Traditional recommendation systems (RecSys) suffer from two problems: the exploitation-exploration dilemma and the cold-start problem. One solution to solving the exploitation-exploration dilemma is the contextual bandit policy, which adaptively exploits and explores user interests. As a result, the contextual bandit policy achieves increased rewards in the long run. The contextual bandit policy, however, may cause the system to explore more than needed in the cold-start situations, which can lead to worse short-term rewards. Cross-domain RecSys methods adopt transfer learning to leverage prior knowledge in a source RecSys domain to jump start the cold-start target RecSys. To solve the two problems together, in this paper, we propose the first applicable transferable contextual bandit (TCB) policy for the cross-domain recommendation. TCB not only benefits the exploitation but also accelerates the exploration in the target RecSys. TCB's exploration, in turn, helps to learn how to transfer between different domains. TCB is a general algorithm for both homogeneous and heterogeneous domains. We perform both theoretical regret analysis and empirical experiments. The empirical results show that TCB outperforms the state-of-the-art algorithms over time.

Cold start is always a key challenge for building real-life recommendation systems. Thanks to the ever-growing multi-modal data in the mobile Internet age and the latest deep learning techniques, transfer-learning based cross-domain recommendation starts to play a crucial role in tackling the cold start problem and to provide "warm-start" recommendation for new users. At Cheetah Mobile, we apply transfer learning to build personalized recommendation systems for both advertisement and content scenarios, serving 600+ millions monthly active mobile users. In particular, we leveraged the app install & usage and many other mobile data, built a Unified User Profile (UUP) by using transfer learning and deep learning, and developed cross-domain personalized Ad and news recommendation. Our approaches enable us to solve the cold start problem with close to full coverage of our user base while yielding significant CTR increase and better user experience.

People around the world are more connected today than ever before. By making phone calls, sending text messages and participating in online chats, mobile users are frequently interacting with their social connections through multiple communication channels. This trend is expected to continue with the emergence of immensely popular communication apps on mobile devices. Intuitively, these interactions on users' mobile phones can reveal valuable information regarding their social relationship with their phone contacts. Understanding such relationship can help provide new services and improve users' mobile experience. In this paper, we explore the opportunity to deeply understand these social relationship through mining mobile communication data. By building an on-device mining framework called Commsense, we show that automatically learning and understanding such relationship can efficiently support useful applications such as categorizing mobile contacts, identifying their relative importance, and automatically managing mobile contacts with very little human interference.

Energy consumption and mobile data cost are two key factors affecting users’ willingness to participate in mobile crowd-sensing tasks. While data-plan (DP) users are mostly concerned with energy consumption, non-data-plan (NDP) users are more sensitive to data cost. Traditional ways of data uploading in mobile crowdsensing tasks often go to two extremes: either in real time or completely offline after the whole task is over. In this paper, we propose effSense—an energy-efficient and cost-effective data uploading framework, which utilizes adaptive uploading schemes within fixed data uploading cycles. In each cycle, effSense empowers the participants with a distributed decision making scheme to choose the appropriate timing and network to upload data. effSense reduces data cost for NDP users by maximally offloading data to Bluetooth/WiFi gateways or DP users encountered; it reduces energy consumption for DP users by piggybacking data on a call or using more energy-efficient networks rather than initiating new 3G connections. By leveraging the predictability of users’ calls and mobility, effSense selects proper uploading strategies for both user types. Our evaluation with the MIT reality mining and Nodobo datasets shows that effSense can reduce 55%–65% energy consumption for DP users, and 48%–52% data cost for NDP users, respectively, compared to traditional uploading schemes.

This paper studies mobile sensing in a complete distributed and opportunistic scheme. We present a novel sensing strategy for sensing nodes without movement constraints. This strategy offers information sharing and sensor scheduling that maximizes the benefits from collaboration between the nodes. We evaluate the strategy with two real world datasets, where mobile devices sense the surroundings and share the information with nearby devices via Bluetooth. We show that data sharing of energy-hungry sensing (e.g., GPS) via nearby devices can save a substantial amount of energy. Moreover, we formulate the offline measurement scheduling as a non-linear optimization problem and compare the offline optimal results with our online scheduling performance.

With the growth of web documents, the ambiguity of personal name becomes more common and brings poor performance of web search. Identifying a correct personal entity from the a piece of or the whole document is still a very challenging problem, especially for Chinese websites. In this paper, we propose a novel Ontology-based approach for Personal Name Disambiguation (named "OnPerDis"). This approach has two main steps: first, we construct person ontology (PO) with rich conceptual modeling as well as a large set of supporting instances, second, for a given personal name on the web, we create a temporary instance and extract features from the web documents, calculate the similarity between this temporary instance and the instances in the PO. The one with the highest similarity score is chosen as the appropriate personal name. Our extensive evaluations with two rich real-life datasets (CIPS-SIGHAN 2012 NERD and Chinese web documents) shows OnPerDis' efficacy on personal name disambiguation on the Web.

With the large-scale adoption of GPS equipped mobile sensing devices, positional data generated by moving objects (e.g., vehicles, people, animals) are being easily collected. Such data are typically modeled as streams of spatio-temporal (x,y,t) points, called trajectories . In recent years trajectory management research has progressed significantly towards efficient storage and indexing techniques, as well as suitable knowledge discovery. These works focused on the geometric aspect of the raw mobility data. We are now witnessing a growing demand in several application sectors (e.g., from shipment tracking to geo-social networks) on understanding the semantic behavior of moving objects. Semantic behavior refers to the use of semantic abstractions of the raw mobility data, including not only geometric patterns but also knowledge extracted jointly from the mobility data and the underlying geographic and application domains information. The core contribution of this article lies in a semantic model and a computation and annotation platform for developing a semantic approach that progressively transforms the raw mobility data into semantic trajectories enriched with segmentations and annotations. We also analyze a number of experiments we did with semantic trajectories in different domains.

Both sensor coverage maximization and energy cost minimization are the fundamental requirements in the design of real-life mobile sensing applications, e.g., (1) deploying environmental sensors (like CO2, fine particle measurement) on public transports to monitor air pollution, (2) analyzing smart phone embedded sensors (like GPS, accelerometer) to recognize people daily activities. However sensor coverage and energy cost contradict each other: the higher frequency mobile sensing takes, the more energy is used, and vise versa. In this paper, we design a novel two-step mobile sensing process ("OptiMoS") to achieve optimal mobile sensing that can effectively balance sensor coverage and energy cost. In the first step, OptiMoS divides the continuous mobile sensor readings into several segments, where the readings in one segment are highly-correlated rather than readings amongst different segments. In the second step, OptiMoS identifies optimal sampling for the sensor readings in each segment, where the selected readings can guarantee reasonably high sensor coverage with limited sampling rate. Various greedy & near-optimal segmentation and sampling methods are designed in OptiMoS, and are evaluated using real-life environmental data from mobile sensors.

Geo-tagged image is an ideal source for the discovery of popular travel places. However, the aspects of popular venues for daily-life purposes like dining and shopping are often missing in the mined locations from geo-tagged images. Fortunately check-in websites provide us a unique opportunity of analyzing people's preferences in their daily lives to complement the knowledge mined from geo-tagged images. This paper presents a novel approach for the discovery of Areas of Interest (AoI). By analyzing both geo-tagged images and check-ins, the approach exploits travelers' flavors as well as the preferences of daily-life activities of local residents to find AoI in a city. The proposed approach consists of two major steps. Firstly, we devise a density-based clustering method to discover AoI, mainly based on the image densities but also reinforced by the secondary densities from the images' neighboring venues. Then we propose a novel joint authority analysis framework to rank AoI. The framework simultaneously considers both the location-location transitions, and the user-location relations. An interactive presentation interface for visualizing AoI is also presented. The approach is tested with very large datasets for Shanghai city. They consist of 49,460 geo-tagged images from Panoramio.com, and 1,361,547 check-ins from the check-in website Qieke.com. By evaluating the ranking accuracy and quality of AoI, we demonstrate great improvements of our method over compared methods.

GPS devices allow recording the movement track of the moving object they are attached to. This data typically consists of a stream of spatio-temporal (x,y,t) points. For application purposes the stream is transformed into finite subsequences called trajectories. Existing knowledge extraction algorithms defined for trajectories mainly assume a specific context (e.g. vehicle movements) or analyze specific parts of a trajectory (e.g. stops), in association with data from chosen geographic sources (e.g. points-of-interest, road networks). We investigate a more comprehensive semantic annotation framework that allows enriching trajectories with any kind of semantic data provided by multiple 3rd party sources.

With the prevalence of GPS-embedded mobile devices, enormous amounts of mobility data are being collected in the form of trajectory - a stream of (x,y,t) points. Such trajectories are of heterogeneous entities - vehicles, people, animals, parcels etc. Most applications primarily analyze raw trajectory data and extract geometric patterns. Real-life applications however, need a far more comprehensive, semantic representation of trajectories. This paper demonstrates the automatic construction and visualization capabilities of SeMiTri - a system we built that exploits 3rd party information sources containing geographic information, to semantically enrich trajectories. The construction stack encapsulates several spatio-temporal data integration and mining techniques to automatically compute and annotate all meaningful parts of heterogeneous trajectories. The visualization interface exhibits different levels of data abstraction, from low-level raw trajectories (i.e. the initial GPS trace) to high-level semantic trajectories (i.e. the sequence of interesting places where moving objects have passed and/or stayed).

Data mediation and interoperation have already become one of the central topics of IT for decades. Since the Web appears, this problem has been exploded due to the increasing amount of data and Web users. This paper takes look at the social tagging behavior. It proposes upper tag ontology (UTO) to model tag data from three popular social tagging networks (del.icio.us, youTube, flickr). In this paper, we discuss how to crawel tag data and store them according to UTO in RDF triple store. Furthermore, based on integrated RDF triple store, we provide three scenarios to show the search possibility of the integrated tag data.

To actually bridge the gap between business perspective and technical perspective, the prerequisite is to provide a comprehensive process modeling framework for business processes. Different from the previous traditional process methodologies, our work is neither only industrial process graphic-modeling nor pure theoretical studies. We mainly focus on the semantically-enhanced process description model. We propose the semantic modeling framework for business processes, i.e. BPMO: basically, we determine the description requirements for the whole business process lifecycle involving process discovery, composition and execution; furthermore, we refine the comprehensive semantic Web services conceptual model WSMO and make specific extensions to realize the BPMO modeling framework.