A Study On Effective Spatio-textual Data Integration And Delivery

With the the growing popularity of mobile devices(e.g.smartphones)and the rapid development of mobile Internet technology,location-based services have been widely deployed and well accepted.These services generate large amounts of spatio-textual data,which contain both geographical location and respective textual description.The key factors to take advantage of them to benefit our daily life are two-fold: first,data quality,as high-quality data bring better user experience;second,the way of accessing the data,user should feel convenient to obtain their interested data.In this dissertation,we start the study from the above two aspects: on one hand,we improve the data quality,which aims to improve the data accuracy as well as reduce its redundancy;on the other hand,we study how to efficiently deliver spatio-textual information to relevant users and meet users' different requests.In particular,the contents and contributions of the dissertation include:(1)Crowdsourcing based spatio-textual data refining: to improve the data accuracy,a crowdsourcing based method is employed to help select correct words(tokens)for data.We formalize the crowdsourcing based framework to handle the problem.First,we develop an effective inference model which utilizes the spatial information of workers and points of interests(POIs)to measure the worker quality and the POI influence in a finer-granularity,and utilize it to infer results.Based on the inference model,we propose an adaptive task assignment algorithm to further improve the inference accuracy,which assigns a group of tasks for each available worker by maximizing the accuracy improvement.(2)Top-k spatio-textual data integration: an effective method of top-k data integration is developed to reduce the data redundancy.Traditional methods merely consider the optimization on spatial and textual components at the same time.To address the problem,we propose a spatio-textual signature-based top-k similarity join framework,which efficiently prunes irrelevant pairs of data through signatures.We find the order of accessing the signatures has a significant effect on the performance.Thus we propose a best-first method that preferentially accesses signatures with large upper bounds.It accelerates the process of generate top-k result.We further optimize the spatio-textual signatures and propose extend signatures to improve the pruning power.(3)Spatio-textual data delivery: to meet users' different requests of delivery,we formulate and study the parameterized delivery problem for spatio-textual information.To address the problem,we propose the spatial-oriented prefix and devise a filter-verification framework.Next we present the region-aware prefix which utilizes hierarchical spatial index to reduce the prefix size and supports region pruning.Then we seamlessly integrate multi-words prefix filtering techniques into the framework and propose the spatio-textual prefix to further enhance pruning power.