Search

Today, with the expansion of online social networks and their impact on various aspects of human life, investigating the interactions between users and identifying influential users for various advertising applications and accelerating or preventing the dissemination of information has been the focus of researchers. One of the fundamental researches is investigating the fact that the similarity of users' characteristics along with their interests leads to new relationships in the friendship network, a concept known as homophily. The study of homophily can provide significant insight into the flow of information and behaviors in a community to analyze the formation of online communities. In recent years, the emergence of location-based social networks (LBSNs) has created massive datasets by sharing spatial and temporal information better than ever before. This issue enables researchers to analyze the behavioral patterns of users and their impact on their social connections and friends. Throughout the present paper, a framework is being defined to examine the effect of combining structural similarity and homophily in determining users' social influence under two scenarios. The experiments simulate performance of nodes on three LBSNs: Gowalla, Foursquare, and Brightkite. By calculating the correlation coefficient for the similarity methods applied, it can be displayed that with the increase in homophily, the correlation of the proposed method and the social influence increases. A new measure of centrality is also introduced by using the topological structure of the user's communication network, such as the eigenvector centrality along with the values of friendship influence and the number of spatial movements of the user. The results show that our proposed centrality matches up to 85% with baseline methods. [ABSTRACT FROM AUTHOR]

Recently, the analysis of heterogeneous networks has become more popular due to the growing number of social networks. These networks are capable of covering a variety of nodes and edges. The members of these networks usually have metadata whose analysis can lead to the discovery of knowledge. One way to analyze such data is clustered where high-quality clustering requires effective similarity calculation. Most of the existing clustering methods do not pay attention to the use of metadata or the characteristics of network members. On the other hand, they are only able to process small and medium-sized networks due to the amount of memory and execution speed. This paper presents a hybrid approach for heterogeneous network clustering to overcome these problems. The structural similarity in this approach is calculated by the graph embedding method, which we call learning-based. Attribute similarity is calculated by a scoring method that we call similarity-based. In the experimental study, we compared the proposed method with collaborative approaches based on similarity on the real-world networks. The experimental findings demonstrate the superiority of the proposed method in terms of entropy, memory consumption, execution time, and density in certain cases. [ABSTRACT FROM AUTHOR]

A recommender system is a model that automatically recommends some meaningful cases (such as clips/films/goods/items) to the clients/people/consumers/users according to their (previous) interests. These systems are expected to recommend the items according to the users' interests. There are two traditional general recommender system models, i.e., Collaborative Filtering Recommender System (ColFRS) and Content-based Filtering Recommender System (ConFRS). Also, there is another model that is a hybrid of those two traditional recommender systems; it is called Hybrid Recommender System (HRS). An HRS usually outperforms simple traditional recommender systems. The problems such as scalability, cold start, and sparsity belong to the main problems that any recommender system should solve. The memory-based (modeless) recommender systems benefit from good accuracies. But they suffer from a lack of admissible scalability. The model-based recommender systems suffer from a lack of admissible accuracies. But they benefit from good scalability. In this paper, it is tried to propose a hybrid model based on an automatically improved ontology to solve the scalability, cold start, and sparsity problems. Our proposed HRS also uses an innovative approach of clustering as an augmented section. When there are enough ratings, it uses a collaborative filtering approach to predict the missing ratings. When there are not enough ratings, it uses a content-based filtering approach to predict the missing ratings. In the content-based filtering section of our proposed HRS, ontology concepts are used to improve the accuracy of ratings' prediction. If our target client is severely sparse, we cannot trust even the ratings predicted by the content-based filtering section of our proposed HRS. Therefore, our proposed HRS uses additive clustering to improve the prediction of the missing ratings if the target client is severely sparse. It is experimentally shown that our model outperforms many of the newly developed recommender systems. [ABSTRACT FROM AUTHOR]

In recent years, various studies have been conducted on SVMs and their applications in different area. They have been developed significantly in many areas. SVM is one of the most robust classification and regression algorithms that plays a significant role in pattern recognition. However, SVM has not been developed significantly in some areas like large-scale datasets, unbalanced datasets, and multiclass classification. Efficient SVM training in large-scale datasets is of great importance in the big data era. However, as the number of samples increases, the time and memory required to train SVM increase, making SVM impractical even for a medium-sized problem. With the emergence of big data, this problem becomes more significant. This paper presents a novel distributed method for SVM training in which a very small subset of training samples is used for classification, which reduces the problem size and thus the required memory and computational resources. The solution of this problem almost converges to standard SVM. This method includes three steps: first, detecting a subset of distributed training samples, second, creating local models of SVM and obtaining partial vectors, and finally combining the partial vectors and obtaining the global vector and the final model. In addition, the datasets which suffer from unbalanced number of samples and tend to the majority class, the proposed method balances the samples of the two classes and it can be used in unbalanced datasets. The empirical results show that using this method is efficient for large-scale problems. [ABSTRACT FROM AUTHOR]

Recommender systems are widely used in many applications. They can be viewed as the predictor systems that are to suggest accurate and highly preferred items to consumers or clients. These systems can be considered to be information filtering systems. They counter some important challenges such as cold start (it means the absence of enough data for a new item to make accurate recommendations), scalability, and sparsity. The memory-based recommender systems have high accuracy but lack scalability. Also, the model-based systems are scalable but not accurate. Current recommender systems use hybrid methods to deal with the most important shortages of traditional filtering approaches. Current recommender systems are usually a hybrid of content-based filtering and collaborative filtering, and so on. In this paper, a hybrid recommender system is presented to meet the stated challenges, increase system performance and provide more accurate recommendations. This system uses both content-based filtering and collaborative filtering. In addition, using an automatically collected wordnet, we create an ontology that has been used in the content-based filtering section of our proposed approach. Furthermore, this framework applies KNN (k nearest neighbors) algorithm and clustering to improve its functionality. The proposed system is evaluated on a real benchmark. The experimentations show the proposed method has a better performance compared with the current superior related methods. The experimentations also show that our recommender system has desirable scalability compared with the state-of-the-art recommender systems. [ABSTRACT FROM AUTHOR]

Despite all the developments in recommender systems and utilizations of successful application models in the industry, it can be said that there is still a need to improve various parts of these systems in order to enhance their effectiveness and scope of application. Recommender Systems (RS) are well-known in the field of e-commerce and are expected to provide customers with important goods and items (including music and movies). In traditional recommender systems suffer from important challenges and problems such as cold start, scalability, and data dispersion. Recently, some of these challenges have been successfully overcomed to an acceptable extent by the advantages of the combined use of these methods. The K-nearest neighbors (KNN)-based Recommender Systems (KRS) are among the most powerful recommender engines that are currently available. In these systems, the rating of a target item is predicted based on the average rating of similar items, where the similarity is defined based on a similarity measure and the average rating of items is treated as a feature. In this paper, KRS is developed by combining the following approaches: (a) using the mean and variance of the item ratings as item features to identify the similar items (i.e., item-based KRS or IKRS); (b) using the mean and variance of the user ratings as the user features to identify the similar users (i.e., user-based KRS or UKRS); (c) using weighted average to combine the neighboring user/item ratings; and (d) using ensemble learning. In this study, some methods, i.e., EVMRS (Ensemble Variance-Mean based Recommender System) and EWVMRS (Ensemble Weighted Variance-Mean based Recommender System) are discussed and an improved EWVMRSG (Ensemble Weighted Variance-Mean Based Recommender system enriched by Gaussian mixture model (GMM)) is expanded, which are all user-based and involve using Mean Distance as the measure of similarity between the users/items to find neighboring users/items, but then they use unweighted average, weighted average, and weighted averaging based on the full-covariance GMM, respectively, for prediction. Experimental evaluations show that the EVMRS, EWVMRS, and EWVMRSG methods, which all use ensemble learning, are the most accurate methods among those developed and evaluated in this study. Depending on the dataset, the EWVMRSG mothed achieves a 20–30% lower absolute error than that of the original Mean based Recommender Systems (MRS). In terms of execution time, the proposed method is comparable to the original MRS models and is much faster than the Slope-one, P-kNN, and C-kNN RSs. [ABSTRACT FROM AUTHOR]

In recent years, researchers from academic and industrial fields have become increasingly interested in social network data to extract meaningful information. This information is used in applications such as link prediction between people groups, community detection, protein module identification, etc. Therefore, the clustering technique has emerged as a solution to finding similarities between social network members. Recently, in most graph clustering solutions, the structural similarity of nodes is combined with their attribute similarity. The results of these solutions indicate that the graph's topological structure is more important. Since most social networks are sparse, these solutions often suffer from insufficient use of node features. This paper proposes a hybrid clustering approach as an application for link prediction in heterogeneous information networks (HINs). In our approach, an adjacency vector is determined for each node until, in this vector, the weight of the direct edge or the weight of the shortest communication path among every pair of nodes is considered. A similarity metric is presented that calculates similarity using the direct edge weight between two nodes and the correlation between their adjacency vectors. Finally, we evaluated the effectiveness of our proposed method using DBLP, Political blogs, and Citeseer datasets under entropy, density, purity, and execution time metrics. The simulation results demonstrate that while maintaining the cluster density significantly reduces the entropy and the execution time compared with the other methods. [ABSTRACT FROM AUTHOR]

Achieving an efficient and reliable method is essential to interpret a user's brain wave and deliver an accurate response in biomedical signal processing. However, EEG patterns exhibit high variability across time and uncertainty due to noise and it is a significant problem to be addressed in mental task as motor imagery. Therefore, fuzzy components may help to enable a higher tolerance to noisy conditions. With the advent of Deep Learning and its considerable contributions to Artificial intelligence and data analysis, numerous efforts have been made to evaluate and analyze brain signals. In this study, to make use of neural activity phenomena, the feature extraction preprocessing is applied based on Multi-scale filter bank CSP. In the following, the hybrid series architecture named EEG-CLFCNet is proposed which extract the frequency and spatial features by Compact-CNN and the temporal features by the LSTM network. However, the classification results are evaluated by merging the fully connected network and fuzzy neural block. Here, the proposed method is further validated by the BCI competition IV-2a dataset and compare with two hyperparameter tuning methods, Coordinate-descent and Bayesian optimization algorithm. The proposed architecture that used fuzzy neural block and Bayesian optimization as tuning approach, results in better classification accuracy compared with the state-of-the-art literatures. As results shown, the remarkable performance of the proposed model, EEG-CLFCNet, and the general integration of fuzzy units to other classifiers would pave the way for enhanced MI-based BCI systems. [ABSTRACT FROM AUTHOR]

Nowadays, social networks are inseparable part of people everyday life. Large amounts of check-ins collected by location-based social networks (LBSNs) are rich information sources of social and spatio-temporal aspects of human life. In LBSNs, a co-occurrence refers to an event that a pair (almost) simultaneously appears at the same location. There are very few efforts on co-occurrence prediction in LBSNs in the literature. This paper aims to predict the co-occurrences between pairs of users, whether in the future or the past. This can be used in epidemiological studies like today COVID-19 pandemic. We have identified several effective spatial/temporal/social factors. For each, the method of the predictability measurement has been proposed. The spatial, regional, categorical and hierarchical-temporal aspects have been measured with inverse diversity, which is based on Entropy. In general, lower entropy implies higher predictability. Coincidences are unintentional encounters between users. Unlike previous studies that eliminated them, we have been considered them as a co-occurrence type and their effects have been measured and considered in the prediction task. By utilizing these identified factors along with the friendship information, a novel framework has been developed to enable the co-occurrence pattern inference among user pairs. Furthermore, valid patterns have been identified to prevent predictions based on expired patterns. To the best of our knowledge, this issue has not been studied in any other human behavior investigations. Evaluations with two real LBSN datasets, show the efficacy of our co-occurrence inference framework, outperforming other comparable method. [ABSTRACT FROM AUTHOR]

As one of the significant issues in social networks analysis, the influence maximization problem aims to fetch a minimal set of the most influential individuals in the network to maximize the number of influenced nodes under a diffusion model. Several approaches have been proposed to tackle this NP-hard problem. The traditional approaches failed to develop an efficient and effective solution due to the exponential growth of the size of social networks (due to massive computational overhead). In this paper, a three-stage framework based on the community detection approach is devised, namely LGFIM. In the first stage, the search space was controlled by partitioning the network into communities. Simultaneously, three heuristic methods were presented for modifying the community detection algorithm to extract the optimal communities: core nodes selection, capacity constraint on communities, and communities combination. These extracted communities were highly compatible with the information propagation mechanism. The next stages apply a scalable and robust algorithm at two different levels of the network: 1. Exploring the local scope of communities to select the most influential nodes of each community and construct the potential influential nodes set 2. Exploring the global scope of the network to select the target influential nodes among potential influential nodes set. Experimental results on various real datasets proved that LGFIM could achieve remarkable results compared with the state-of-the-art algorithms, especially acceptable influence spread, much better running time, and more applicable to massive social networks. [ABSTRACT FROM AUTHOR]

Entity resolution (ER) is a process that identifies duplicate records referring to a real-world entity and links them together in one or more datasets. As a first step toward reducing the number of required record comparisons, blocking methods attempt to group records that are likely to match. A proper evaluation of blocking methods for selecting the best one has a direct effect on the ultimate ER performance. Currently, the available metrics for evaluating blocking techniques exclusively assess their actual potential. However, it is possible to deduce new pairs from the identified ones in dirty datasets due to transitive closure between matching record pairs. In the present study, a modification of current metrics is proposed to obtain a more accurate evaluation of blocking methods taking into account transitive closure and the potential of blocking methods. Comparing the existing and proposed metrics for ten available blocking algorithms on two dirty datasets demonstrates that the proposed metrics correlate significantly with ER final performance. [ABSTRACT FROM AUTHOR]

Headline generation is a challenging subtask of abstractive text summarization, which its output should be a summary, shorter than one sentence. It would be precious to develop a dataset for the evaluation of abstractive summarization methods on this task in the Persian language. There are several datasets for headline generation in Persian, most of which are not large enough to be used by more sophisticated methods of text summarization, such as deep learning models. Moreover, all of these datasets are focused on daily news and there is no dataset for summarizing scientific Persian papers. In this article, we present "ElmNet," a headline generation dataset of about 400,000 abstract/headline pairs of scientific papers, gathered from six major publishers for scientific articles in Persian. We, moreover, evaluate the performance of the most important deep learning-based headline generation methods, on the proposed dataset. The results prove the comparability of the performance of the state-of-the-art methods on this task, to their results on the existing English datasets. [ABSTRACT FROM AUTHOR]

Digital social media has played a key role in tourism and hospitality industry. The use of machine and deep learning has been effective in market segmentation and customers' preference prediction through social big data analysis. This paper develops a new method to analyze large set of open data in social networking sites for travellers segmentation and predict tourists' choice preferences using dimensionality reduction and deep learning techniques. Deep belief network was used for predicting the travellers' choice preferences from their past ratings and online reviews. Self-organizing map was also used for clustering the travellers' online ratings and reviews. The feature extraction is performed using latent Dirichlet allocation as an unsupervised learning technique. To improve the effectiveness of learning, a dimensionality reduction technique, higher-order singular value decomposition, is performed on the clusters for the prediction of missing values and traveller–traveller similarity calculation. The proposed method was evaluated on travellers' online reviews and ratings which were crawled from TripAdvisor. The results showed the robustness of the proposed method in analysing the large text-based reviews and numerical datasets in tourism context. [ABSTRACT FROM AUTHOR]

Mental health problems have always existed in human life. However, factors such as lifestyle changes and industrialization affect modern human mental health. A common mental illness is depressive disorder, and a large number of patients are not even aware they have it. Due to the harmful influences of depressive disorder on quality of life, timely and accurate diagnosis is a matter of extreme prominence. The aim of this study is to design a web-based expert system for diagnosing depression using the fuzzy Delphi method by estimating the weights and importance of depression symptoms. Fuzzy Logic is adopted to calculate the level of depression. Two thresholds are obtained, namely lack of depression and severe depression. Finally, the level of depression for each person is estimated by virtue of the calculated value’s proximity to these two opposing points. Java Expert System Shell is used to build the knowledge base. Sensitivity and specificity analysis are performed with 238 participants. The proposed system appears helpful for everyone, from ordinary persons to specialists in medical environments. It can also be useful to train psychology students in the area of diagnostic reasoning. [ABSTRACT FROM AUTHOR]

Each clustering algorithm usually optimizes a qualification metric during its learning process. The qualification metric in traditional clustering algorithms considers all the features of under-consideration dataset equally; it means each feature participates in the clustering process equivalently. Considering that some features have more information than the others in a dataset (due to their lower information or their higher variances, etc.), we proposes a fuzzy weighted clustering algorithm. We name this new clustering algorithm, Fuzzy Weighted Locally Adaptive Clustering (FWLAC) algorithm. The proposed FWLAC algorithm is capable of handling imbalanced clustering. However, FWLAC algorithm suffers from its sensitivity to the two parameters that should be tuned manually. The performance of FWLAC algorithm is affected by well-tuning of its parameters. So the paper proposes two solutions to well-tuning of its two parameters. In the first solution, we propose a simple clustering ensemble framework to show the sensitivity of the WLAC algorithm to its manual well-tuning. Although it is not a try-and-error procedure, it is like a grid search, where we use different pairs of values for both parameters h and h. Per each pair of values for parameters h and h, the algorithm produces a partitioning. So after the grid search, we obtain a large number of partitionings. We break any of the partitionings into its clusters, and then they form an ensemble of clusters. Finally the consensus partitioning is extracted from them by a consensus function. The algorithm is not data dependent at all. For all datasets, we use a similar grid search and a similar set of values for the parameters h and h. In this way we have proposed an alternative solution to parameter selection. We use a selection phase to select/remove some clusters from our ensemble of clusters to obtain an elite ensemble of clusters. To do this, a stability measure, normalized mutual information (NMI), was used to validate a cluster. The paper shows the effectiveness of the proposed clustering frameworks both theoretically and experimentally. [ABSTRACT FROM AUTHOR]

Clustering ensembles combine multiple partitions of data into a single clustering solution of better quality. Inspired by the success of supervised bagging and boosting algorithms, we propose non-adaptive and adaptive resampling schemes for the integration of multiple independent and dependent clusterings. We investigate the effectiveness of bagging techniques, comparing the efficacy of sampling with and without replacement, in conjunction with several consensus algorithms. In our adaptive approach, individual partitions in the ensemble are sequentially generated by clustering specially selected subsamples of the given dataset. The sampling probability for each data point dynamically depends on the consistency of its previous assignments in the ensemble. New subsamples are then drawn to increasingly focus on the problematic regions of the input feature space. A measure of data point clustering consistency is therefore defined to guide this adaptation. Experimental results show improved stability and accuracy for clustering structures obtained via bootstrapping, subsampling, and adaptive techniques. A meaningful consensus partition for an entire set of data points emerges from multiple clusterings of bootstraps and subsamples. Subsamples of small size can reduce computational cost and measurement complexity for many unsupervised data mining tasks with distributed sources of data. This empirical study also compares the performance of adaptive and non-adaptive clustering ensembles using different consensus functions on a number of datasets. By focusing attention on the data points with the least consistent clustering assignments, whether one can better approximate the inter-cluster boundaries or can at least create diversity in boundaries and this results in improving clustering accuracy and convergence speed as a function of the number of partitions in the ensemble. The comparison of adaptive and non-adaptive approaches is a new avenue for research, and this study helps to pave the way for the useful application of distributed data mining methods. [ABSTRACT FROM AUTHOR]