Recommending social activities, such as watching movies or having dinner is a typical function found in social networks or e-commerce sites. Besides certain websites which manage activity-related locations (e.g., foursquare.com), many items on product sale platforms (e.g, groupon.com) can naturally be mapped to social activities. For example, movie tickets can be thought of as activity items, which map to social activity "watch a movie". Traditional recommenders estimate the degree of interest by the target user to candidate activity items and, accordingly, promote the top-k activity items to the user. However, these systems ignore an important characteristic of social activities: people like to participate in them with their folks. This paper is the first one to consider this fact and improves the effectiveness of recommendation in two directions. First, we show that people, more often than not, prefer to find partners before participation in social activities. This means that if a system recommends an activity item alone, the user may give up the item, if she cannot think of a partner to attend the activity together. Therefore, we study the problem of activity-partner recommendation; i.e., for each recommended activity item, find a suitable partner for the user. This (i) saves the users time for finding activity partners, (ii) increases the likelihood that the activity item will be selected by the user, (iii) improves the effectiveness of recommender systems to users overall and enkindles their social enthusiasm. Our partner recommender is built upon the users historical attendance preferences, their social context, and geographic information. Second, we explore the impact of finding suitable partners to improve the effectiveness of recommending activities to users. Assuming that users tend to select the activities for which they can find suitable partners, we propose a partner-aware activity recommendation model, which integrates this hypothesis into conventional recommendation approaches. Our method first estimates the probability that the target user can find suitable partners for each candidate activity item, and then considers this probability when ranking them. Finally, the recommended items not only match the users' interests, but also have high chances to be selected by the users, because the users can find suitable partners to attend the corresponding activities together. We conduct experiments on real data that evaluate the effectiveness of activity-partner recommendation and partner-aware activity recommendation. The results verify that (i) suggesting partners greatly improves the likelihood that a recommended activity item is selected by the target user and (ii) considering the existence of suitable partners in the ranking of recommended items improves the accuracy of recommendation significantly.
Users preferences, and consequently their ratings and re- views to items, change over time. Likewise, characteristics of items are also time-varying. By dividing data into time periods, temporal Recommender Systems (RSs) improve recommendation accuracy by exploring the temporal dynamics in user rating data. However, temporal RSs have to cope with rating sparsity in each time period. Meanwhile, reviews generated by users contain rich information about their preferences, which can be exploited to address rating sparsity and further improve the performance of temporal RSs. In this paper, we develop a temporal rating model with topics that jointly mines the temporal dynamics of both user-item ratings and reviews. Studying temporal drifts in reviews help us understand item rating evolutions and user interest changes over time. Our model also automatically splits the review text in each time period into interim words and intrinsic words. By linking interim words and intrinsic words to short-term and long-term item features respectively, we jointly mine the temporal changes in user and item latent features together with the associated review text in a single learning stage. Through experiments on 28 real world datasets collected from Amazon, we show that the rating prediction accuracy of our model significantly outperforms the existing state-of-art RS models. And our model can automatically identify representative interim words in each time period as well as intrinsic words cross all time periods. This can be very useful in understanding the time evolution of users preferences and items characteristics.
DanMu, an emerging type of user-generated comment has been increasingly popular in recent years. Many online video platforms such as Tudou.com have provided the DanMu function. Unlike traditional online review such as reviews at Youtube.com that are outside videos, the DanMu is scrolling marquee comment, which is overlaid directly on top of the video and synchronized to a specific playback time. Such comments are displayed as streams of moving subtitles overlaid on the video screen. Viewers could easily write DanMus while watching videos and the written DanMus will be immediately overlaid onto the video and displayed to writers themselves and other viewers as well. Such DanMu systems have greatly enabled users to communicate with each other in a much more direct way creating a real-time sense of sharing watching experience. Although we see much unique natures of DanMu and great impact on online video systems, to the best of our knowledge, there is no work that has provided a comprehensive study on DanMu. In this paper, as a pilot study, we analyse the unique characteristics of DanMu from various perspectives. Specifically, we first illustrate some unique distributions of DanMus by comparing with traditional reviews (TReviews) that we collected from a real DanMu-enabled online video system. Second we discover two interesting patterns in DanMu data: herding effect and multiple-burst phenomena that are significantly different from those in TRviews and reveal important insights about the growth of DanMus on a video. Towards exploring antecedents of both herding effect and multiple-burst phenomena, we propose to further detect leading DanMus within bursts because those leading DanMus make the most contribution to both patterns. A framework is proposed to detect leading DanMus that effectively combines multiple factors contributing to lead DanMus. Based on the identified characteristics of DanMu, finally we propose to predict the distribution of future DanMus (i.e., the growth of DanMus), which is important for many DanMu-enabled online video systems. This prediction task includes two aspects: one is to predict which videos future DanMus will be posted for; the other one is to predict which segments of a video future DanMus will be posted on, we develop two sophisticated models to solve both problems. Finally, intensive experiments are conducted with a real-world data set to validate all methods developed in this paper.
The increasing availability of structured machine-processable knowledge in the Web of Data calls for machine learning methods to support standard reasoning based services (such as query-answering and logic inference). Inductive and transductive reasoning algorithms can efficiently exploit statistical regularities in the inherently incomplete knowledge bases distributed across the Web. This paper focuses on the problem of predicting missing class-memberships and property values of individual resources in Web ontologies. We propose a transductive learning method for inferring missing properties about individuals: given a class-membership/property value learning problem, we address the task of identifying relations which are likely to link similar individuals, and efficiently propagating knowledge across such (possibly diverse) relations. Our experimental evaluation demonstrates the effectiveness of the proposed method.
Internet and Web technologies have changed our lives in ways we are not even fully aware. In the near future, Internet will interconnect more than 50 billion of things of the real world, nodes will sense billions of features of interest and properties, and things will be represented with Web-based bi-directional services with high-dynamic content and real-time data. This is the new era of the Internet and the Web of Things. The emergence of such paradigms implies the evolution and integration of the systems which they interact with. Thereby, it is essential to develop abstract models for representing, and simulating the Web of Things in order to establish new approaches. A model of the Web of Things based on a structured XML representation is described in this paper. We also present a simulator whose ultimate goal is to encapsulate the expected dynamics of the Web of Things, for the future development of Information Retrieval (IR) systems. The sim- ulator generates a real-time collection of XML documents, which contain spatio-temporal contexts, textual and sensed information with highly dynamic dimensions. The simulator is characterized among others for its flexibility and versatility to represent real-world scenarios and a unique perspective from information retrieval, we tested the simulator in terms of fundamentals variables.
A Knowledge graph is a graph with entities of different types as nodes and various relations among them as edges. The constructions of knowledge graphs in the past decades facilitate many applications, such as link prediction, web search analysis, question answering, etc. Knowledge graph embedding aims to represent entities and relations in a large-scale knowledge graph as elements in a continuous vector space. Existing methods, e.g., TransE, TransH and TransR, learn the embedding representation by defining a global margin-based loss function over the data. However, the optimal loss function is determined during experiments whose parameters are examined among a closed set of candidates. Moreover, embeddings over two knowledge graphs with different entities and relations share the same set of candidate loss functions, ignoring the locality of both graphs. This leads to the limited performance of embedding related applications. In this paper, a locally adaptive translation method for knowledge graph embedding, called TransA, is proposed to find the optimal loss function by adaptively determining its margin over different knowledge graphs. Then the convergence of TransA is verified from the aspect of its uniform stability. To make the embedding methods up-to-date when new vertices and edges are added into the knowledge graph, the incremental algorithm for TransA, called iTransA, is proposed by adaptively adjusting the optimal margin over time. Experiments on two benchmark data sets demonstrate the superiority of the proposed method, as compared to the-state-of-the-art ones.
Mobile applications consume device energy for their operations, and the fast rate of battery depletion on mobile devices poses a major usability hurdle. After the display, data communication is the second-biggest consumer of mobile device energy. At the same time, software applications that run on mobile devices represent a fast-growing product segment. Typically, these applications serve as front-end display mechanisms, which fetch data from remote servers and display the information to the user in an appropriate format -- incurring significant data communication overheads in the process. In this work, we propose methods to reduce energy overheads in mobile devices due to data communication by leveraging data caching technology. A review of existing caching mechanisms revealed that they are primarily designed for optimizing performance and cannot be easily ported to mobile devices for energy savings. Further, architectural differences between traditional client-server and mobile communications infrastructures make the use of existing caching technologies unsuitable in mobile devices. In this paper, we propose a set of two new caching approaches specifically designed with the constraints of mobile devices in mind: (a) a response caching approach, and (b) an object caching approach. Our experiments show that, even for a small cache size of 250 MB, object caching can reduce energy consumption on average by 45%, compared to the no-cache case, and response caching can reduce energy consumption by 20% compared to the no-cache case. The benefits increase with larger cache sizes. These results demonstrate the efficacy of our proposed method, and raise the possibility of significantly extending mobile device battery life.
In recent times, automated business processes and web services have become ubiquitous in diverse application spaces. Efficient composition of web services in real time while providing necessary Quality of Service (QoS) guarantees is a computationally complex problem and several heuristic based approaches have been proposed to compose services optimally. In this paper, we present the design of a scalable QoS-aware service composition mechanism which balances the computational complexity of service composition with the QoS guarantees of the composed service and achieves scalability. On one hand, we handle the case of a single QoS parameter using an intelligent search and pruning mechanism in the composed service space and show that our methodology yields near optimal solution on real benchmarks. On the other hand, we handle the case of multiple QoS parameters using aggregation techniques. As a final contribution, we explore search time versus solution quality trade-off using parameterized search algorithms that produce better quality solutions at the cost of time. We present experimental results to show the efficiency of our proposed mechanism.