Journal of Intelligent & Fuzzy Systems - Volume 32, issue 2

Authors: Vathi, Eleni | Siolas, Georgios | Stafylopatis, Andreas

Article Type: Research Article

Abstract: In recent years, the popularity of social networks has grown dramatically. Understanding and revealing the underlying community structure of these complex networks is an area of great interest, with a plethora of applications. In this paper, we present a methodology for identifying user communities on Twitter. Initially, Twitter features such as the shared content, the users’ interactions and the following relationships between the users are utilized to define a number of similarity metrics. These metrics are then used to compute the similarity between each pair in a set of Twitter users and by extension to group these users into communities. …Subsequently, we propose a novel method based on latent Dirichlet allocation to extract the topics discussed in each community and eliminate those which consist of everyday words. Additionaly, we introduce a method for automatically generating labels for the non-trivial topics. The methodology is evaluated with a real-world dataset created using the Twitter Searching API. Show more

Keywords: Community detection, topic modeling, Twitter

DOI: 10.3233/JIFS-169125

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1265-1275, 2017

Authors: Tran, Van Cuong | Hoang, Dinh Tuyen | Nguyen, Ngoc Thanh | Hwang, Dosam

Article Type: Research Article

Abstract: In recent years, information extraction from tweets has been challenging for researchers in the fields of knowledge discovery and data mining. Unlike formal text, such as news articles and pieces of longer content, tweets are of a specific nature: short, noisy, and with dynamic content. Thus, it is difficult to apply the traditional natural language processing algorithms to analyze them. Active learning is well-suited to many problems in natural language processing, especially when unlabeled data may be abundant, but labeled data is limited. The method proposed here aims to minimize annotation costs while maximizing the desired performance from the model. …The method recognizes named entities from tweet streams on Twitter by using an active learning method with different query strategies. The tweets are queried for labeling by a human annotator based on query-by-committee, uncertainty-based sampling, and diversity-based sampling. The experimental evaluations of the proposed method on tweet data achieved better results than random sampling. Show more

Keywords: Named entity recognition, active learning, tweet streams, query strategy

DOI: 10.3233/JIFS-169126

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1277-1287, 2017

Authors: Jędrzejowicz, Joanna | Jędrzejowicz, Piotr

Article Type: Research Article

Abstract: The paper proposes two variants of the ensemble distance-based and Naive-Bayes online classifiers with data reduction. In the first variant the reduced dataset is obtained through applying bias-correction fuzzy clustering. In the second we used the kernel-based fuzzy clustering as the data reduction tool. It is assumed that vectors of data with unknown class label arrive one by one, and that there is available an initial chunk of data with known class labels serving as the initial training set. Classification is carried-out in rounds. Each round involves a number of the classification decisions equal to the chunk size. For each …round a set of base classifiers is constructed using different distance metrics. Set of base classifiers is extended with the Naive-Bayes classifier. The unknown label of each incoming vector is determined through weighted majority voting. After each round has been completed the training set is replaced by the fresh one and the classification process is continued. The approach is validated through computational experiment involving a number of datasets often used for testing data streams mining algorithms. Show more

Keywords: Online classification, kernel-based fuzzy clustering, bias-correction fuzzy clustering

DOI: 10.3233/JIFS-169127

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1289-1296, 2017

Authors: Nguyen, Loan T.T. | Trinh, Truc | Nguyen, Ngoc-Thanh | Vo, Bay

Article Type: Research Article

Abstract: Mining frequent closed itemsets (FCIs) is important in mining non-redundant (minimal) association rules. Therefore, many algorithms have been developed for mining FCIs with reduced mining time and memory usage. For mining FCIs, algorithms use the minimum support threshold, minSup , to prune itemsets. However, using a fixed minSup is not suitable for mining top-rank-k FCIs. A large threshold will lead to a small number of generated FCIs, leading to insufficient FCIs to query when k is large. On the other hand, a small minSup will generate a huge number of generated FCIs, leading to large runtimes and high memory usage. In …this paper, we propose a method for mining top-rank-k FCIs without using a fixed minimum support threshold. A strategy is first used to eliminate 1-items that cannot generate FCIs belonging to top-rank-k FCIs. Next, based on the set of candidate 1-items, we propose TRK-FCI, a DCI-Plus-based algorithm, for mining top-rank-k FCIs. In the process of mining top-rank-k FCIs, TRK-FCI automatically increases minSup according to the mined FCIs, efficiently pruning itemsets that cannot belong to top-rank-k FCIs. We also modify the dynamic bit vector (DBV) structure and apply it to reduce memory usage and runtime in the TRK-FCI-DBV algorithm. Experimental results show that TRK-FCI-DBV is more efficient than TRK-FCI for various databases. Show more

Keywords: DCI-Plus, dynamic bit vectors, frequent closed itemsets, top-rank-k frequent closed itemsets

DOI: 10.3233/JIFS-169128

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1297-1305, 2017

Authors: Rodríguez-Fernández, Víctor | Menéndez, Héctor D. | Camacho, David

Article Type: Research Article

Abstract: Unmanned Aerial Vehicles (UAVs) are starting to provide new possibilities to human societies and their demand is growing according to the new industrial application fields for these revolutionary tools. The current systems are still evolving, specially from an Artificial Intelligence perspective, which is increasing the different tasks that UAVs can perform. However, the current state still requires a strong human supervision. As a consequence, a good preparation for UAV operators is mandatory due to some of their applications might affect human safety. During the training process, it is important to measure the performance of these operators according to different factors …that can help to decide what operators are more suitable for different kinds of missions creating operator profiles. Having this goal in mind, this work aims to present an extensive and robust methodology to automatically extract different performance profiles from the training process of operators in an UAV simulation environment. Our method combines the definition of a set of performance metrics with clustering techniques to define operators profiles, ensuring that the behavior discrimination is suitable and consistent. Show more

Keywords: UAVs, Human-Robot Interaction, computer-based simulation, clustering, performance metrics, behavioral analysis

DOI: 10.3233/JIFS-169129

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1307-1319, 2017

Authors: Leon, Florin | Curteanu, Silvia

Article Type: Research Article

Abstract: The concept of a large margin is central to support vector machines and it has recently been adapted and applied for nearest neighbour classification. In this paper, a modification of this method is proposed in order to be used for regression problems. This model also allows the use of a set of prototypes with different distance metrics, which can increase the flexibility of the method especially for problems with large number of instances. The learning of the distance metrics is performed by two optimization methods, namely an evolutionary algorithm and an approximate differential approach. A real world problem, i.e. the …prediction of the corrosion resistance of some alloys containing titanium and molybdenum is considered as a case study. It is shown that the suggested method provides very good results compared to other well-known regression algorithms. Show more

Keywords: Large margin, nearest neighbour regression, prototypes, distance metrics, evolutionary algorithm, approximate differential optimization

DOI: 10.3233/JIFS-169130

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1321-1332, 2017

Authors: Plata, Diego Rueda | Ramos-Pollán, Raúl | González, Fabio A.

Article Type: Research Article

Abstract: This work proposes a supervised layer-wise strategy to train deep convolutional neural networks (DCNs) particularly suited for small, specialized image datasets. DCNs are increasingly being used with considerable success in image classification tasks and trained over large datasets (with more than 1M images and 10 K classes). Pre-trained successful DCNs can then be used for new smaller datasets (10 K to 100 K images) through a transfer learning process which cannot guarantee competitive a-priori performance if the new data is of different or specialized nature (medical imaging, plant recognition, etc.). We therefore seek out to find competitive techniques to train DCNs for such …small datasets, and hereby describe a supervised greedy layer-wise method analogous to that used in unsupervised deep networks. Our method consistently outperforms the traditional methods that train a full DCN architecture in a single stage, yielding an average of over 20% increase in classification performance across all DCN architectures and datasets used in this work. Furthermore, we obtain more interpretable and cleaner visual features. Our method is better suited for small, specialized datasets since we require a training cycle for each DCN layer and this increases its computing time almost linearly with the number of layers. Nevertheless, it still remains as a fraction of the computing time required to generate pre-trained models with large generic datasets, and poses no additional requirements on hardware. This constitutes a solid alternative for training DCNs when transfer learning is not possible and, furthermore, suggests that state of the art DCN performance with large datasets might yet be improved at the expense of a higher computing time. Show more

Keywords: Convolutional networks, deep learning, greedy layer-wise training

DOI: 10.3233/JIFS-169131

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1333-1342, 2017

Authors: Cañizares, Pablo C. | Merayo, Mercedes G. | Vara, Juan M.

Article Type: Research Article

Abstract: This paper introduces a new high-level domain-specific modelling language, LAnt , for the design of ant colony optimization algorithms. This language is used to represent the main elements required to define the structure of the algorithms and to capture the specific constraints associated to the problem to solve. It aims to support users with low experience in the development of solutions based on this paradigm. The proposal has been implemented as an Eclipse plug-in, including an editor and an integrated code generator.

Keywords: Artificial intelligence, ant colony optimization, model driven engineering, domain specific language

DOI: 10.3233/JIFS-169132

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1343-1354, 2017

Authors: Salehi, Saber | Selamat, Ali | Kuca, Kamil | Krejcar, Ondrej | Sabbah, Thabit

Article Type: Research Article

Abstract: Spam email problem is a major shortcoming of email technology for computer security. In this research, a granular classifier model is proposed to discover hyper-boxes in the geometry of information granules for spam detection in three steps. In the first step, the k-means clustering algorithm is applied to find the seed_points to build the granular structure of the spam and non-spam patterns. Moreover, the key part of the spam and non-spam classifiers’ structure is captured by applying the interval analysis through the high homogeneity of the patterns. In the second step, PSO algorithm is hybridized with the k-means to optimize …the formalized information granules’ performance. The size of the hyperboxes is expanded away from the center of the granules by PSO. There are some patterns that do not placed in any of the created clusters and known as noise points. In the third step, the membership function in fuzzy sets is applied to solve the noise points’ problem by allocating the noise points through the membership grades. The proposed model is evaluated based on the accuracy, misclassification and coverage criteria. Experimental results reveal that the performance of our proposed model is increased through applying Particle Swarm Optimization and fuzzy set. Show more

Keywords: Spam detection, hyperbox geometry of classifiers, granular classifier, membership functions, particle swarm optimisation, k-means clustering algorithm

DOI: 10.3233/JIFS-169133

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1355-1363, 2017

Authors: Huk, M.

Article Type: Research Article

Abstract: In this paper, we show that a contextual neural network with artificial neurons performing a conditional aggregation of signals can be trained by the generalized backpropagation algorithm. To allow this algorithm to be used for training contextual neural networks, we derive appropriate generalized delta rules. Our approach is constructed on the basis of introduced generalized representation of the aggregation function in an ordered groups space and division of its attention function into binary scan-path and contribution functions. The advantage of the proposed representation is that it clarifies the description of the aggregation process by using Stark’s scan-path theory and allows …us to achieve results independent from the actual form of the attention functions used during aggregation. As such, the proposed solution is valid for the whole presented family of conditional aggregation functions and is a considerable extension of the previously reported results. In particular, the obtained results are valid for the introduced exemplary attention functions which illustrate performed calculations. Moreover, the presented solution can be further extended by considering real valued, non-binary contribution functions inside ordered aggregation functions. Especially promising are its possible applications in large deep neural networks and energy-limited systems. Show more

Keywords: Contextual neural networks, conditional aggregation, classification, dynamic inputs selection, selective attention

DOI: 10.3233/JIFS-169134

Citation: Journal of Intelligent & Fuzzy Systems, vol. 32, no. 2, pp. 1365-1376, 2017