Search

Inferring user profiles based on texts created by users on social networks has a variety of applications in recommender systems such as job offering, item recommendation, and targeted advertisement. The problem becomes more challenging when working with short texts like tweets on Twitter, or posts on Facebook. This work aims at proposing an integrated framework based on Dempster–Shafer theory of evidence, word embedding, and k -means clustering for user profiling problem, which is capable of not only working well with short texts but also dealing with uncertainty inherently in user texts. The proposed framework is essentially composed of three phases: (1) Learning abstract concepts at multiple levels of abstraction from user corpora; (2) Evidential inference and combination for user modeling; and (3) User profile extraction. Particularly, in the first phase, a word embedding technique is used to convert preprocessed texts into vectors which capture semantics of words in user corpus, and then k -means clustering is utilized for learning abstract concepts at multiple levels of abstraction, each of which reflects appropriate semantics of user profiles. In the second phase, by considering each document in user corpus as an evidential source that carries some partial information for inferring user profiles, we first infer a mass function associated with each user document by maximum a posterior estimation, and then apply Dempster’s rule of combination for fusing all documents’ mass functions into an overall one for the user corpus. Finally, in the third phase, we apply the so-called pignistic probability principle to extract top- n keywords from user’s overall mass function to define the user profile. Thanks to the ability of combining pieces of information from many documents, the proposed framework is flexible enough to be scaled when input data coming from not only multiple modes but different sources on web environments. Besides, the resulting profiles are interpretable, visualizable, and compatible in practical applications. The effectiveness of the proposed framework is validated by experimental studies conducted on datasets crawled from Twitter and Facebook.

Argumentative agents in AI are inspired by how humans reason by exchange of arguments. Given the same set of arguments possibly attacking one another (Dung's AA framework) these agents are bound to accept the same subset of those arguments (aka extension) unless they reason by different argumentation semantics. However humans may not be so predictable, and in this paper we assume that this is because any real agent's reasoning is inevitably influenced by her own preferences over the arguments. Though such preferences are usually unobservable, their effects on the agent's reasoning cannot be washed out. Hence by reconstructing her reasoning process, we might uncover her hidden preferences, which then allow us to predict what else the agent must accept. Concretely we formalize and develop algorithms for such problems as uncovering the hidden argument preference relation of an agent from her expressed opinion, by which we mean a subset of arguments or attacks she accepted from a given AA framework; and uncovering the collective preferences of a group from a dataset of individual opinions. A major challenge we addressed in this endeavor is to deal with “answer sets” of argument preference relations which are generally exponential or even infinite. So we start by developing a compact representation for such answer sets called preference states. Preference revelation tasks are then structured as derivations of preference states from data, and reasoning prediction tasks are reduced to manipulations of derived preference states without enumerating the underlying (possibly infinite) answer sets. We also apply the presented results to two non-trivial problems: learning preferences over rules in structured argumentation with priorities – an open problem so far; and analyzing public polls in apparently deeper ways than existing social argumentation frameworks allow.

Bagasse’s pyrolysis oil has low calorific value, high water content, and acidity. The main components of the pyrolysis oil are acids/esters, alcohols, aldehydes/ketones, furan compounds, phenol com...

Currently, the number of deliveries handled by transportation logistics is rapidly increasing because of the significant growth of the e-commerce industry, resulting in the need for improved functional vehicle routing measures for logistic companies. The effective management of vehicle routing helps companies reduce operational costs and increases its competitiveness. The vehicle routing problem (VRP) seeks to identify optimal routes for a fleet of vehicles to deliver goods to customers while simultaneously considering changing requirements and uncertainties in the transportation environment. Due to its combinatorial nature and complexity, conventional optimization approaches may not be practical to solve VRP. In this paper, a new optimization model based on reinforcement learning (RL) and a complementary tree-based regression method is proposed. In our proposed model, when the RL agent performs vehicle routing optimization, its state and action are fed into the tree-based regression model to assess whether the current route is feasible according to the given environment, and the response received is used by the RL agent to adjust actions for optimizing the vehicle routing task. The procedure repeats iteratively until the maximum iteration is reached, then the optimal vehicle route is returned and can be utilized to assist in decision making. Multiple logistics agency case studies are conducted to demonstrate the application and practicality of the proposed model. The experimental results indicate that the proposed technique significantly improves profit gains up to 37.63% for logistics agencies compared with the conventional approaches.

This paper aims to develop a new efficient framework named Secure Decentralized Training Framework (SDTF) for Privacy Preserving Deep Learning models. The main feature of the proposed framework is its capable of working on a decentralized network setting that does not need a trusted third-party server while simultaneously ensuring the privacy of local data with a low cost of communication bandwidth. Particularly, we first propose a so-called Efficient Secure Sum Protocol (ESSP) that enables a large group of parties to jointly calculate a sum of private inputs. ESSP can work not only with integer number but also with floating point number without any data conversion. We then propose a Secure Model Sharing Protocol that enables a group of parties securely train and share the local models to be aggregated into a global model. Secure Model Sharing Protocol exploits randomization techniques and ESSP to protect local models from any honest-but-curious party even n - 2 of n parties colluding. Eventually, these protocols are employed for collaborative training decentralized deep learning models. We conduct theoretical evaluation of privacy and communication cost as well as empirical experiments on balance class image datasets (MNIST) and an unbalance class text dataset (UCI SMS Spam). These experiments demonstrate the proposed approach can obtain high accuracy (i.e. 97% baseline accuracy in only 10 training rounds with MNIST, 100 training rounds with SMS Spam) and robust to the heterogeneity decentralized network, with non-IID and unbalance data distributions. We also show a reduction in required rounds of training to achieve the accuracy baseline by 5 × as compared to Downpour SGD. It is shown that the proposed approach can achieve both the privacy at the level of cryptographic approaches and efficiency at the level of randomization techniques, while it also retains higher model’s utility than differential privacy approaches.

This paper proposes a new outlier detection approach that measures the degree of outlierness for each instance in a given dataset. The proposed model utilizes a mass-based dissimilarity measure to address the weaknesses of neighbor-based outlier models while detecting local outliers in the dataset within a variety of data point densities. In particular, it first applies a hierarchical partitioning technique to generate a set of tree-like nested structure partitions for the input dataset, and then a mass-based dissimilarity measure is defined to quantify the dissimilarity between two data instances given the generated hierarchical partition structure. After that, for each data instance, a context set is obtained by gathering the neighbors around it with the $k$ lowest mass dissimilarities, and based on those context sets, a mass-based local outlier score model is introduced to compute the outlierness for each individual instance. The proposed approach fundamentally changes the perspective of the outlier model by using the mass-based measurement instead of the distance-based functions used in most neighbor-based methods. A comprehensive experiment conducted on both synthetic and real-world datasets demonstrates that the proposed approach is not only competitive with the existing state-of-the-art outlier detection models but is also an efficient and effective alternative for local outlier detection methods.

In this paper, we aim at developing a new collaborative filtering recommender system using soft ratings, which is capable of dealing with both imperfect information about user preferences and the sparsity problem. On the one hand, Dempster-Shafer theory is employed for handling the imperfect information due to its advantage in providing not only a flexible framework for modeling uncertain, imprecise, and incomplete information, but also powerful operations for fusion of information from multiple sources. On the other hand, in dealing with the sparsity problem, community context information that is extracted from the social network containing all users is used for predicting unprovided ratings. As predicted ratings are not a hundred percent accurate, while the provided ratings are actually evaluated by users, we also develop a new method for calculating user-user similarities, in which provided ratings are considered to be more significant than predicted ones. In the experiments, the developed recommender system is tested on two different data sets; and the experiment results indicate that this system is more effective than CoFiDS, a typical recommender system offering soft ratings.

The performance of a partitional clustering algorithm is influenced by the initial random choice of cluster centers. Different runs of the clustering algorithm on the same data set often yield different results. This paper addresses that challenge by proposing an algorithm named k-PbC, which takes advantage of non-random initialization from the view of pattern mining to improve clustering quality. Specifically, k-PbC first performs a maximal frequent itemset mining approach to find a set of initial clusters. It then uses a kernel-based method to form cluster centers and an information-theoretic based dissimilarity measure to estimate the distance between cluster centers and data objects. An extensive experimental study was performed on various real categorical data sets to draw a comparison between k-PbC and state-of-the-art categorical clustering algorithms in terms of clustering quality. Comparative results have revealed that the proposed initialization method can enhance clustering results and k-PbC outperforms compared algorithms for both internal and external validation metrics.

Mining uncertain frequent patterns (UFPs) from uncertain databases was recently introduced, and there are various approaches to solve this problem in the last decade. However, systems are often faced with the problem of too many UFPs being discovered by the traditional approaches to this issue, and thus will spend a lot of time and resources to rank and find the most promising patterns. Therefore, this paper introduces a task named mining top-k UFPs from uncertain databases. We then propose an efficient method named TUFP (mining Top-k UFPs) to carry this out. Effective threshold raising strategies are introduced to help the proposed algorithm reduce the number of generated candidates to enhance the performance in terms of the runtime as well as memory usage. Finally, several experiments on the number of generated candidates, mining time, memory usage and scalability of TUFP and two state-of-the-art approaches (CUFP-mine and LUNA) were conducted. The performance studies show that TUFP is efficient in terms of mining time, memory usage and scalability for mining top-k UFPs.

The imbalanced dataset is a crucial problem found in many real-world applications. Classifiers trained on these datasets tend to overfit toward the majority class, and this problem severely affects classifier accuracy. This ultimately triggers a large cost to cover the error in terms of misclassifying the minority class especially in credit-granting decision when the minority class is the bad loan applications. By comparing the industry standard with well-known machine learning and ensemble models under imbalance treatment approaches, this study shows the potential performance of these models towards the industry standard in credit scoring. More importantly, diverse performance measurements reveal different weaknesses in various aspects of a scoring model. Employing class balancing strategies can mitigate classifier errors, and both homogeneous and heterogeneous ensemble approaches yield the best significant improvement on credit scoring.

High-utility itemset mining (HUIM) has become a key phase of the pattern mining process, which has wide applications, related to both quantities and profits of items. Many algorithms have been proposed to mine high-utility itemsets (HUIs). Since these algorithms often return a large number of discovered patterns, a more compact and lossless representation has been proposed. The recently proposed closed high utility itemset mining (CHUIM) algorithms were designed to work with certain types of databases (e.g., those without probabilities). In fact, real-world databases might contain items or itemsets associated with probability values. To effectively mine frequent patterns from uncertain databases, several techniques have been developed, but there does not exist any method for mining CHUIs from this type of databases. This work presents a novel and efficient method without generating candidates, named CPHUI-List, to mine closed potential high-utility itemsets (CPHUIs) from uncertain databases. The proposed algorithm is DFS-based and utilizes the downward closure property of high transaction-weighted probabilistic mining to prune non-CPHUIs. It can be seen from the experiment evaluations that the proposed algorithm has better execution time and memory usage than the CHUI-Miner.

Multimodal transportation has become a main focus of logistics systems due to environmental concerns, road safety issues, and traffic congestion. Consequently, research and policy interests in multimodal freight transportation problems are increasing. However, there are major challenges in the development of multimodal transportation associated with inherent risks and numerous uncertainties. Since risks are potential threats that directly impact logistics and transportation systems, comprehensive risk analysis should be carried out. Risk analysis is a critical process of identifying and analyzing significant issues to help industry mitigate those risks. However, identifying and prioritizing risks is more complex because of the ambiguity of the relevant data. This study proposes the integration of the fuzzy analytic hierarchy process (FAHP) and data envelopment analysis (DEA) for identifying and assessing quantitative risks. The proposed FAHP-DEA methodology uses the FAHP method to determine the weights of each risk criterion. The DEA method is employed to evaluate the linguistic variables and generate the risk scores. The simple additive weighting (SAW) method is used to aggregate risk scores under different risk criteria into an overall risk score. A case study of the coal industry demonstrates that the proposed risk analysis model is practical and allows users to more accurately prioritize risks while selecting an optimal multimodal transportation route. The process raises user's attention to the high-priority risks and is useful for industries in optimizing a multimodal transportation route under risk decision criteria.

Machine learning and data mining techniques have been developed rapidly in recent times. In tasks such as classification, machine learning techniques have been shown to equal to and even surpass human performance. However, high performance models are usually complex, opaque and have low interpretability thus making it difficult to explain the underlying behaviors of those models that lead to the final outcomes. In many domains such as medicine and healthcare, interpretability is one of the most important factors when considering the adoption of those models. In this paper, we propose a two-stage binary classification system applicable for healthcare (or general) data that benefits from a high level of interpretability and can at the same time achieve the results comparable to commonly used classification techniques. The motivation behind the proposed system is the lack of effective classification methods for handling data generated by various distributions (such as healthcare or banking data) that can harmonize both performance and interpretability perspectives. In this work, we tackle the problem by applying divide and conquer strategy on a new disentangled representation of the underlying data. The merit of our system is evaluated by a classification experiment with a wide range of real data and popular transparent and black-box models. Furthermore, a use case in data of sepsis patients staying in the ICU (Intensive Care Unit) is depicted to prove the interpretability of the proposed model.

In mitigating occupational hazards, there is often a need to use administrative controls such as job rotation over a prolonged period until the hazards can be eliminated or mitigated to safe levels. This research develops a noise-safe job-rotation optimization model that accounts for learning, forgetting, and boredom effects. Our analysis focuses on the case of human-paced and labor-intensive operations, considering the trade-off between safety and productivity. A case of multi-skilled workers that have heterogeneous skill levels with varying problem sizes is used to demonstrate the model s capabilities. A genetic algorithm and a randomized greedy algorithm are developed and shown to be effective in solving large-scale safe job rotation problems. Our results also show how the boredom and forgetting effects create productivity delays when job rotation is used.

This article aims to propose a novel hybrid forecasting model involving autoregressive integrated moving average (ARIMA), artificial neural networks (ANNs) and k-means clustering. The single models and k-means clustering are used to build the hybrid forecasting models in different levels of complexity (i.e. ARIMA; hybrid model of ARIMA and ANNs; and hybrid model of k-means, ARIMA, and ANN). To obtain the final forecasting value, the forecasted values of these three models are combined with the weights generated from the discount mean square forecast error (DMSFE) method. The proposed model is applied to three well-known data sets: Wolf's sunspot, Canadian lynx and the exchange rate (British pound to US dollar) to evaluate the prediction capability in three measures (i.e. MSE, MAE, and MAPE). In addition, the prediction performance of the proposed model is compared to ARIMA; ANNs; Khashei and Bijari's model; and the hybrid model of k-means, ARIMA, and ANN. The obtained results show that the proposed model gives the best performance in MSE, MAE, and MAPE for all three data sets.

Millions of players are active on gaming service platforms such as Twitch and Steam every month, but few researchers pay attention to their comments and experience. Among many types of games, hardcore games attract game enthusiasts with highly great difficulty. Hardcore games require players to invest much time to learn, so they have the highest engagement in the game genre. The Dark Souls series has always been a landmark masterpiece in hardcore games. It attracts many hardcore players with its challenging, subversive narrative, rich gameplay, and builds. The analysis of the success of the Dark Souls series is meaningful and inspiring for game developers. Many studies have analyzed the impact of Dark Souls games on player behaviors from a psychological and cultural perspective in current Game Research. However, few studies have investigated the attraction of the Dark Souls series from the players' perspective. Therefore, this research uses a topic model on massive review data of the Dark Soul series on Steam to reveal players' concerns. It aims to explore the charm of the Dark Souls series. The Dark Souls trilogy of three games produced by From Software and are available on Steam. Steam is the most significant worldwide digital distribution platform that offers comprehensive service, including installing and automatically updating games and community features such as friends lists and groups, cloud storage, and in-game voice and chat functionality. Since both Dark Souls (DS1) and Dark Souls 3 (DS3) was highly rated in this franchise, this study focuses on the masterpiece of the trilogy: Dark Souls. In the experiment, we collected a new review dataset of the Dark Souls series from Steam, including approximately 100,000 DS3 and 23,000 DS1 reviews up to October 2021. Then we used the Latent Dirichlet Allocation (LDA) model to categorize reviews from both datasets. Consequently, we uncovered 15 and 14 topics from DS1 and DS3, respectively. Among them, we used 13 topics that appear in both games to find out topics with high frequencies and positive ratings. The results indicate that most game players are concerned with five common topics: "experience", "combat", "character", "item", and "difficulty”, which are related to in-game items and boss fight design. Specifically, these topics appear in both games with a frequency and positive rating of over 10%. We also found stratification on "device" significantly lower than other topics, reflecting the developer's less optimizing game control when DS series were ported. Generally, the analysis results from this research provide high interpretability that can further support other studies in this field.

Daily electricity peak load forecasting is important for electricity generation capacity planning. Accurate forecasting leads to saving on excessive electricity generating capacity, while maintaining the stability of the power system. The main challenging tasks in this research field include improving forecasting accuracy and reducing computational time. This paper proposes a hybrid model involving variational mode decomposition (VMD), empirical mode decomposition (EMD), fast Fourier transform (FFT), stepwise regression, similar days selection (SD) method, and artificial neural network (ANN) for daily electricity peak load forecasting. Stepwise regression and similar days selection method are used for input variable selection. VMD and FFT are applied for data decomposition and seasonality capturing, while EMD is employed for determining an appropriate decomposition level for VMD. The hybrid model is constructed to effectively forecast special holidays, which have different patterns from other normal weekdays and weekends. The performance of the hybrid model is tested with real electricity peak load data provided by the Electricity Generating Authority of Thailand, the leading power utility state enterprise under the Ministry of Energy. Experimental results show that the hybrid model gives the best performance while saving computation time by solving the problems in input variable selection, data decomposition, and imbalance data of normal and special days in the training process.

Measuring the similarity between materials is essential for estimating their properties and revealing the associated physical mechanisms. However, current methods for measuring the similarity between materials rely on theoretically derived descriptors and parameters fitted from experimental or computational data, which are often insufficient and biased. Furthermore, outliers and data generated by multiple mechanisms are usually included in the dataset, making the data-driven approach challenging and mathematically complicated. To overcome such issues, we apply the Dempster–Shafer theory to develop an evidential regression-based similarity measurement (eRSM) method, which can rationally transform data into evidence. It then combines such evidence to conclude the similarities between materials, considering their physical properties. To evaluate the eRSM, we used two material datasets, including 3[Formula: see text] transition metal–4[Formula: see text] rare-earth binary and quaternary high-entropy alloys with target properties, Curie temperature, and magnetization. Based on the information obtained on the similarities between the materials, a clustering technique is applied to learn the cluster structures of the materials that facilitate the interpretation of the mechanism. The unsupervised learning experiments demonstrate that the obtained similarities are applicable to detect anomalies and appropriately identify groups of materials whose properties correlate differently with their compositions. Furthermore, significant improvements in the accuracies of the predictions for the Curie temperature and magnetization of the quaternary alloys are obtained by introducing the similarities, with the reduction in mean absolute errors of 36% and 18%, respectively. The results show that the eRSM can adequately measure the similarities and dissimilarities between materials in these datasets with respect to mechanisms of the target properties.

Fuzzy cluster analysis plays an essential role in addressing unclear boundaries between clusters in data and aims to group objects into fuzzy clusters based on their similarities. In this paper, we propose a new method for fuzzy clustering of data with categorical attributes. Specifically, we first introduce a method for kernel-based representation of cluster centers in which the underlying distribution of categorical values within a cluster center is estimated as a weighted sum of the uniform distribution and their frequency distribution. We then extend the k-centers clustering method by applying this newly proposed method of cluster center presentation for fuzzy clustering of categorical data. The effectiveness and efficiency of the proposed method are demonstrated by conducting experiments on 16 realworld datasets and comparing the results with those of existing methods. In addition, our research can be regarded as the first attempt to apply a fuzzy silhouette scoring method that includes internal coherence and external separation of fuzzy clusters into clustering of categorical data.

Unsupervised anomaly detection in high-dimensional data is crucial for both machine learning research and industrial applications. Over the past few years, the logistics agencies’ operation efficiency decreased due to the lack of understanding how best to handle potential client requests, while current anomaly detection approaches might be inefficient in distinguishing normal and abnormal behaviours from the high-dimensional data. Although previous studies continue to improve detection models, they suffer from the inability to preserve vital information while performing a dimensional reduction process. In this study, the authors aim to improve anomaly detection by proposing an ensemble method that is built and trained on two hybrid models. Eventually, after two trained hybrid models were introduced, an ensemble probability rule was applied to combine their prediction results for performing final decision-making of anomaly detection. To demonstrate the practical use of our proposed model, we have set up a case study with a logistics agency and the experiment shows that the proposed model improved accuracy by 0.88 over current models.

Decision support for Supplier Selection and Order Allocation (SSOA) is an important application area of multiple criteria decision making (MCDM) problems. In Amid et al. (Int J Prod Econ 131(1):139–145, 2011) proposed and developed a weighted maximin model to ensure the weight-consistent solution for SSOA in an MCDM problem under an uncertain environment. Essentially, this model is based on a weight-consistent constraint and a maximin aggregation operator. This paper reanalyzes the weighted maximin model in terms of the weight-consistent constraint, and then proposes a general weight-consistent model for SSOA in MCDM problems under uncertainty. In this paper, two existing models are reviewed and compared with the proposed model. Three datasets with different ranges of fuzzy demand and full factorial patterns of criteria weights are used to test the performances of the related models. The results showed that the proposed model always generates a weight-consistent Pareto-optimal solution in all cases, while the other existing models do not.

This article proposes an uncertain target-oriented QFD approach to service standardization-based service design with an application to bank window service, based on a probabilistic interpretation o...

Time series classification is one of the most important issues in time series data mining. This problem has attracted more and more attention of researchers in recent years. Among proposed methods in literature, 1-Nearest Neighbor (1-NN), its variants and improvements have been widely considered as hard to be beaten on classification of time series. In this paper, we propose a novel non-parametric method to classify time series. The proposed method, namely Weighted Local Dynamic Time Warping Barycenter Averaging k -Nearest Neighbors (WLDBAk-NN), is an improvement of Local Mean-based k -Nearest Neighbors (LMk-NN) algorithm. It improves LM k -NN in that it replaces the local mean vectors by local Dynamic Time Warping Barycenter (DBA) vectors calculated using our method, namely Weighted DBA (WDBA). By experiments, we show that (i) WLDBAk-NN outperforms the Weighted Local Mean-based k -Nearest Neighbors (WLMk-NN) algorithm, and (ii) both WLMk-NN and WLDBAk-NN outperform 1-NN, LMk-NN, k -Nearest Centroid Neighbors (k-NCN), and LMk-NCN in 85 time series datasets of UCR Time Series Classification Archive. The experimental results also show that new local mean vectors used in WLMk-NN and WLDBAk-NN significantly contribute to the improvement of the performance of time series classification.