Search

Ride-on-demand (RoD) services use dynamic prices to balance the supply and demand to benefit both drivers and passengers, as an effort to improve service efficiency. However, dynamic prices also create concerns for passengers: the “unpredictable” prices sometimes prevent them from making quick decisions at ease. It is thus necessary to give passengers more information to tackle this concern, and predicting dynamic prices is a possible solution. We focus on fine-grained dynamic price prediction – predicting the price for every single passenger request. Price prediction helps passengers understand whether they could get a lower price in neighboring locations or within a short time, thus alleviating their concerns. The prediction is performed by learning the relationship between dynamic prices and features extracted from multi-source urban data. There are linear or non-linear models as candidates for learning, and using different models leads to varying implications on accuracy, interpretability, model training procedures, etc. We train one linear and one non-linear model as representatives, and evaluate their performance from different perspectives based on real service data. In addition, we interpret feature contribution, at different levels, based on both models and figure out what features or datasets contribute the most to dynamic prices. Finally, based on evaluation results, we provide discussions on model selection under different circumstances, and propose a way to combine the two models. Our hope is that the study not only serves as an accurate prediction for passengers, but also provides concrete guidance on how to choose between models to improve the prediction.

International audience; Recent years witness the rapidly-growing business of ride-on-demand (RoD) services such as Uber, Lyft and Didi. Unlike taxi services, these emerging transportation services use dynamic pricing to manipulate the supply and demand, and to improve service responsiveness and quality. Despite this, on the drivers' side, dynamic pricing creates a new problem: how to seek for passengers in order to earn more under the new pricing scheme. Seeking strategies have been studied extensively in traditional taxi service, but in RoD service such studies are still rare and require the consideration of more factors such as dynamic prices, the status of other transportation services, etc. In this paper, we develop ROD-Revenue, aiming to mine the relationship between driver revenue and factors relevant to seeking strategies, and to predict driver revenue given features extracted from multi-source urban data. We extract basic features from multiple datasets, including RoD service, taxi service, POI information, and the availability of public transportation services, and then construct composite features from basic features in a product-form. The desired relationship is learned from a linear regression model with basic features and high-dimensional composite features. The linear model is chosen for its interpretability-to quantitatively explain the desired relationship. Finally we evaluate our model by predicting drivers' revenue. We hope that ROD-Revenue not only serves as an initial analysis of seeking strategies in RoD service, but also helps increasing drivers' revenue by offering useful guidance.

As part of the commemoration of the 50th anniversary of SIGCOMM, a panel discussion on "Past and Future Directions in Networking Research" was held during the SIGCOMM conference on August 21, 2019. The panelists were Dah Ming Chiu (Chinese University of Hong Kong), Manya Ghobadi (Massachusetts Institute of Technology), Mark Handley (University College London), Jennifer Rexford (Princeton University), and Lixia Zhang (University of California, Los Angeles). The discussion was moderated by Bruce Maggs (Duke University and Akamai Technologies).

In emerging ride-on-demand (RoD) services, dynamic pricing plays an important role in regulating supply and demand and improving service efficiency. Despite this, it also makes passenger anxious: whether the current price is low enough, or otherwise, how to get a lower price. It is thus necessary to provide more information to ease the anxiety, and predicting the prices is one possible solution. In this study, the authors predict the dynamic prices to help passengers learn if there is a lower price around. They first use entropy of historical prices to characterize the predictability of prices in different locations and claim that different prediction algorithms should be used to balance between efficiency and accuracy. They present an ensemble learning approach to price prediction and compare it with two baseline predictors, namely a Markov and a neural network predictor. The performance evaluation is based on the real data from a major RoD service provider. Results verify that the two baseline predictors work well in locations with different levels of predictabilities, and that ensemble learning significantly increases the prediction accuracy. Finally, they also evaluate the effects of prediction, i.e., the probability that passengers could benefit from the prediction and get a lower price.

In emerging ride-on-demand (RoD) services such as Uber and Didi (in China), dynamic prices play an important role in regulating supply and demand, trying to improve the service quality for both drivers and passengers. In this paper, we take a new perspective to study RoD services besides the supply or demand, and focus on passengers' reaction to dynamic prices. Passengers' reaction can be regarded as a process of searching for the best price before getting on a car, and the searching process reflects passengers' demand elasticity -- “how eager they are requesting a ride”. We collect data of passengers' reaction from a real RoD service provider in China, and analyze the patterns of passengers' reaction. The analysis results show that both the dynamic prices and passengers' demand elasticity influence their reaction. We then adopt and extend a previous model for sequential search from a known distribution to understand passengers' reaction, and use our data to obtain the search costs under various circumstances, which could be interpreted as passengers' demand elasticity. Insights on the search cost and other relevant quantities are discussed. Our expectation is that the result of the study should be helpful not only for service providers in designing dynamic pricing algorithms, but also for passengers and policy makers in understanding the effects and implications of dynamic pricing.

The net neutrality debate has been centered on the question: should Internet service providers (ISPs) be allowed to differentiate services for Internet content traffic? The concern is that the differentiation imposed by selfish ISPs might discriminate content providers (CPs) and harm social welfare. Although market competition among ISPs would alleviate the problem and moderate the necessity for net neutrality regulations, the problem remains in monopolistic access markets. We focus on such a market and study paid prioritization where CPs voluntarily pay for prioritizing their traffic under shared capacity. We study an ISP’s pricing strategy, CPs’ choices of priority, and the resulting system equilibrium, based on which we derive the utility of the ISP and CPs as well as social welfare. This paper shows that: 1) an ISP’s optimal pricing leads to an efficient differentiation among CPs, such that social welfare is close to its maximum; 2) although ISPs might inhibit capacity deployment in the short run, price regulation could solve this issue; and 3) under medium system scale and capacity cost, ISPs would have strong incentives to expand capacity under paid prioritization. From a welfare perspective, our results suggest that paid prioritization could be superior to the imposition of net neutrality regulations.

Video recommendation has become an essential part of online video services. Cold start, a problem relatively common in the practical online video recommendation service, occurs when the user who needs video recommendation has no viewing history. (Cold start consists of the new-user problem and the new-item problem. In this paper, we discuss the new-user one.) A promising approach to resolve this problem is to capitalize on information in online social networks (OSNs): Videos viewed by a user’s friend may be good candidates for recommendation. However, in practice, this information is also quite limited, either because of insufficient friends or lack of abundant viewing history of friends. In this work, we utilize social groups with richer information to recommend videos. It is common that users may be affiliated with multiple groups in OSNs. Through members within the same group, we can reach a considerably larger set of users and hence more candidate videos for recommendation. In this paper, by collaborating with Tencent Video, we propose a social-group-based algorithm to produce personalized video recommendations by ranking candidate videos from the groups a user is affiliated with. This algorithm was implemented and tested in the Tencent Video service system. Compared with two state-of-the-art methods, the proposed algorithm not only improves the click-through rate, but also recommends more diverse videos.

The streaming of panoramic videos comes with a new challenge: adapting the streamed content to user attention. In this paper, by making some abstraction we study the key issues of the problem: (1) online content switching strategies, and (2) offline segmentation in preparation for adaptive streaming. Our work demonstrates how the problem depends fundamentally on user behavior, represented by user attention trajectories. Although we only relied on a typical real-world measured trajectory and two artificially generated benchmark trajectories, we point out how more knowledge of user attention can help optimize both the online and offline strategies.

In this paper, we study the posting behavior of OSN users, in particular the posting frequency and temporal patterns, and consider possible interpretations of how users use the platform. At the aggregate (macro) level, we find two distinct peaks, one during morning working hours, and one in the evening. The morning peak is more pronounced for frequent posters, while the evening peak is pronounced in the remaining users. We postulate that this difference results from qualitatively different usage of the OSN platform (e.g. for work, with customers, etc.) than purely social interactions (e.g., friends, family, etc.). We also study user posting behavior at an individual (micro) level and apply LDA to cluster user temporal patterns, interpret our results. Our study provides possibly new insights into user activity in today's OSNs, and suggests a framework for profiling users based on their posting activities. In the process, we provide a novel application of LDA, to temporal user posting behavior by equating the time epochs of posts to words in documents. We believe our approach will complement other methods of user profiling based on static demographic information and friendship network information.

In past studies, Online Social Networks (OSNs) is commonly assumed to build on triadic closure, implying that alters in a user’s ego network form either a single connected component, or a small number of connected components. In real-world OSNs, we find a significant number of users with a different ego network pattern consisting of a more diverse social circle with many friends not connected to each other. We conjecture this is caused by the increasing use of OSNs for functional (e.g. business or marketing) rather than traditional socializing activities. We refer to the resulting prototypical users as functional and social users respectively. In this paper, we use a manually tagged dataset (from Tencent social platform) to identify these two type of users and demonstrate their different friend circle patterns using examples. To help sort out functional users from social users, we develop metrics to measure diversity of a user’s friend circle, borrowing concepts from classic works on structural holes and community detection. We show how the different measures of diversity perform in classifying the two types of users. Then we combine the structural diversity measures and behavioral measures to train machine leaning models. We further study ego network diversity in groups of users with different demographics (profession, gender and age). Our results bring new insights to the heterogeneous nature of today’s OSNs and help better profile users. Our study also shed new light on structural hole theory and Dunbar’s number in the OSN context.

All large scale online video systems, e.g. Netflix and Youku, make a significant investment on video recommendations that can dramatically affect video information diffusion processes among users. However, there is a lack of efficient methodology to interpret how various recommendation mechanisms affect information diffusion processes resulting in the difficulty to evaluate video recommendation efficiency. In this work, we propose to quantify and explain video recommendation mechanisms by using epidemic models to mine video view count traces. It is well known that an epidemic model is an efficient approach to model information diffusion processes; while view count traces can be viewed as the results of video information diffusion driven by video recommendations. Thus, we propose a framework based on extended epidemic models to quantify and interpret two recommendation mechanisms, i.e., direct and word-of-mouth (WOM) recommendations, by fitting video view count traces collected from Tencent Video, a large scale online video system in China. Our approach is a novel methodology to evaluate video recommendation mechanisms, and a new perspective to interpret how recommendation mechanisms drive view count evolution. Refereed/Peer-reviewed

Online video-on-demand(VoD) services invariably maintain a view count for each video they serve, and it has become an important currency for various stakeholders, from viewers, to content owners, advertizers, and the online service providers themselves. There is often significant financial incentive to use a robot (or a botnet) to artificially create fake views. How can we detect fake views? Can we detect them (and stop them) efficiently? What is the extent of fake views with current VoD service providers? These are the questions we study in this article. We develop some algorithms and show that they are quite effective for this problem. Refereed/Peer-reviewed

There are two types of P2P systems satisfying two different user demands: 1) file downloading and 2) video-on-demand (VoD) streaming. An example of file downloading is the original BitTorrent, and examples for VoD streaming include various commercial P2P-based VoD streaming systems such as that offered by PPLive. We have a hypothesis-by combining a type: 1) system and 2) system as a single P2P system, both the file downloading users and the streaming users of the same video will benefit in performance. The reasoning is that at any moment, only a subset of the file downloading peers can provide good service to VoD streaming peers and the VoD streaming peers are only good at providing service to a different subset of the file downloading peers. The former subset is the set of peers close to completing the downloading of the video file; whereas the latter subset is the set of peers starting to download a video. In this paper, we propose a novel design for a mesh-based video distribution system without depending on video replication on streaming peers. We produce simple back-of-the-envelop analysis to show its effectiveness. Then, we further validate our design and compare it with other designs through simulation and experiments in practical networking environment by implementing a prototype. Refereed/Peer-reviewed

In emerging ride-on-demand (RoD) services such as Uber or Didi (in China), dynamic pricing plays an important role in regulating supply and demand, trying to make such service, to some extent, more convenient for passengers. Despite the convenience, dynamic pricing also exerts mental burden on passengers: they wonder whether the current price is low enough to accept, or if it is not, what they could do to get a lower price. Without extra information, passengers sometimes feel anxious and lose satisfaction. It is thus necessary to provide more information to relieve the anxiety, and price prediction is one of the solutions. In this paper we predict the dynamic prices to help passengers understand whether they could get a lower price in neighboring locations or within a short time. We first divide a city into rectangular cells, and use entropy and the temporal correlation of prices to characterize the predictability of prices of each cell. Based on the predictability of prices, we claim that different prediction algorithms should be used in different city areas, to balance between efficiency and accuracy. We design and implement two predictors, namely a Markov predictor and a neural network predictor, and evaluate their performance based on the real data we collected from a major RoD service provider in China. The results verify that the Markov predictor works well enough in highly-predictable areas, and the neural network predictor, while requiring more computation time, works better in areas with lower predictabilities. Finally, we also evaluate the effects of our prediction, i.e., the probability that passengers (in different city areas) could benefit from the prediction and get a lower price.

Smart-home is becoming increasingly popular in recent years, and it introduces a new content retrieval paradigm - delay-insensitive downloading. In this new paradigm, users do not require the content retrieval task to finish as soon as possible, but only set a deadline for it. We study the role of this paradigm in the traffic engineering of a chunk-based cloud storage service. We propose that it could help to reduce the high intra-datacenter traffic at peak resulting from the chunk-based architecture, by delaying users' content requests when necessary. Because of the introduction of the new paradigm, we consider the co- existence of three applications (downloading, streaming, delay-insensitive downloading) in the service, and try to understand the best way to delay users' requests. We conduct an incentive-based study for the evaluation of schemes of delaying users' requests: the service provider pays incentives to users to promote this new paradigm. The incentive, as well as users' application-specific QoE on the service, is modelled, and a framework for the evaluation is presented. We then apply our framework to study several proposed delay schemes and present both quantitative and qualitative results.

Nowadays digital services, such as cloud computing and network access services, allow dynamic resource allocation and virtual resource isolation. This trend can create a new paradigm of flexible pricing schemes. A simple pricing scheme is to allocate multiple isolated service classes with differentiated prices, namely Paris Metro Pricing (PMP). The benefits of PMP are its simplicity and applicability to a wide variety of general digital services, without considering specific performance guarantees for different service classes. The central issue of our study is whether PMP is economically viable, namely whether it will produce more profit for the service provider and whether it will achieve more social welfare. Prior studies had only considered specific models and arrived at conflicting conclusions. In this article, we identify unifying principles in a general setting and derive general sufficient conditions that can guarantee the viability of PMP. We further apply the results to analyze various examples of digital services., Comment: This paper appears in ACM Transactions on Internet Technology (ToIT), Special Issue on Pricing and Incentives in Networks and Systems, Vol. 14, No. 12, Issue 2-3, pp12:1-12:21, Oct 2014. A preliminary version has been presented at IEEE INFOCOM 2010. in C-K Chau (2014)