Expert Systems with Applications: X最新文献

Data extracted from software repositories is used intensively in Software Engineering research, for example, to predict defects in source code. In our research in this area, with data from open source projects as well as an industrial partner, we noticed several shortcomings of conventional data mining approaches for classification problems: (1) Domain experts’ acceptance is of critical importance, and domain experts can provide valuable input, but it is hard to use this feedback. (2) Evaluating the quality of the model is not a matter of calculating AUC or accuracy. Instead, there are multiple objectives of varying importance with hard to quantify trade-offs. Furthermore, the performance of the model cannot be evaluated on a per-instance level in our case, because it shares aspects with the set cover problem. To overcome these problems, we take a holistic approach and develop a rule mining system that simplifies iterative feedback from domain experts and can incorporate the domain-specific evaluation needs. A central part of the system is a novel multi-objective anytime rule mining algorithm. The algorithm is based on the GRASP-PR meta-heuristic but extends it with ideas from several other approaches. We successfully applied the system in the industrial context. In the current article, we focus on the description of the algorithm and the concepts of the system. We make an implementation of the system available.

Deep Learning (DL) has recently become a topic of study in different applications including healthcare, in which timely detection of anomalies on Electrocardiogram (ECG) can play a vital role in patient monitoring. This paper presents a comprehensive review study on the recent DL methods applied to the ECG signal for the classification purposes. This study considers various types of the DL methods such as Convolutional Neural Network (CNN), Deep Belief Network (DBN), Recurrent Neural Network (RNN), Long Short-Term Memory (LSTM), and Gated Recurrent Unit (GRU). From the 75 studies reported within 2017 and 2018, CNN is dominantly observed as the suitable technique for feature extraction, seen in 52% of the studies. DL methods showed high accuracy in correct classification of Atrial Fibrillation (AF) (100%), Supraventricular Ectopic Beats (SVEB) (99.8%), and Ventricular Ectopic Beats (VEB) (99.7%) using the GRU/LSTM, CNN, and LSTM, respectively.

Citation screening is a labour-intensive part of the process of a systematic literature review that identifies citations eligible for inclusion in the review. In this paper, we present an automatic text classification approach that aims to prioritise eligible citations earlier than ineligible ones and thus reduces the manual labelling effort that is involved in the screening process. e.g. by automatically excluding lower ranked citations. To improve the performance of the text classifier, we develop a novel neural network-based feature extraction method. Unlike previous approaches to citation screening that employ unsupervised feature extraction methods to address a supervised classification task, our proposed method extracts document features in a supervised setting. In particular, our method generates a feature representation for documents, which is explicitly optimised to discriminate between eligible and ineligible citations.

The generated document representation is subsequently used to train a text classifier.

Experiments show that our feature extraction method obtains average workload savings of 56% when evaluated across 23 medical systematic reviews. The proposed method outperforms 10 baseline feature extraction methods by approximately 6% in terms of the WSS@95% metric.

This paper examines the role of textual and unstructured data in the credit risk assessment of sovereigns. Specifically, in this paper, a novel approach to understand and predict sovereign ratings is proposed. For that purpose, information embedded in the annual country reports issued by the European Commission is used. The model employs a neural-network-based document embedding known as document to vector (Doc2Vec) to convert each country report into a numerical vector, which is then used as features into a logistic regression. The model is trained using information from 2011 to 2019 and it correctly predicts the 70.27% of country ratings in the test sample, improving slightly the results obtained using only macroeconomic variables.

This paper deals with the Unrelated Parallel Machine scheduling problem with Setups and Resources (UPMSR) with the objective of minimizing makespan. Processing times and setups depend on machine and job. The necessary resources could be: specific resources for processing, needed for processing a job on a machine; specific resources for setups, needed to do the previous setup before a job is processed on a machine; shared resources, understanding these as unspecific resources that could also be needed in both processing or setup. The number of scarce resources depends on machine and job. As an industrial example, in a plastic processing plant molds are the specific resource for processing machines, cleaning equipment is the specific resource for setups and workers are the unspecific shared resource to operate processing machines and setup cleaning equipment. A mixed integer linear program is presented to model this problem. Also a three phase algorithm based on mathematical exact method is introduced. Model and algorithm are tested in a comprehensive and extensive computational campaign. Tests show good results for different combinations of useE of resources and in most cases come to less than 2.7% of gap against lower bound for instances of 400 jobs.

This paper provides a dependency based framework for converting natural language to SPARQL. We present a tool known as PAROT (which echos answers from ontologies) which is able to handle user’s queries that contain compound sentences, negation, scalar adjectives and numbered list. PAROT employs a number of dependency based heuristics to convert user’s queries to user’s triples. The user’s triples are then processed by the lexicon into ontology triples. It is these ontology triples that are used to construct SPARQL queries. From the experiments conducted, PAROT provides state of the art results.

In this study is examined the role of inventories and assets in the financial and shareholder value creation of a company. Research builds several Data Envelopment Analysis (DEA) models (staged), and tests their connections with each other. Research concerns publicly traded manufacturing and trade companies of Finland and three Baltic States (Estonia, Latvia and Lithuania) during the years 2010–2018. Logical and in two stages proceeding DEA efficiency model gets statistical significance, and there is support that inventory and asset related measures will lead to revenue, profits and cash flow, which together will eventually result in higher shareholder value (like stated in operations and supply chain management theories such as theory of constraints). However, this finding has weakness as explanation power is low, and there is a lot of noise. It could also be so that inventories and assets are part of bunch of other inputs, which together directly create shareholder value. Therefore, it remains as an open question whether inventory and assets should be managed through classical and logical stages in companies through organization hierarchy, or if inventory and assets should be just a part of group of factors, which together aim to increase shareholder value.

The manufacturing of products at low maturity levels (referred to as physical product development) requires knowledge intensive nonconformance problem solving, yet constituting a major difficulty in industry. Due to the exponential increase of failure cost during the product development process however, problems have to be effectively remedied as early as possible. Facing shortened innovation cycles, problem solving efficiency simultaneously constitutes a competitive factor. The purpose of this theoretical review is therefore the analysis of relevant approaches contributing to knowledge-based problem solving in physical product development, to synthesize a comprehensive construct as well as to derive novel conceptualizations. The latter demonstrably emerges from natural language processing, case ontologies and machine-/deep learning support, embedded in a distributed case-based reasoning architecture. Building on this, we likewise encourage researchers and professionals to propose new studies dedicated to the field of problem solving in physical product development.