Web server-based reliable landmarks,cannot define the error bound of reliable landmarks,and the insufficient geolocation ability of single source landmarks.In order to solve these problems which restrict the improvement of geolocation effect,this dissertation focuses on the street-level landmark acquisition,street-level landmark evaluation and multi-source landmark fusion.The main work and innovations are as follows:1.Because online map-based landmark mining algorithms are constrained by the online map service itself,a street-level landmark mining algorithm based on radar search is proposed.Initially,the region in which landmarks will be mined is divided into square sub-regions of appropriate size.Then,the radar search service of an online map is used to perform a recursive query request for each sub-region with the granularity dynamically adjusted,and finally all the street-level candidate landmarks in the sub-region are obtained.Landmark mining experiments in Taiwan and Hong Kong show that the algorithm has significant better performance on number,distribution,and coverage area of landmark,based on the existing method.2.Considering existing street-level landmark evaluation methods have low accuracy and strict constraints,this dissertation analyzes the causes and evaluation idea of invalid Web-based candidate landmarks,and proposes Evaluator,a Web-based landmark evaluation approach.Evaluator adopts the idea of decision tree to filter invalid landmarks layer by layer,and comprehensively assesses candidate landmarks with public data and services to obtain reliable landmarks with quantitative reliability.This dissertation proposes DNS distributed query algorithm to effectively resolve all IP addresses of a domain name,which provides data support for Evaluator to filter candidate landmarks.Meanwhile,this dissertation also proposes Reverse verification algorithm to obtain all domain names of an IP address,which provides an important reference to calculate the reliability of a reliable landmark.In addition,gradient descent is used to estimate the evaluation parameters,which effectively improves the robustness of Evaluator.Evaluation and Geolocation experiments of 5 cities in China and the US show that Evaluator can effectively exclude invalid candidate landmarks,and significantly improves the evaluation coverage and geolocation accuracy based on the latest landmark evaluation method.3.The current street-level landmark evaluation methods,which just give a relative reliablity value,cannot determine the error range of reliable landmarks.Therefore,a street-level landmark evaluation algorithm that can estimate the upper bound of landmark error is proposed.Firstly,the city of candidate landmarks are verified by IP location databases.Secondly,candidate landmarks are grouped through their last-hop routers,and then divided into several clusters by E-Apriori algorithm based on their location.Thirdly,the Land-mark reliability probability model is used to calculate the probability of the last-hop router location range.Finally,the upper error bound of the landmark is determined by the locatoin range of the last-hop router.The landmark evaluation experiments in Hong Kong and Beijing show that the proposed algorithm can effectively select reliable landmarks and determine the upper error bound of the landmark.The reliable landmarks obtained by the algorithm significantly improve the geolocation effect of landmarks based on the latest methods.4.Due to the limited number of landmarks and sometimes the low accuracy for a single landmark data source,it is difficult to ensure the geolocation effect.In order to integrate landmarks from multiple sources and improve the geolocation effect,this dissertation proposes Lusion,a multi-source landmark fusion algorithm.First,we perform data preprocessing on the landmark sources to ensure the coincident granularity of the landmarks is a data source as well as the independence between the data sources,for the purpose of meeting the integration requirements.Second,we extend the IP addresses in the landmark data sources to improve the fusion effect.Third,we model the location data of the street-level and city-level landmarks using the landmark location mixture model.Finally,we use the expectation maximization algorithm to estimate the location of the landmarks.Experment with simulated data and that with real world data in Hong Kong and Zhengzhou respectively show that the algorithm can effectively fuse city-level and street-level landmark data sources.The accuracy and geolocation effect of landmarks from Lusion are significantly improved based on the original landmark data set.