Jiali Mao

TSCluWin: Trajectory Stream Clustering over Sliding Window

The popularity of GPS-embedded devices facilitates online monitoring of moving objects and analyzing movement behaviors in a real-time manner. Trajectory clustering acts as one of the most important trajectory analysis tasks, and the researches in this area have been studied extensively in the recent decade. Due to the rapid arrival rate and evolving feature of stream data, little effort has been devoted to online clustering trajectory data streams. In this paper, we propose a framework that consists of two phases, including a micro-clustering phase where a number of micro-clusters represented by compact synopsis data structures are incrementally maintained, and a macro-clustering phase where a small number of macro-clusters are generated based on micro-clusters. Experimental results show that our proposal is both effective and efficient to handle streaming trajectories without compromising the quality.

TrajSpark: A Scalable and Efficient In-Memory Management System for Big Trajectory Data

The widespread application of mobile positioning devices has generated big trajectory data. Existing disk-based trajectory management systems cannot provide scalable and low latency query services any more. In view of that, we present TrajSpark, a distributed in-memory system to consistently offer efficient management of trajectory data. TrajSpark introduces a new abstraction called IndexTRDD to manage trajectory segments, and exploits a global and local indexing mechanism to accelerate trajectory queries. Furthermore, to alleviate the essential partitioning overhead, it adopts the time-decay model to monitor the change of data distribution and updates the data-partition structure adaptively. This model avoids repartitioning existing data when new batch of data arrives. Extensive experiments of three types of trajectory queries on both real and synthetic dataset demonstrate that the performance of TrajSpark outperforms state-of-the-art systems.

HyMU: A Hybrid Map Updating Framework

Accurate digital map plays an important role in mobile navigation. Due to the ineffective updating mechanism, existing map updating methods cannot guarantee completeness and validity of the map. The common problems of them involve huge computation and low precision. More importantly, they scarcely consider inferring new roads on sparse unmatched trajectories. In this paper, we first address the issue of finding new roads in sparse trajectory area. On the basis of sliding window model, we propose a two-phase hybrid framework to update the digital map with inferred roads, called HyMU, which takes full advantage of line-based and point-based strategies. Through inferring road candidates for consecutive time windows and merging the candidates to form missing roads, HyMU can even discover new roads in sparse trajectory area. Therefore, HyMU has high recall and precision on trajectory data of different density and sampling rate. Experimental results on real data sets show that our proposal is both effective and efficient as compared to other congeneric approaches.

Feature Grouping-Based Outlier Detection Upon Streaming Trajectories

Outlier detection acts as one of the most important analysis tasks for trajectory stream. In stream scenarios, such properties as unlimitedness, time-varying evolutionary, sparsity, and skewness distribution of trajectories pose new challenges to outlier detection technique. Trajectory outlier detection techniques mainly focus on finding trajectory that is dissimilar to the majority of the others, which is based on the hypothesis that they are probably generated by a different mechanism. Most distance-based methods tend to utilize a function (e.g., weighted linear sum) to measure the similarity of two arbitrary objects provided that representative features have been extracted in advance. However, this kind of method is not tailored to identify the outlier which is close to its neighbors according to some features, but behaves significantly different from its neighbors in terms of the other features. To address this issue, we propose a feature grouping-based mechanism that divides all the features into two groups, where the first group (Similarity Feature) is used to find close neighbors and the second group (Difference Feature) is used to find outliers within the similar neighborhood. According to the feature differences among local adjacent objects in one or more time intervals, we present two outlier definitions, including local anomaly trajectory fragment (TF-outlier) and evolutionary anomaly moving object (MO-outlier ). We devise a basic solution and then an optimized algorithm to detect both types of outliers. Experimental results show that our proposal is both effective and efficient to detect outliers upon trajectory data streams.

DT-KST: Distributed Top-k Similarity Query on Big Trajectory Streams

During the past decade, with the widespread use of smartphones and other mobile devices, big trajectory data are generated and stored in a distributed way. In this work, we focus on the distributed top-k similarity query over big trajectory streams. Processing such a distributed query is challenging due to the limited network bandwidth. To overcome this challenge, we propose a communication-saving algorithm DT-KST (Distributed Top-K Similar Trajectories). DT-KST utilizes the multi-resolution property of Haar wavelet, and devises a level-increasing communication strategy to tighten the similarity bounds. Then, a local pruning strategy is imported to reduce the amount of data returned from distributed nodes. Theoretical analysis and extensive experiments on a real dataset show that DT-KST outperforms the state-of-the-art approach in terms of communication cost.

Cloned Vehicle Behavior Analysis Framework

Cloned vehicles brought tremendous harm to transportation management and public safety, which necessitates an efficient detection mechanism to discern the behaviors of cloned vehicles. The ubiquitous inspection spots deployed in the city have been collecting moving information of passing vehicles. Thus the positional sequences of inspection spots that vehicles passed by could form into their travelling traces. This provides us unprecedented opportunity to detect cloned vehicles. In this paper, we first propose a framework to discern the behaviors of cloned vehicles, called CVAF. It consists of three parts, including cloned vehicle detection, trajectory differentiation using matching degree-based clustering, and behavior pattern extraction. The experimental results on the real-world data show that our CVAF framework can identify cloned vehicle and discern their behavior patterns effectively. Our proposal can assist traffic control and public security department to solve the crime of cloned vehicle.

MDTK: Bandwidth-Saving Framework for Distributed Top-k Similar Trajectory Query

During the past decade, with the popularity of smartphones and other mobile devices, big trajectory data is generated and stored in a distributed way. In this work, we focus on the DTW distance based top-k query over the distributed trajectory data. Processing such a query is challenging due to the limited network bandwidth and the computation overhead. To overcome these challenges, we propose a communication-saving framework MDTK (Multi-resolution based Distributed Top-K). MDTK sends the bounding envelopes of the reference trajectory from coarse to finer-grained resolutions and devises a level-increasing communication strategy to gradually tighten the proposed upper and lower bound. Then, distance bound based pruning strategies are imported to reduce both the computation and communication cost. Besides, we embed techniques including: indexing, early-stopping and cascade pruning, to improve the query efficiency. Extensive experiments on real datasets show that MDTK outperforms the state-of-the-art method.

Online clustering of streaming trajectories

With the increasing availability of modern mobile devices and location acquisition technologies, massive trajectory data of moving objects are collected continuously in a streaming manner. Clustering streaming trajectories facilitates finding the representative paths or common moving trends shared by different objects in real time. Although data stream clustering has been studied extensively in the past decade, little effort has been devoted to dealing with streaming trajectories. The main challenge lies in the strict space and time complexities of processing the continuously arriving trajectory data, combined with the difficulty of concept drift. To address this issue, we present two novel synopsis structures to extract the clustering characteristics of trajectories, and develop an incremental algorithm for the online clustering of streaming trajectories (called OCluST). It contains a micro-clustering component to cluster and summarize the most recent sets of trajectory line segments at each time instant, and a macro-clustering component to build large macro-clusters based on micro-clusters over a specified time horizon. Finally, we conduct extensive experiments on four real data sets to evaluate the effectiveness and efficiency of OCluST, and compare it with other congeneric algorithms. Experimental results show that OCluST can achieve superior performance in clustering streaming trajectories.

Discovering Underground Roads from Trajectories Without Road Network

With the wide application of GPS-enabled electronic devices, huge amounts of positional information data have been accumulated, so that it’s critical to discover inherent knowledge from such massive data. In this paper, we address this topic by proposing two issues, including how to discover the underpasses for pedestrians to cross the roads, and how to discover the tunnels providing passageways for vehicles. Subsequently, we propose a three-step framework to deal with the issues, including an incremental clustering phase, a sub-trajectory detecting phase and a cluster filtering phase. Experiments upon real-life data sets demonstrate the effectiveness and efficiency of the proposed framework.

Learning User Credibility on Aspects from Review Texts

Spammer detection has been popularly studied these years which aims at filtering unfair or incredible customers. Most users have different backgrounds or preferences so that they make distinct reviews/ratings, however they can not be treated as spammers. To date, the existing previous spammer detection technology has limited usability. In this paper, we propose a method to calculate user credibility on multi-dimensions by considering users difference related to their personalities e.g. background and preference. Firstly, we propose to evaluate customer credibilities on aspects with the consideration of different concerns given by different customers. A boot-strapping algorithm is applied to detect the intrinsic aspects of review text and the aspect ratings are assigned by mining semantic polarity. Then, an iteration algorithm is designed for estimating credibilities by considering the consistency between individual ratings and overall ratings on aspects. Finally, experiments on the real dataset demonstrate that our method outperforms baseline systems.