Accurate forecasting of sales/consumption is particularly important for marketing because this information can be used to adjust marketing budget allocations and overall marketing strategies. Recently, online social platforms have produced an unparalleled amount of data on consumer behavior. However, two challenges have limited the use of these data in obtaining meaningful business marketing insights. First, the data are typically in an unstructured format, such as texts, images, audio, and video. Second, the sheer volume of the data makes standard analysis procedures computationally unworkable. In this study, we combine methods from cloud computing, machine learning, and text mining to illustrate how online platform content, such as Twitter, can be effectively used for forecasting. We conduct our analysis on a significant volume of nearly two billion Tweets and 400 billion Wikipedia pages. Our main findings emphasize that, by contrast to basic surface-level measures such as the volume of or sentiments in Tweets, the information content of Tweets and their timeliness significantly improve forecasting accuracy. Our method endogenously summarizes the information in Tweets. The advantage of our method is that the classification of the Tweets is based on what is in the Tweets rather than preconceived topics that may not be relevant. We also find that, by contrast to Twitter, other online data (e.g., Google Trends, Wikipedia views, IMDB reviews, and Huffington Post news) are very weak predictors of TV show demand because users tweet about TV shows before, during, and after a TV show, whereas Google searches, Wikipedia views, IMDB reviews, and news posts typically lag behind the show. Data, as supplemental material, are available at http://dx.doi.org/10.1287/mksc.2015.0972 .

Abstract. In scientific workflow systems, temporal consistency is critical to ensure the timely completion of workflow instances. To monitor and guarantee the correctness of temporal consistency, temporal constraints are often set and then verified. However, most current work adopts user specified temporal constraints without considering system performance, and hence may result in frequent temporal violations that deteriorate the overall workflow execution effectiveness. In this paper, with a systematic analysis of such problem, we propose a probabilistic strategy which is capable of setting coarse-grained and finegrained temporal constraints based on the weighted joint distribution of activity durations. The strategy aims to effectively assign a set of temporal constraints which are well balanced between user requirements and system performance. The effectiveness of our work is demonstrated by an example scientific workflow in our scientific workflow system.

To tackle the issue in deep crowd sensing, a Time and Location Correlation Incentive (TLCI) scheme is proposed for deep data gathering in crowdsourcing networks. In TLCI scheme, a metric named "Quality of Information Satisfaction Degree" (QoISD) is to quantify how much collected sensing data can satisfy the application's QoI requirements mainly in terms of data quantity and data coverage. Two incentive algorithms are proposed to satisfy QoISD with different view. The first algorithm is to ensure that the application gets the specified sensing data to maximize the QoISD. Thus, in the first incentive algorithm, the reward for data sensing is to maximize the QoISD. The second algorithm is to minimize the cost of the system while meeting the sensing data requirement and maximizing the QoISD. Thus, in the second incentive algorithm, the reward for data sensing is to maximize the QoISD per unit of reward. Finally, we compare our proposed scheme with existing schemes via extensive simulations. Extensive simulation results well justify the effectiveness of our scheme. The QoISD can be optimized by 81.92%, and the total cost can be reduced by 31.38%.

No abstract

Privacy risks of recommender systems have caused increasing attention. Users' private data is often collected by probably untrusted recommender system in order to provide high-quality recommendation. Meanwhile, malicious attackers may utilize recommendation results to make inferences about other users' private data. Existing approaches focus either on keeping users' private data protected during recommendation computation or on preventing the inference of any single user's data from the recommendation result. However, none is designed for both hiding users' private data and preventing privacy inference. To achieve this goal, we propose in this paper a hybrid approach for privacy-preserving recommender systems by combining differential privacy (DP) with randomized perturbation (RP). We theoretically show the noise added by RP has limited effect on recommendation accuracy and the noise added by DP can be well controlled based on the sensitivity analysis of functions on the perturbed data. Extensive experiments on three large-scale real world datasets show that the hybrid approach generally provides more privacy protection with acceptable recommendation accuracy loss, and surprisingly sometimes achieves better privacy without sacrificing accuracy, thus validating its feasibility in practice.