Accurate water demand forecasting is crucial for sustainable resource management. State-of-the-art prediction methods building on machine- and deep learning perform well in capturing complex consumption patterns in time series data and provide accurate predictions. However, a major challenge for fresh water suppliers that has remained widely unaddressed so long, is how to transfer prediction models of different geographic regions to another area for which there is no specific knowledge or data available. We call this concept spatial transferability. Based on a real-world data set of ten regions in North-East Italy, we propose and compare different approaches to enable spatial transferability of prediction models. Our approaches build on the one hand on the aggregation of the available data sets, i.e. the combination of the different training inputs, and on the other hand on ensemble techniques, i.e. the combination of the different models' outputs. The underlying basic model types range from rather simple linear models to more complex ones, like long short term memory. Results show that by applying our proposed techniques, the induced transfer error can be kept below one percent on average.

A common issue for companies is that the volume of product orders may at times exceed the production capacity. We formally introduce two novel problems dealing with the question which orders to discard or postpone in order to meet certain (timeliness) goals, and try to approach them by means of model-based diagnosis. In thorough analyses, we identify many similarities of the introduced problems to diagnosis problems, but also reveal crucial idiosyncracies and outline ways to handle or leverage them. Finally, a proof-of-concept evaluation on industrial-scale problem instances from a well-known scheduling benchmark suite demonstrates that one of the two formalized problems can be well attacked by out-of-the-box model-based diagnosis tools.

S.516-523

Optimal production planning in the form of job shop scheduling problems (JSSP) is a vital problem in many industries. In practice, however, it can happen that the volume of jobs (orders) exceeds the production capacity for a given planning horizon. A reasonable aim in such situations is the completion of as many jobs as possible in time (while postponing the rest). We call this the Job Set Optimization Problem (JOP). Technically, when constraint programming is used for solving JSSPs, the formulated objective in the constraint model can be adapted so that the constraint solver addresses JOP, i.e., searches for schedules that maximize the number of timely finished jobs. However, also highly specialized solvers which proved very powerful for JSSPs may struggle with the increased complexity of the reformulated problem and may fail to generate a JOP solution given practical computation timeouts. As a remedy, we suggest a framework for solving multiple randomly modified instances of a relaxation of the JOP, which allows to gradually approach a JOP solution. The main idea is to have one module compute subset-minimal job sets to be postponed, and another one effectuating that random job sets are found. Different algorithms from literature can be used to realize these modules. Using IBM’s cutting-edge CP Optimizer suite, experiments on well-known JSSP benchmark problems show that using the proposed framework consistently leads to more scheduled jobs for various computation timeouts than a standalone constraint solver approach.

Constraint answer set programming (CASP) is a family of hybrid approaches integrating answer set programming (ASP) and constraint programming (CP). These hybrid approaches have already proven to be very successful in various domains. In this paper we present first evaluation results for the CASP solver ASCASS, which provides novel methods for defining and exploiting problem-dependent search heuristics. Beyond the possibility of using already built-in problem-independent heuristics, ASCASS allows on the ASP level the definition of problem-dependent variable selection, value selection and pruning strategies, which guide the search of the CP solver. The proof-of-concept evaluation was carried out on benchmark instances of the real world Partner Units Problem (PUP). Due to a sophisticated heuristic, which cannot be represented by other ASP or CASP solvers, ASCASS shows superior performance.

Constraint Satisfaction Problems (CSPs) have the big advantage of a succinct, declarative and easy to understand representation form. Unfortunately, solving CSPs is NP-complete in the general case. In order to cope with this, common CSP frameworks offer the possibility to use different built-in heuristics. However, the provided built-in heuristics are often not suitable to significantly boost solution calculation. Also the facilities for expressing domain-specific heuristics in a declarative manner within the CSP framework are typically very limited (e.g. by defining a static variable selection order)and thus are often not applicable. As a consequence such domain-specific heuristics are often implemented by means of custom propagators or custom constraints (e.g. a special constraint for bin packing problems) forcing domain experts and knowledge engineers to leave the declarative world and implement the heuristics in a procedural manner. In this paper we propose a new declarative language for expressing domain specific heuristics for CSPs which can be easily integrated in every CSP framework. We also describe a prototype implementation within a state-of-the-art CSP solver and present proof of concept results on real world configuration problem instances.

The Partner Units Problem (PUP) constitutes a challenging real-world configuration problem with diverse application domains such as railway safety, security monitoring, electrical engineering, or distributed systems. Although using the latest problem-solving methods including Constraint Programming, SAT Solving, Integer Programming, and Answer Set Programming, current methods fail to generate solutions for midsized real-world problems in acceptable time. This paper presents the QuickPup algorithm based on backtrack search combined with smart variable orderings and restarts. QuickPup outperforms the available methods by orders of magnitude and thus makes it possible to automatically solve problems which couldn’t be solved without human expertise before. Furthermore, the runtimes of QuickPup are typically below one second for real-world problem instances.

The Partner Units Problem (PUP) is a computationally challenging configuration problem with diverse application domains, such as railway safety or electrical engineering. The recently formulated Quick Pup algorithm has made it possible for the first time to automatically derive solutions for real world-sized problems, such that the Partner Units Configuration Problem can be seen as practically solved. Further challenges, which were clearly out of reach before, can be tackled now. This paper addresses two practically relevant problems related to the PUP. The first problem is to calculate solutions for legacy instances of the PUP. The difference to the original PUP is that the solver is given a partial solution which has to be extended to a complete solution. The second problem is about adapting partial legacy solutions which cannot be extended to complete solutions and thus making a consistent complete solution possible. All presented approaches are evaluated on behalf of a newly defined set of benchmark instances.

Recommender systems are common web applications which support users in finding suitable products in large and/or complex product domains. Although state-of-the-art systems manage to accomplish the task of finding and presenting suitable products they

Constraint sets can become inconsistent in different contexts. We are interested in identifying minimal sets of constraints that have to be adapted or deleted in order to restore consistency. In this paper we sketch a highly efficient divide-and-conquer based diagnosis approach which identifies minimal sets of faulty constraints in a given over-constrained problem. This approach is specifically applicable in scenarios where the efficient identification of leading (preferred) diagnoses is crucial.

Recommender systems are wide-spread web applications which can effectively support users in finding suitable products in a large and/or complex product domain. Although state-of-the-art systems manage to accomplish the task of finding and presenting suitable products they show big deficits in the applied model of human behavior. Time limitations, cognitive capacities, and willingness to cognitive effort bound rational decision taking which can lead to unforeseen side effects and furthermore to sub-optimal decisions. Decoy effects are cognitive phenomenons which are omni-present on result pages. State-of-the-art recommender systems are completely unaware of such effects. Due to the fact that such effects constitute one source of irrational decisions their identification and, if necessary, the neutralization of their biasing potential is extremely important. This paper introduces an approach for identifying and minimizing decoy effects on recommender result pages. To undergird the presented approach we present the results of a corresponding user study which clearly proofs the concept.

Recommenders support effective product retrieval processes for online users. These systems propose repair actions in situations where no solution can be found and derive recommendations including a set of explanations as to why a certain product has been selected. In this context utility constraints (scoring rules) have to be defined which specify the way utilities of products, explanations, and repair alternatives are determined. Such constraints can be faulty which means that they calculate rankings in a way not expected by marketing and sales experts. The maintenance and repair of such constraints is an extremely error-prone task. In this paper we present an intelligent environment which supports the automated adaptation of faulty utility constraints taking into account existing marketing and sales requirements. In this context we discuss experiences from commercial projects.

Internet users often face the challenge of identifying the most suitable product out of some product assortment available on a certain e-sales platform. Recommender systems can substantially alleviate this typically complex task. Since the rise of such systems a lot of effort has been done in developing different recommendation approaches and algorithms, which all of them have certain strengths and weaknesses. What has been widely ignored by the recommender community so far are the potentials and impacts of psychological and decision theoretical phenomenons, which already have been investigated and applied in the field of marketing. Such phenomenons promise big capability to support users in decision making when facing a comparison situation. This paper concentrates on two classes of phenomenons, which are decoy effects and serial position effects. Tightly coupled to these phenomenons is the problem of getting the utility function of a recommender right, as this function serves both as the basis of result set calculation as well as the fundament of exploitation of above mentioned phenomenons. Putting all these aspects together an extended architecture for recommender systems will be proposed in the end of the paper.

Recommender applications support decision-making processes by helping online customers to identify products more effectively. Recommendation problems have a long history as a successful application area of Artificial Intelligence (AI) and the interest in recommender applications has dramatically increased due to the demand for personalization technologies by large and successful e-Commerce environments. Knowledge-based recommender applications are especially useful for improving the accessibility of complex products such as financial services or computers. Such products demand a more profound knowledge from customers than simple products such as CDs or movies. In this paper we focus on a discussion of AI technologies needed for the development of knowledge-based recommender applications. In this context, we report experiences from commercial projects and present the results of a study which investigated key factors influencing the acceptance of knowledge-based recommender technologies by end-users.