Research On User Characteristics Modeling In Crowdsensing Based On Implicit Feedback And Data Quality

In recent years,with the popularization and development of mobile terminals and sensing technology,crowdsensing has been widely and effectively used in a large number of fields.Task allocation and user selection are important functions of the crowdsensing platform.At present,many research works have designed different scenarios and proposed effective task matching schemes,which have improved the service quality of the crowdsensing platform.However,most of these studies are modeled from the perspective of the platform,which usually ignore the characteristics of users themselves,especially the preference characteristic and reliability characteristic.In real scenarios,these two characteristics often determine the user’s experience of participating in tasks,whether the system can quickly recruit enough users and the quality of task completion,therefore,it is necessary to accurately describe user preference and reliability.This thesis mainly studies the modeling of preference characteristic and reliability characteristic of crowdsensing users and introduces these two characteristics into the user recruitment model.Crowdsensing system stores the implicit feedback information of the interaction between users and tasks.Combined with the analysis of the recommender system,the user’s preference for tasks can be extracted from the implicit feedback.The reliability characteristic of users depends on the quality of data submitted by users,so adopting a ground truth discovery model is an effective solution.The work of this thesis is to improve the existing recommendation system and ground truth discovery model based on implicit feedback and data quality respectively,so that it can perform better in crowdsensing scenarios and more accurately describe user preference and reliability characteristics,and then build a characteristic-driven user recruitment model.The thesis mainly includes the following contents:1)The modeling of user preference characteristic based on implicit feedback.By using the metric learning method,the three types of implicit feedback which contain the browsing and participation of tasks are firstly analyzed,and the metric learning samples are represented as quadruples.Considering that the users’ preference focuses on different aspects such as task location,time and type,a metric learning method based on the weighting of task labels is designed,thus the measurement function is obtained.Designing a metric learning loss function based on quadruple samples.Quantitatively analyzing the breadth of user preferences at the level of task participation and task interaction and establishing a dynamic sample spacing mechanism for metric learning.Designing a regularization strategy based on multi-level distance to avoid the overfitting of metric learning.Designing experiments to study the parameter setting of the model,the effectiveness of submodules,and the superiority of the model.2)The modeling of user reliability characteristic based on data quality.Building a basic data quality driven truth discovery model.Considering the spatiotemporal correlation of truths,the result of the last truth discovery is used as one of the effective data sources if the number of users is sparse.Introducing tasks with known truths,setting the reliabilities of users in the initial state of the system,and setting the reliabilities of newly added users in the middle,synchronously shrinking the reliability weights of other users.Combined with the previous truth and the completion of the tasks with known truths,an optimized truth discovery model is established,and the Lagrange multiplier method is used to derive the iterative formulas of truths and reliabilities.Designing experiments to verify the accuracy advantages of the model under different parameter settings.3)Crowdsensing user recruitment research driven by preference and reliability characteristics.Designing multi-round user recruitment mechanism in crowdsensing.The user preference and reliability characteristics are normalized,combined with the time limit and the number of remaining recruits,an optimization model that maximizes preference with reliability constraint is established.The model is transformed into a 0-1 knapsack problem with multiple knapsacks and it is solved by using the artificial fish swarm algorithm.Designing experiments to verify the performance advantages of the model in terms of recruitment efficiency and task completion quality,proving the necessity and advancement of studying user characteristics.