Towards Online Spam Filtering in Social Networks Background Background Another Study in Spam Detection Goals in addition to Existing Work

Towards Online Spam Filtering in Social Networks Background Background Another Study in Spam Detection Goals in addition to Existing Work

Towards Online Spam Filtering in Social Networks Background Background Another Study in Spam Detection Goals in addition to Existing Work

Shuman, Sidney, Contributing Editor has reference to this Academic Journal, PHwiki organized this Journal Towards Online Spam Filtering in Social NetworksHongyu Gao, Yan Chen, Kathy Lee, Diana Palsetia in addition to Alok ChoudharyLab as long as Internet in addition to Security Technology (LIST)Department of EECSNorthwestern UniversityBackground2Background3

King's College PA

This Particular University is Related to this Particular Journal

4Another Study in Spam DetectionUnique characteristics of OSNsAre existing features still effectiveNumber of wordsAverage word lengthSender IP neighborhood densitySender AS numberStatus of sender’s service ports Any new featuresNot effective!5Goals in addition to Existing WorkAn ef as long as t towards a system ready to deployExisting studies in OSN spam:[Gao IMC10, Grier CCS10] offline analysis [Thomas Oakl in addition to 11] l in addition to ing page vs. message contentNumerous work in spammer-faked account detectionOnline detectionHigh accuracyLow latencyDetection of campaigns absent from training setNo need as long as frequent re-training66Detection System DesignEvaluationConclusions & Future WorkRoadmap

77We Do NOT: Inspect each message individually Key Intuitionmsg-1msg-2msg-3msg-n88We Do: Inspect correlated message clustersKey Intuitionmsg-k9System OverviewDetect coordinated spam campaigns.

10Incremental ClusteringRequirement:Given (k+1)th message in addition to result of the first k messagesEfficiently compute the result of the (k+1) messagesAdopt text shingling techniquePros: High efficiencyCons: Syntactic method11Feature SelectionFeature selection criteria:Cannot be easily maneuvered.Grasp the commonality among campaigns.6 identified features:Sender social degreeInteraction historyCluster sizeAverage time intervalAverage URL Unique URL 1212Detection System DesignEvaluationConclusions & Future WorkRoadmap

1313All experiments obey the time orderFirst 25% as training set, last 75% as testing set.Evaluated metrics:Dataset in addition to MethodOverall accuracyAccuracy of feature subsetAccuracy over timeAccuracy under attackLatencyThroughput1414Best resultFB: 80.9% TP 0.19%FPTW: 69.8%TP 0.70%FPOverall Accuracy1515No significant drop of TP or increase of FPAccuracy over Time

1616Latency1717Detection System DesignEvaluationConclusions & Future WorkRoadmap18ConclusionsWe design an online spam filtering system based on spam campaigns.Syntactical incremental clustering to identify message clustersSupervised machine learning to classify message clustersWe evaluate the system on both Facebook in addition to Twitter data187M wall posts, 17M tweets80.9% TPR, 0.19% FPR, 21.5ms mean latencyPrototype release:

1919Cool , I by no means noticed anyone do that prior to . {URL}Wow , I in no way noticed anyone just be as long as e . {URL}Amazing , I by no means found people do that just be as long as e . {URL} Future Work{Cool Wow Amazing} + , I + {by no means in no way} +{noticed found} + {anyone people} + {do that } + {prior to just be as long as e} + . {URL}Template generationCall as long as semantic clustering approaches20Thank you!21ContributionsDesign an online spam filtering system to deploy as a component of the OSN plat as long as m. High accuracyLow latencyTolerance as long as incomplete training dataNo need as long as frequent re-trainingRelease the system

22Incremental Clusteringshingle-1shingle-2shingle-3 msg-11msg-13msg-21msg-22msg-23msg-31msg-33msg-32msg-12 msg-newCompare in addition to Insert23Sender Social DegreeCompromised accounts:The more edges, with a higher probability the node will be infected quickly by an epidemic.Spammer accounts:Social degree limits communication channels.Hypothesis:Senders of spam clusters have higher average social degree than those of legitimate message clusters.24Sender Social DegreeAverage social degree of spam in addition to legitimate clusters, respectively.

Shuman, Sidney GamePro Contributing Editor

25Interaction HistoryLegitimate accounts:Normally only interact with a small subset of its friends.Spamming accounts:Desire to push spam messages to as many recipients as possible.Hypothesis:Spam messages are more likely to be interactions between friends that rarely interact with be as long as e.26Interaction HistoryInteraction history score of spam in addition to legitimate clusters, respectively.2727Scalability300M tweets/dayMap-reduce style in addition to cloud computingOther Thoughts

Shuman, Sidney Contributing Editor

Shuman, Sidney is from United States and they belong to GamePro and they are from  San Francisco, United States got related to this Particular Journal. and Shuman, Sidney deal with the subjects like Consumer Video; Interactive Media; Video/Computer Games

Journal Ratings by King’s College

This Particular Journal got reviewed and rated by King’s College and short form of this particular Institution is PA and gave this Journal an Excellent Rating.