Crowdsourcing Data Application Based On User Perception In Social Network

Social network provides a platform for users with similar interests or varied expertise to solve specific problems by using collective thinking.Crowdsourcing system can be regarded as a small social network with clear objectives.Based on common interest or public good,users come together voluntarily to contribute perception experience,and exploit the social functionality to achieve common objectives.Common objectives include retrieving hot events based on collective intelligence,solving large-scale problems with distributed perceptual data sources,etc.The crowdsourcing model based on user perception in social network is the combination of social network and crowdsourcing perception,which meets the characteristics of open collaboration crowdsourcing.Based on the characteristics,this thesis regards user communities as crowdsourcing perceptual data sources,and distinguishs the contributed perception data into three layers.The basic layer corresponds to social interaction oriented entity information.The key layer corresponds to collaborative perception oriented sensing data.The extended layer corresponds to target domain oriented event description.Three levels of crowdsourcing data are of great significance for user characteristic extraction,high-quality data collection,hot event tracking and so on.The main contributions of this thesis are highlighted as below:(1)Cross-network entity alignment based on crowdsourcing entity information.Crowdsourcing entity information include heterogeneous attribute information and associated structure information,and can be used to identify the implicit correspondence between cross-network entities,i.e.,matching user accounts of the same natural person across networks.The thesis proposes a Matrix Factorization based Representation learning(MFRep)model to study the multi-entity alignment problem,including cross-network users and associated entities.MFRep considers cross-network structure difference in real scenarios,and infers multi-entity correspondences simultaneously in a mutually reinforcing way.Cross-network entity alignment based on crowdsourcing entity information facilitates building a comprehensive user profile.(2)Crowdsourced team formation based on crowdsourcing sensing data.Users’ online footprints retrieved from crowdsourcing entity information can act as the clue to collect historical sensing data contributed by user communities.Driven by swarm intelligence applications,crowdsourcing sensing data include numerical data such as environmental information perception,and category data such as consumer survey/product review.This thesis proposes a social team crowdsourcing mechanism QEM-Crowd(Quality Estimation Model for team Crowdsourcing),including a sensing quality estimation model named QEM and a task allocation algorithm named CSSelection.Based on contribution estimation of different users on crowdsourcing tasks,CS-Selection greedily selects multiple users with varied expertise for crowdsourced team formation.By modeling task valuation maximization problem as a Submodular Cost Submodular Knapsack(SCSK)problem,CS-Selection guarantees team connectivity and achieves the near-optimal solution of the submodular optimization problem.(3)Event detection and tracking based on crowdsourcing event description.Crowdsourcing sensing data contributed by team members can be viewed as text description information of task relevant topics/events.This,along with topic discussion forms crowdsourcing event description information.Social network users can be regarded as crowdsourcing sensors of social hot events.Crowdsourcing event descriptions include the detection indicators of hot events,the content changes of events and the temporal and spatial characteristics of events.To deal with continuously-arriving event descriptions,this thesis proposes an event detection and tracking framework CyberEM(Event Evolution Model with respect to Cybersecurity domain).CyberEM exploits semantically coherent linguistic structure to characterize and detect hot events,and designs event aggregation algorithm to track potentially evolutionary events.In conclusion,this thesis comprehensively discusses open collaboration crowdsourcing model based on social network user perception and its application value.It provides a new way to solve the specific application problems based on the collective thinking of social network users.