Search

Existing techniques for solving large Multidimensional Knapsack Problems (MKP) aim at reducing the number of variables (items). While these approaches solve problems with many items efficiently, their performance declines with increasing number of constraints. We propose empirical orthogonal constraint generation (EOCG) to reduce the number of constraints. The intuition is that, geometrically, each constraint is a dimension of a hypercube, with capacity as side length, while items correspond to vectors with their weights as coordinates along the dimensions–the basis vectors. MKP problems guarantee the feasibility of a solution by one constraint on the coordinate sum for each dimension. In contrast, EOCG aims at reducing the number of dimensions to be constrained by using new basis vectors to represent the optimal item set with less coordinates. The key challenge is that a concise problem representation has to be formulated so that its solution also solves the original MKP. EOCG allows for this transfer by successively choosing new dimensions so that capacity violations on the next dimension added decline with a steep descent until a valid and optimal solution is found. We evaluate EOCG on established benchmark instances, which EOCG often solves in lower dimensions. EOCG finds the currently best-known solutions to all high-dimensional Chu and Beasley instances, improves the best-known solutions to two Glover and Kochenberger instances, and proves the optimality of a solution to another instance.

Recommender systems (RS) play a key role in e-commerce by preselecting presumably interesting products for customers. Hybrid RSs using a weighted average of individual RSs’ predictions have been widely adopted for improving accuracy and robustness over individual RSs. While for regression tasks, approaches to estimate optimal weighting schemes based on individual RSs’ out-of-sample errors exist, there is scant literature in classification settings. Class prediction is important for RSs in e-commerce, as here item purchases are to be predicted. We propose a method for estimating weighting schemes to combine classifying RSs based on the variance-covariance structures of the errors of individual models' probability scores. We evaluate the approach on a large real-world ecommerce data set from a European telecommunications provider, where it shows superior accuracy compared to the best individual model as well as a weighting scheme that averages the predictions using equal weights.

We propose a method for learning weighting schemes in weighted hybrid recommender systems (RS) that is based on statistical forecast and portfolio theory. An RS predicts the future preference of a set of items for a user, and recommends the top items. A hybrid RS combines individual RS in making the predictions. To determine the weighting of individual RS, we learn so-called optimal weights from the covariance matrix of available error data of individual RS that minimize the error of a combined RS. We test the method on the well-known MovieLens 1M dataset, and, contrary to the “forecast combination puzzle”, stating that a simple average (SA) weighting typically outperforms learned weights, the out-of-sample results show that the learned weights consistently outperform the individually best RS as well as an SA combination.

A primary task of customer relationship management (CRM) is the transformation of customer data into business value related to customer binding and development, for instance, by offering additional products that meet customers’ needs. A customer’s purchasing history (or sequence) is a promising feature to better anticipate customer needs, such as the next purchase intention. To operationalize this feature, sequences need to be aggregated before applying supervised prediction. That is because numerous sequences might exist with little support (number of observations) per unique sequence, discouraging inferences from past observations at the individual sequence level. In this paper the authors propose mechanisms to aggregate sequences to generalized purchasing types. The mechanisms group sequences according to their similarity but allow for giving higher weights to more recent purchases. The observed conversion rate per purchasing type can then be used to predict a customer’s probability of a next purchase and target the customers most prone to purchasing a particular product. The bias–variance trade-off when applying the models to target customers with respect to the lift criterion are discussed. The mechanisms are tested on empirical data in the realm of cross-selling campaigns. Results show that the expected bias–variance behavior well predicts the lift achieved with the mechanisms. Results also show a superior performance of the proposed methods compared to commonly used segmentation-based approaches, different similarity measures, and popular class predictors. While the authors tested the approaches for CRM campaigns, their parameterization can be adjusted to operationalize sequential features of high cardinality also in other domains or business functions.

Numerous forecast combination techniques have been proposed. However, these do not systematically outperform a simple average (SA) of forecasts in empirical studies. Although it is known that this is due to instability of learned weights, managers still have little guidance on how to solve this “forecast combination puzzle”, i.e., which combination method to choose in specific settings. We introduce a model determining the yet unknown asymptotic out-of-sample error variance of the two basic combination techniques: SA, where no weightings are learned, and so-called optimal weights that minimize the in-sample error variance. Using the model, we derive multi-criteria boundaries (considering training sample size and changes of the parameters which are estimated for optimal weights) to decide when to choose SA. We present an empirical evaluation which illustrates how the decision rules can be applied in practice. We find that using the decision rules is superior to all other considered combination strategies.

Nowadays corporate data centers leverage virtualization technology to cut operational and management costs. Virtualization allows splitting and assigning physical servers to virtual machines (VM) that run particular business applications. This has led to a new stream in the capacity planning literature dealing with the problem of assigning VMs with volatile demands to physical servers in a static way such that energy costs are minimized. Live migration technology allows for dynamic resource allocation, where a controller responds to overload or underload on a server during runtime and reallocates VMs in order to maximize energy efficiency. Dynamic resource allocation is often seen as the most efficient means to allocate hardware resources in a data center. Unfortunately, there is hardly any experimental evidence for this claim. In this paper, we provide the results of an extensive experimental analysis of both capacity management approaches on a data center infrastructure. We show that with typical workloads of transactional business applications dynamic resource allocation does not increase energy efficiency over the static allocation of VMs to servers and can even come at a cost, because migrations lead to overheads and service disruptions.

Financial services within corporations usually are part of an information system on which many business functions depend. As of the importance of forecast quality for financial services, means of forecast accuracy improvement, such as data-driven statistical prediction techniques and/or forecast support systems, have been subject to IS research since decades. In this paper we consider means of forecast improvement due to regular patterns in forecast revisioning. We analyze how business forecasts are adjusted to exploit possible improvements for the accuracy of forecasts with lower lead time. The empirical part bases on an unique dataset of experts' cash flow forecasts and accountants' actuals realizations of companies in a global corporation. We find that direction and magnitude of the final revision in aggregated forecasts can be related to suggested targets in earnings management, providing the means of improving the accuracy of longer-term cash flow forecasts.

The Minitrack on Service Analytics is part of the Decision Analytics, Mobile Services and Service Science Track of the 48th Annual Hawaii International Conference on System Sciences (HICSS-48) on January 5-8, 2015. Service Analytics describe all processes of capturing, processing, and analyzing data taken from a service system – in order to improve, extend, and personalize the service provided and to create new value for both the provider and the customer. The modern view on services focuses on the cocreation of value between providers and customers— leveraging knowledge, skills, and resources of both partners from an overall system point of view. In most service systems, however, the service providers have no access to data related to the service usage by their customers. On the other hand, an increasing volume of data will be collected either by the users/customers themselves (e.g. through mobile/smart phones) or by technologies like smart metering in energy services, telematics in automotive and mobility services, RFID in logistics, condition sensors usage in engineering, or data capture solutions in healthcare. Research topics addressed in this year’s minitrack and in future installations of the minitrack include the applicability of basic and advanced analytics to different service systems, the state-of-the-art of service analytics methodologies and tool-support, and the investigation of benefits resulting from the application of service analytics. This minitrack will serve as a forum for researchers and practitioners to share progress in the study of these and related themes. Submissions on, but not limited to, the following topics are encouraged: ·Web Usage Mining and Web Personalization ·Data Mining / Machine Learning applied to Services ·Recommender Systems for Services ·Social Network Analytics applied to Services ·Privacy Issues resulting from Service Analytics ·Fraud Analytics for Service Systems ·Analysis and Prediction of User Behavior in Mobile Phone Systems ·Analysis and Prediction of Driver Behavior in Traffic Situations ·Analysis and Exploitation of Floating Car Data ·Electricity Consumption Analysis using Smart Meter Data ·Analytics for Healthcare Services ·Analysis and Prediction of IT Service Demand Patterns ·Analysis of Service Problem Reports ·Industrial Service Analytics and Optimization ·Sports Analytics

Fur Anwendungen im Bereich der Industrie 4.0 stellt die zunehmende Flut an verfugbaren Daten ein enormes Potenzial zur verbesserten Planung und Steuerung von Geschafts-, Produktions- und Logistikprozessen dar. Nutzbar wird dieses Potenzial jedoch nur, wenn Unternehmen den Prozess der Datentransformation – von der Gewinnung, Integration und Analyse bis hin zur Entscheidung und Bereitstellung – in gute unternehmerische Entscheidungen in seiner Ganzheit beherrschen. Idealerweise sind Business-Anwender unabhangig von IT-Experten in der Lage, ein solches System flexibel und gezielt auf sich dynamisch andernde, oft komplexe Geschaftsumfelder anzupassen. Da aktuell verfugbare Ansatze diese Anforderungen nicht erfullen, ist es Ziel dieses Beitrages, technologische und methodologische Ansatzpunkte fur die Entwicklung einer intelligenten, auf die Anforderungen zukunftiger Fertigungs- und Logistikprozesse ausgerichteten Decision Support Pipeline im Kontext von Industrie 4.0 zu liefern. Anhand konkreter Beispiele verdeutlichen wir das Potenzial dieses integrierten, prozessorientierten Ansatzes.

Today, the frontier for using data to make business decisions has shifted, and high-performing service companies are building their competitive strategies around data-driven insights that produce impressive business results. In principle, the ever-growing amount of available data would allow for deriving increasingly precise forecasts and optimised input for planning and decision models. However, the complexity resulting from considering large volumes of high-dimensional, fine-grained, and noisy data in mathematical models leads to the fact that dependencies and developments are not found, algorithms do not scale, and traditional statistics as well as data-mining techniques collapse because of the well-known curse of dimensionality. Hence, in order to make big data actionable, the intelligent reduction of vast amounts of data to problemrelevant features is necessary and advances are required at the intersection of economic theories, service management, dimensionality reduction, advanced analytics, robust prediction, and computational methods to solve managerial decisions and planning problems., Enterprise Modelling and Information Systems Architectures, Vol 9, No 1 (2014)