SoC PhD Forum 2014

Poster Abstracts

^M ^M




Data Mining & Database



Poster Presenter: Wang Xiaoli (

Title: Efficient 3-in-1 inverted index and its real applications

In traditional relational databases, data are modeled as tables. However, most real life data cannot be simply modeled as tables, but as complex structures like sequences, trees and graphs. Existing systems typically cater to the storage of complex structures separately, which can result in a waste of resources. Moreover, many applications may require the storage of various complex structures, and it is not easy to adapt existing systems to support such applications. In this dissertation, we aim to develop a unified framework, denoted by 3-in-1, that can support the efficient storage and retrieval of various complex structures (i.e., sequences, trees, and graphs).



Poster Presenter: Jovian Lin (

Title: Addressing Cold-Start in App Recommendation: Latent User Models Constructed from Twitter Followers

As a tremendous number of mobile applications (apps) are readily available, users have difficulty in identifying apps that are relevant to their interests. Recommender systems that depend on previous user ratings (i.e., collaborative filtering, or CF) can address this problem for apps that have sufficient ratings from past users. But for apps that are newly released, CF does not have any user ratings to base recommendations on, which leads to the cold-start problem. In this paper, we describe a method that accounts for nascent information culled from Twitter to provide relevant recommendation in such cold-start situations. We use Twitter handles to access an app's Twitter account and extract the IDs of their Twitter-followers. We create pseudo-documents that contain the IDs of Twitter users interested in an app and then apply latent Dirichlet allocation to generate latent groups. At test time, a target user seeking recommendations is mapped to these latent groups. By using the transitive relationship of latent groups to apps, we estimate the probability of the user liking the app. We show that by incorporating information from Twitter, our approach overcomes the difficulty of cold-start app recommendation and significantly outperforms other state-of-the-art recommendation techniques.



Poster Presenter: Le Thuy Ngoc (

Title: Semantics for XML keyword search

Since XML has become a standard for information exchange over the Internet, more and more data are represented as XML. XML keyword search has been attracted a lot of interests because it provides a simple and user-friendly interface to query XML documents. Existing approaches for XML keyword search can be classified into two types: tree-based approaches and graph-based approaches based on whether the considered XML document contains ID/IDREFs or not. The tree-based approaches are for XML documents with no ID/IDREF and mainly follow the Lowest Common Ancestor (LCA) semantics (and thus they are also called LCA-based approaches), while the graph-based approaches are for XML documents with ID/IDREFs and apply the Steiner tree semantics. These tree-based and graph-based approaches work well for certain types of XML documents. However, since these approaches only rely on the structure of XML documents but do not consider the semantics of Objects, Relationships among objects, Attributes of objects, and Attribute of relationships (referred to as ORA-semantics), they may suffer from several problems, including meaningless answers, missing answers, duplicated answers, and schema dependence. Therefore, we propose to use ORA-semantics for XML keyword search to address the above problems.  We classify nodes in XML documents into different types such as object class, object identifier (OID), object attribute, relationship attribute, etc. ORA-semantics provides the type of each node in XML data. Based on ORA-semantics, we can first distinguish an object node from an arbitrary node in XML data, e.g., attribute and value. Then we can detect whether the two object nodes referring to the same object based on object class and OID.  These identifications enable us to solve the above problems of the existing approaches.



Poster Presenter: Xiao Liu (

Title: Differentially Private Recommendation Systems

Recommendation systems are becoming increasingly important as a tool to manage information overload. In a collaborative recommendation system, however, privacy may be a concern, as users' opinions are used to generate recommendations for others. We consider an interactive recommendation system model in which users cast votes on objects, and the system repeatedly recommends objects and then collects feedback on its recommendations. We consider differential privacy and investigate the trade-off between the quality of the recommendations and the privacy of the users for any (epsilon, delta)-differentially private recommendation algorithm. In particular: i) We show a lower bound for the privacy of any non-trivial collaborative recommendation system. ii) We propose an (epsilon, delta)-differentially private recommendation algorithm with recommendation quality within a constant factor of optimal, and with privacy parameters epsilon and delta (almost) within constant factors of optimal.



Poster Presenter: Chen Tao (

Title: Understanding and Classifying Image Tweets

Social media platforms now allow users to share images alongside their textual posts. These image tweets make up a fast-growing percentage of tweets, but have not been studied in depth unlike their text-only counterparts. We study a large corpus of image tweets in order to un- cover what people post about and the correlation between the tweet’s image and its text. We show that an impor- tant functional distinction is between visually-relevant and visually-irrelevant tweets, and that we can successfully build an automated classifier utilizing text, image and social con- text features to distinguish these two classes, obtaining a macro F1 of 70.5%.





Poster Presenter: Gan Tian (

Title: Temporal Encoded F-formation System for Social Interaction Detection

In the context of a social gathering, such as a cocktail party, the memorable moments are generally captured by professional photographers or by the participants. The latter case is often undesirable because many participants would rather enjoy the event instead of being occupied by the photo-taking task. Motivated by this scenario, we propose the use of a set of cameras to automatically take photos. Instead of performing dense analysis on all cameras for photo capturing, we first detect the occurrence and location of social interactions via F-formation detection. In the sociology literature, F-formation is a concept used to define social interactions, where each detection only requires the spatial location and orientation of each participant. This information can be robustly obtained with additional Kinect depth sensors. In this paper, we propose an extended F-formation system for robust detection of interactions and interactants. The extended F-formation system employs a heat-map based feature representation for each individual, namely Interaction Space (IS), to model their location, orientation, and temporal information. Using the temporally encoded IS for each detected interactant, we propose a best-view camera selection framework to detect the corresponding best view camera for each detected social interaction. The extended F-formation system is evaluated with synthetic data on multiple scenarios. To demonstrate the effectiveness of the proposed system, we conducted a user study to compare our best view camera ranking with human’s ranking using real-world data.



Poster Presenter: Ma Keng Teck (

Title: VIP: A unifying framework for computational eye-gaze research

Eye-gaze is an emerging modality in many research areas and applications. The reference model for saliency research is bi-directional: top-down and bottom-up. In biometric research, the identity of a person can be inferred from eye-gaze. In human-computer interface, eye-gaze is a response of the interactions between the tasks and the visual stimulus. These models are incomplete and we will propose the VIP framework. This unifying framework captures the dependence of eye-gaze on Visual stimuli, Intent, and Person, making it more complete and subsuming all existing models. We conducted extensive user experiments to collect the VIP dataset. It is the first eye-gaze dataset to contain all 3 factors. The utility of our framework is illustrated with the application of inferring the viewer's personal traits. The accuracy of 0.92 is achieved for classification of Introvert/Extrovert. We further demonstrate the superiority and completeness of our VIP framework by showing that incorporating personal traits into saliency model will produce more accurate fixation predictions.



Poster Presenter: Tang Rui Ming (

Title: The Price Is Righth

We study the relationship between quality and price of data. We proposed a theoretical and practical pricing framework for a data market in which data consumers can trade data quality for discounted prices. In most data markets, prices are prescribed and accuracy is determined by the data. Instead, we consider a model in which accuracy can be traded for discounted prices: ``what you pay for is what you get". The data market model consists of data consumers, data providers and data market owners. The data market owners are brokers between the data providers and data consumers. A data consumer proposes a price for the data that she requests. If the price is less than the price set by the data provider, then she gets an lower quality value. Based on this principle, we first consider pricing relational data based on the quality dimension of accuracy, then we consider pricing XML data based on the quality dimension of completeness. We also study the price of queries for cases that data consumers request for data in forms of queries. We propose a generic data pricing model that is based on minimal provenance, i.e. minimal sets of tuples contributing to the result of a query. We show that the proposed model fulfils desirable properties such as contribution monotonicity, bounded-price and contribution arbitrage-freedom. We present a baseline algorithm to compute the exact price of a query based on our pricing model. We show that the problem is NP-hard. We therefore devise, present and compare several heuristics. We conduct a comprehensive experimental study to show their effectiveness and efficiency.



Poster Presenter: Minhui Zhu (

Title: An Image-based Representation for Peer-Assisted Rendering in Networked Virtual Environments

Networked virtual environments (NVEs) are becoming an important class of distributed multimedia applications. However, due to the limitations in rendering capability and network connection of mobile devices, it is still difficult to guarantee the mobile users the interactive navigation experience as good as PC users in NVEs. The state of the art approach to ease mobile access to NVEs involves a powerful server and image-based rendering techniques. It does reduce the rendering workload on mobile devices significantly with the assisting server, but at the same time, the server can also be easily overwhelmed as the client number increases. Therefore, we introduce a new and more scalable remote rendering technique, called peer-assisted rendering. A resource-constrained client will request part of the rendered scene from other powerful peers with similar viewpoints within the same virtual environment, warps the rendered parts, and merges into its own view. Furthermore, both the peer-assisted and server-assisted approaches adopt image-based representations and the image warping technique to reduce the interaction delay. While having multiple reference images reduces warping artifacts, managing these reference images can be expensive. We then propose a new image-based representation, called the sprite tree, to preserve and organize samples of static content from reference images. A sprite tree is storage-efficient, allowing the reuse of a large number of previously-rendered samples. It could be used as both a server/peer-side cache and a client-side cache to accelerate the rendering.



Poster Presenter: Shanghong ZHAO (

Title: 3D Mesh Preview Streaming

Online galleries of 3D models typically provide two ways to preview amodel before the model is downloaded and viewed by the user: (i) byshowing a set of thumbnail images of the 3D model taken fromrepresentative views (or keyviews); (ii) by showing a video of the 3Dmodel as viewed from a moving virtual camera along a path determined bythe content provider. We propose a third approach called preview streamingfor mesh-based 3D objects: by streaming and showing parts of the meshsurfaces visible along the virtual camera path. The research focuses onthe preview streaming architecture and framework and presents ourinvestigation into how such a system would best handle network congestioneffectively. We present three basic methods: (a) STOP-AND-WAIT, where thecamera pauses until sufficient data is buffered; (b) REDUCE-SPEED, wherethe camera slows down in accordance to reduce network bandwidth; and (c)REDUCE-QUALITY, where the camera continues to move at the same speed butfewer vertices are sent and displayed, leading to lower mesh quality. Wefurther propose two advanced methods: (d) KEYVIEW-AWARE, which trades offmesh quality and camera speed appropriately depending on how close thecurrent view is to the keyviews, and (e) ADAPTIVE-ZOOM, which improvesvisual quality by moving the virtual camera away from the original path. Auser study reveals that our KEYVIEW-AWARE method is preferred over thebasic methods. Moreover, the ADAPTIVE-ZOOM scheme compares favorably tothe KEYVIEW-AWARE method, showing that path adaptation is a viableapproach to handling bandwidth variation.


Network & Systems



Poster Presenter: Guo Xiangfa (

Title: Opportunities and challenges of opportunistic mobile networks

The research on opportunistic mobile networks studies how to utilize the communication chances between mobile nodes. As the short-range-communication chipsets, e.g., WiFi, Bluetooth, NFC, are widely integrated into smart phones and vehicles, more applications rise for mobile opportunistic networks. Example applications include smart phone sensor data communication, mobile social networks, and intelligent transportation systems. All these applications require some common fundamental functions. First, in order to transmit messages using short range ommunication interfaces, two devices need to know the presence of each other at the very first step, which we call neighbor discovery for mobile nodes. Second, when messages need to be transmitted from a source and a destination who cannot directly “see” each other, some nodes in the middle are needed to carry and forward the messages. In this case, an efficient routing scheme for selecting message carriers is essential. Thirdly, in the distributed system formed by mobile nodes, how to efficiently collect “up-to-date” data is key to many applications. The three fundamental functions are crucial to the feasibility and performance of opportunistic mobile networks. In the near future, once opportunistic mobile networks are commercialized, system performance immediately requires theory and system support from the current research results. Our research work covers the three fundamental topics. Our research focuses on the resource efficiency and system performance. From system and theory view, we extensively investigate key factors in system performance for the three topics. Our research results can significantly save resources compared with state of the art. These are either evaluated by systems implementations, or simulations, or both.



Poster Presenter: Wang Hui (

Title: Adaptive Video Delivery with Wireless Multicast

As the growth of HD video traffic outpaces that of wireless network, dynamic adaptation of video bit-rates is proposed to match network conditions. While recent studies demonstrate that the existing adaptive streaming mechanisms are still ineffective when multiple flows are competing for restricted bandwidth. To address this ineffectiveness, we exploit wireless multicast to efficiently utilize the network resource. In particular, our approach has the following three differentiating features: (i) the bandwidth consumption could be reduced with wireless multicast; (ii) instead of using the basic link rate, adaptive multicast link rate is employed; and (iii) intelligent algorithms are suggested to optimally allocate the resource for each user.



Poster Presenter: Padmanabha Venkatagiri Seshadri (

Title: Resource aware selection of user generated content in constrained mobile networks

Proliferation of mobile devices and increasing sophistication of the on-board sensors such as cameras, speaker/microphone etc available in mobile devices has resulted in the availability of enormous quantities of user generated content (UGC) which is supported by descriptive meta-information. This UGC is complementing and even supplanting the traditional content provider services available on the internet. In addition, UGC is also gaining significance as in-situ content. I.e UGC generated at a certain event is disseminated in the vicinity of the event even as it is happening. Such ""proximity social media sharing"" enhances the user experience. On the other hand, existing networking technologies such as 3G/HSDPA, WiFi are finding it hard to scale to support this heavy loads of traffic due to the UGC. Furthermore, mobile devices are also constrained by their battery capacities. Existing research in connection with UGC is mainly focused on processing the content and enriching its description with the sensor data, while, the effort required to retrieve the content is not considered as part of the problem. Techniques applied to reduce the load on 3G/HSDPA include traffic offloading or mobile-to-mobile communication have agnostic to the characteristics of the UGC which is multifaceted. This means the content aspect of the problem and the resources available for its transfer are not connected effectively. We argue that by considering the content characteristics of UGC which could be subjective (such as user feedback) or objective (sensory data, spaciotemporal conditions) and the resource availability of the network, we could ""select"" the UGC that satisfies the needs of the consumers of the content. The contributions of our work includes the design and development of techniques to effectively bridge the gap between the UGC and the resource and select appropriate data. These techniques will be part of a complete operational system which facilitates consumers to access the data and producers of content to contribute their content in an resource aware manner.



Poster Presenter: Lavanya Ramapantulu (

Title: Mix and Match: Analyzing the Energy Efficiency of Heterogeneous Clusters

Traditional datacenter systems advocate the use of high-performance hardware, resulting in increased power consumption and cooling costs. With increasing availability of systems having diverse performance-to-power ratios, we analyze the energy efficiency of mixing high-performance and low-power nodes in a cluster. Using a model-driven analysis, we predict the heterogeneous mix of nodes that is the most energy-efficient while maintaining a given deadline. Considering service demands of the workloads on cores, memory and I/O devices, we derive Pareto-optimal configurations by matching the execution rate of different nodes. Our mix and match approach determines heterogeneous configurations that exhibit a "sweet region", where energy usage reduces linearly as the deadline is relaxed. Our analysis shows that mixing high-performance and low-power nodes is almost always more energy-efficient than homogeneous datacenter clusters.



Poster Presenter: Wang Wei (

Title: Practical WiFi Mesh Networks: from Ground to Air

WiFi mesh networks are designed to provide network connectivity in operating environments where wired infrastructure is infeasible or impractical. They have the advantages of low cost, fast and flexible deployment, distributed management, and fault tolerance. Therefore, WiFi mesh networks are ideal for the applications like disaster response and recovery missions and Internet access for less privileged areas. However, in spite of these advantages, WiFi mesh networks have not seen widespread deployment and many startups of mesh networks have ceased to operate, or moved on to other technologies. To investigate the performance of large-scale WiFi mesh networks, we deployed a 20-node WiFi mesh testbed on campus. Through extensive measurement, we found that one problem that hinders the deployment of WiFi mesh networks is MAC unfairness. For example, a node far away from gateway may incur throughput starvation, which seriously affects user experience.
To this end, we designed and implemented a scheduling protocol that is able to achieve much better fairness performance, without affecting the overall throughput. In addition, we also developed a power control protocol to mitigate the interference among competing mesh nodes, thereby further improving the overall throughput. Since our WiFi mesh testbed is built from off-the-shelf hardware, there is little assumption made in our work, and our solutions can be deployed in a practical system.
Furthermore, we are pushing the envelope of WiFi mesh deployment by incorporating aerial nodes, which are built using the latest quadcopter technology. Our motivation is to allow a mesh network to dynamically expand its coverage from a fixed 2D region to an arbitrary 3D space. This will extend the use of WiFi mesh networks into new operating scenarios where it will simply be impractical to rely on WLANs. To demonstrate the usefulness of our aerial mesh network, we have designed and implemented an aerial node for search and rescue mission. Specifically, our aerial mesh node is able to quickly and reliably detect the background WiFi signal from the smartphone of a lost hiker, thereby expediting the search and rescue operation. Some of our ongoing work on aerial mesh networks includes automatic antenna adjustment for aerial node tracking, multi-hop video streaming, and collaborative flight control using on-board cameras.


Software Engineering & Security



Poster Presenter: Asankhaya Sharma

Title: Exploiting Undefined Behaviors for Efficient Symbolic Execution

Symbolic execution is an important and popular technique used in several software engineering tools for test case generation, debugging and program analysis. As such improving the performance of symbolic execution can have a huge impact on the effectiveness of such tools. In addition, optimizations based on undefined behaviors are an essential part of current C and C++ compilers (like GCC and LLVM). In this paper, we present a technique to systematically introduce undefined behaviors during compilation to speed up the subsequent symbolic execution of the program. We have implemented our technique inside LLVM and tested with an existing symbolic execution engine (Pathgrind). Preliminary results on the SIR repository benchmark are encouraging and show 48% speed up in time and 30% reduction in the number of constraints.



Poster Presenter: Kuldeep Kumar (

Title: Detecting Design Similarity Patterns in Programs

Code clones are repeated program structures of considerable size and significant similarities occurring in various forms within software systems. They play a major role in software maintenance and reuse. Several techniques have been proposed in the literature for detecting code clones. Existing clone detection techniques mainly focus on detecting similar code fragments, files or directories. But many design and analysis patterns show as recurring configurations of collaborating components such as classes. Such collaborative patterns often represent program structures that exhibit specific behavior that is meaningful for developers who need understand programs, reengineer legacy code for reuse, refactor or simply maintain programs. Unfortunately, unless manually documented, collaborative patterns remain implicit in code. Our aim is at time-efficient and scalable methods to detect collaborative patterns.



Poster Presenter: Zhang Chunwang (

Title: Tagged-MapReduce: A General Framework for Secure Computing on Hybrid Clouds

We present tagged-MapReduce, a general extension to MapReduce that supports secure computing with mixed-sensitivity data on hybrid clouds. Tagged-MapReduce augments each key-value pair in MapReduce with a sensitivity tag. This enables fine-grained dataflow control during execution to prevent data leakage as well as supporting expressive security policies and complex MapReduce computations. Security constraints for preventing data leakage impose restrictions on computation and data storage/transfer, hence, we present scheduling strategies that can exploit properties of the map and reduce functions to rearrange the computation for greater efficiency under these constraints while maintaining MapReduce correctness. We present a general security framework for analyzing MapReduce computations in the hybrid cloud which captures how dataflow can leak information through execution. Experiments on Amazon EC2 with our prototype in Hadoop show that we are able to obtain security while effectively outsourcing computation to the public cloud and reducing inter-cloud communication.



Poster Presenter: Zheng Chaodong (

Title: Thwarting Sybil Attacks in Multi-channel Wireless Networks

Sybil attacks refer to the behavior where malicious users dishonestly generate large number of fake identities (also called sybil identities), and use these identities to gain extra benefit or conduct other hostile activity. Researchers have confirmed it can do severe damage to many real-life systems. For example, in a WiFi network where the base station allocates equal bandwidth to each client, an adversary can gain unfair advantage by simulating multiple (fake) clients. Similarly, for voting systems, an adversary can also use sybil attacks to manipulate the final result. Over the past few years, we study how to leverage the characteristics of multi-channel wireless networks to counter sybil attacks. In particular, for both centralized and ad-hoc wireless networks, we have developed protocols that can efficiently thwart sybil attacks and other malicious behavior. Many of our algorithms have relatively low overhead, and can be adopted, or combined with other protocols, to solve other critical problems.



Poster Presenter: Abhijeet Banerjee (

Title: An automated technique to expose energy bugs in Android applications

Smartphones have brought upon radical changes in the mobile industry. However, due to the increasing complexity of smartphone applications, they might consume substantial battery power. This poster presents a test generation framework that exposes anomalous energy behaviours in  Android applications. We call such anomalous energy behaviours as energy hotspots. Energy hotspots capture specific locations in the application that exhibit potential wastage of battery power. For each detected energy hotspot, we also report a sequence of user inputs (e.g. taps or touches  on the smartphone screen) that lead to the same. Our framework systematically combines graph search and learning heuristics to redirect the test generation towards likely energy hotspots. Specifically, learning heuristics are used to compute a subgraph that is more likely to exhibit energy hotspots. Our framework uses real power measurements and executes an application on an Android device without any instrumented code. Our experiments with several applications from Google Play store/F-Droid show the efficacy of our test generation framework. In particular, we have uncovered a number of energy hotspots, some of which cause a substantial wastage of battery power.


Information System



Poster Presenter: Li Mei (

Title: Mobile App Portfolio Management and Developer's Performance: An Empirical Study of the Apple iOS Platform

A critical challenge faced by mobile app developers today is to effectively manage their app portfolio, but this issue has rarely been addressed in the IS academic literature. To address this gap, we empirically examine the relationship between the size and diversity of mobile developers' app portfolio and their performance in a mobile platform. Using a data set from the Apple App Store, and based on panel-level linear model estimation, we find a negative impact of app portfolio diversity on developers' performance. However, this impact decreases with portfolio size. This result implies that a diversified app portfolio hinders developers from fostering core competency, whereas their increasing development experience could mitigate this negative effect. Our results provide mobile app developers with a strategic growth trajectory recommendation, i.e., focus on product specialization during the early growth phases but adopt a diversification strategy after the accumulation of substantial development experience.



Poster Presenter: Sangaralingam Kajanan (

Title: Twitter post filter for Mobile Apps

The Twitter platform has emerged as a leading medium of conducting social commentary, where users remark upon all kinds of entities, events and occurrences. As a result, organizations are starting to mine twitter posts to unearth the knowledge encoded in such commentary. Mobile applications, commonly known as mobile apps, are the fastest growing consumer product segment in the history of human merchandizing, with over 800,000 apps on the Apple platform and over 600,000 on Android. A particularly interesting issue is to evaluate the popularity of specific mobile apps by analyzing the social conversation on them. Clearly, twitter posts related to apps are an important segment of this conversation and have been a main area of research for us. In this respect, one particularly important problem arises due to a name conflict of mobile app names and the names that are used to refer the mobile apps in twitter posts. In this paper, we present a strategy to reliably extract twitter posts that are related to specific apps, but discovering the contextual clues that enable effective filtering of irrelevant twitter posts is our concern. While our application is in the important space of mobile apps, our techniques are completely general and may be applied to any entity class. We have evaluated our approach against a popular Bayesian classifier and a commercial solution. We have demonstrated that our approach is significantly more accurate than both of these. These results as well as other theoretical and practical implications are discussed.



Poster Presenter: Prasanta Bhattacharya (

Title: Beyond UGC: Using Behavioral Analytics to Target Profitable Customers on Social Media

With advances in information and communication technologies (ICT), companies and platforms look to use the increasing volume and diversity of user-generated content (UGC) to predict consumer behavior, but with mixed results. As such, we ask what is the relationship between a consumer’s inherent motivation to participate in public UGC activities and his purchase behavior? To address this question, we use unique datasets from a social network site (SNS) and an offline fashion retailer to investigate the relationship between a user’s online image-seeking behavior (measured by a public-private sentiment divergence metric) and his purchase expenditure. We use both text mining and econometric methods on over 240 million observations to find that: 1) while the exclusive use of public and private content leads to non-significant predictions, divergence of sentiment is a significant predictor of offline purchases, 2) users who engage in SNS due to image-seeking motivations spend less and are more price sensitive, and 3) however, they spend more when exposed to certain site features like brand-pages. Finally, we extend text-mining techniques to extract specific behavioral cues from user content to show that sentiment divergence also correlates to a divergence in actual words and topics discussed. Marketers and platform owners can benefit from our results by designing appropriate features to target such attractive users.



Poster Presenter: Yue Yanzhen (

Title: Co-Navigability, Tracking Fulfillment and Autonomy in Collaborative Online Shopping

This paper explores three functional mechanisms that may help to improve customers' collaborative online shopping experience, i.e. co-navigability, tracking fulfillment, and autonomy. The effects of the three functional mechanisms on collaborative online customers shopping process satisfaction and outcome satisfaction are empirically investigated in a lab survey. The findings indicate that perceived co-navigability has positive effect on both process satisfaction and outcome satisfaction of customers. Meanwhile, the results show that perceived tracking fulfillment and perceived autonomy have positive effect on customers' process satisfaction. Theoretical contributions and practical implications are discussed.