Search

A high percentage of information that propagates through a social network is sourced from different exogenous sources. E.g., individuals may form their opinions about products based on their own experience or reading a product review, and then share that with their social network. This sharing then diffuses through the network, evolving as a combination of both network and external effects. Besides, different individuals (nodes in a social network) have different degrees of exposition to their external sources, as well. Modeling this influence of external sources is important in order to understand the diffusion process and predict future content sharing patterns. Recognizing this fusion of intrinsic (network) effect and exogenous (external) effect, this paper develops a novel fuzzy relative willingness (FRW) model. Leveraging a fuzzy set approach provides a way to handle the uncertainties arising within the human concept of willingness. We demonstrate that FRW is able to accurately identify both top- $k$ most content producers and diffusion effect based on external influence. We also demonstrate that the fuzzy set theory provides a compelling framework to model uncertainties pertaining to the influence as well as the susceptibility of individuals for both network and exogenous effects.

We propose a framework for model calibration based on niching genetic algorithms.It consists of interactive multimodal search, visualization, and parameters analysis.The output of the framework is a set of alternative model calibration solutions.The multimodal framework is applied to two different applications.Experiments showed the framework benefits when interactively validating models. Automated calibration is a crucial stage when validating non-linear dynamic systems. The modeler must control the calibration results and analyze parameter values in an iterative way. In many non-linear models, it is usual to find sets of configuration parameters that may obtain the same model fitting. In these cases, the modeler needs to understand the results' implications and run a sensitivity analysis to check the model validity. This paper presents a framework based on niching genetic algorithms to provide modeler with a set of alternative calibration solutions which also ease the analysis of their parameters, model's response, and sensitivity analysis. The framework is called MOMCA, an integral and interactive solution for model validation which facilitates the implication of decision makers. The core component of MOMCA is its niching genetic algorithm, able to reach various optima in multimodal optimization problems by keeping the necessary diversity. The proposed framework is applied to two different case studies. The first case study is a biological growth model and the second one is a managerial model to improve brand equity. Both applications show the benefits of the framework when providing a set of calibrated models and a way to analyze and perform sensitivity analysis based on the set of solutions.

An incremental learning method for nodes’ classification is presented in the paper. In particular, there is proposed an active scheme algorithm for multi-class classification of nodes’ states that varies over time and depends on information spread in the network. Demonstration of the method is conducted using social network dataset. According to sent messages between nodes, the emotional state of the message sender updates each receiving node’s feature vector and the method tries to classify next emotional state of the receiver. The novelty of the proposed approach lies in applying incremental learning method for non-stationary network environment. There are demonstrated some properties of the proposed method in experiments with real data set, showing that the method can effectively classify the future state of nodes.

This article investigates the possibility of accurate quality prediction of resources generated by communities based on the crowd-generated content. We use data from Wikipedia, the prime example of community-run site, as our object of study. We define the quality as a distribution of user-assigned grades across a predefined range of possible scores and present a measure of distribution similarity to quantify the accuracy of a prediction. The proposed method of quality prediction is based on Markov Random Field and its Loopy Belief Propagation implementation. Based on our results, we highlight key problems in the approach as presented, as well as trade-offs caused by relying solely on network structure and characteristics, excluding metadata. The overall results of content quality prediction are promising in homophilic networks.

Many collections of numbers do not have a uniform distribution of the leading digit, but conform to a very particular pattern known as Benford’s distribution. This distribution has been found in numerous areas such as accounting data, voting registers, census data, and even in natural phenomena. Recently it has been reported that Benford’s law applies to online social networks. Here we introduce a set of rigorous tests for adherence to Benford’s law and apply it to verification of this claim, extending the scope of the experiment to various complex networks and to artificial networks created by several popular generative models. Our findings are that neither for real nor for artificial networks there is sufficient evidence for common conformity of network structural properties with Benford’s distribution. We find very weak evidence suggesting that three measures, degree centrality, betweenness centrality and local clustering coefficient, could adhere to Benford’s law for scalefree networks but only for very narrow range of their parameters.

A framework for multi-label classification extended by Error Correcting Output Codes (ECOCs) is introduced and empirically examined in the article. The solution assumes the base multi-label classifiers to be a noisy channel and applies ECOCs in order to recover the classification errors made by individual classifiers. The framework was examined through exhaustive studies over combinations of three distinct classification algorithms and four ECOC methods employed in the multi-label classification problem. The experimental results revealed that (i) the Bode-Chaudhuri-Hocquenghem (BCH) code matched with any multi-label classifier results in better classification quality; (ii) the accuracy of the binary relevance classification method strongly depends on the coding scheme; (iii) the label power-set and the RAkEL classifier consume the same time for computation irrespective of the coding utilized; (iv) in general, they are not suitable for ECOCs because they are not capable to benefit from ECOC correcting abilities; (v) the all-pairs code combined with binary relevance is not suitable for datasets with larger label sets.

Choosing a proper training set for machine learning tasks is of great importance in complex domain problems. In the paper a new distance measure for training set selection is presented and thoroughly discussed. The distance between two datasets is computed using variance of entropy in groups obtained after clustering. The approach is validated using real domain datasets from debt portfolio valuation process. Eventually, prediction performance is examined.

Relations between objects in various systems, such as hyperlinks connecting web pages, citations of scientific papers, conversations via email or social interactions in Web 2.0 portals are commonly modeled by networks. One of many interesting problems currently studied for such domains is node classification. Due to the nature of the networked data and the unavailability of collection of nodes' broad representation for training in majority of environments, only a very limited data may remain useful for classification. Therefore, there is a need for accurate and efficient algorithms that are able to perform good classification based only on scanty knowledge of network nodes. A new approach of sampling algorithm-LDGibbs, used in the context of collective classification with application of label-dependent features, is proposed in the paper in order to provide more accurate generalization for sparse datasets. Additionally, a new LDBootstrapping algorithm based on label-dependent features has been developed. Both new algorithms include additional steps to extract new input features based on graph structures but limited only to the nodes of a given label. It means that a separate set of structural features is provided for each label. The comparison with the other approaches, in particular with standard Gibbs Sampling and bootstrapping provided satisfactory results and revealed LDGibbs's superiority.

Data mining of electronic health records (eHRs) allows us to identify patterns of patient data that characterize diseases and their progress and learn best practices for treatment and diagnosis. Clinical Prediction Rules (CPRs) are a form of clinical evidence that quantifies the contribution of different clinical data to a particular clinical outcome and help clinicians to decide the diagnosis, prognosis or therapeutic conduct for any given patient. The TRANSFoRm diagnostic support system (DSS) is based on the construction of an ontological repository of CPRs for diagnosis prediction in which clinical evidence is expressed using a unified vocabulary. This paper explains the proposed methodology for constructing this CPR repository, addressing algorithms and quality measures for filtering relevant rules. Some preliminary application results are also presented.

Background Analysis of encounter data relevant to the diagnostic process sourced from routine electronic medical record (EMR) databases represents a classic example of the concept of a learning healthcare system (LHS). By collecting International Classification of Primary Care (ICPC) coded EMR data as part of the Transition Project from Dutch and Maltese databases (using the EMR TransHIS), data mining algorithms can empirically quantify the relationships of all presenting reasons for encounter (RfEs) and recorded diagnostic outcomes. We have specifically looked at new episodes of care (EoC) for two urinary system infections: simple urinary tract infection (UTI, ICPC code: U71) and pyelonephritis (ICPC code: U70). Methods Participating family doctors (FDs) recorded details of all their patient contacts in an EoC structure using the ICPC, including RfEs presented by the patient, and the FDs’ diagnostic labels. The relationships between RfEs and episode titles were studied using probabilistic and data mining methods as part of the TRANSFoRm project. Results The Dutch data indicated that the presence of RfE’s “Cystitis/Urinary Tract Infection”, “Dysuria”, “Fear of UTI”, “Urinary frequency/urgency”, “Haematuria”, “Urine symptom/complaint, other” are all strong, reliable, predictors for the diagnosis “Cystitis/Urinary Tract Infection” . The Maltese data indicated that the presence of RfE’s “Dysuria”, “Urinary frequency/urgency”, “Haematuria” are all strong, reliable, predictors for the diagnosis “Cystitis/Urinary Tract Infection”. The Dutch data indicated that the presence of RfE’s “Flank/axilla symptom/complaint”, “Dysuria”, “Fever”, “Cystitis/Urinary Tract Infection”, “Abdominal pain/cramps general” are all strong, reliable, predictors for the diagnosis “Pyelonephritis” . The Maltese data set did not present any clinically and statistically significant predictors for pyelonephritis. Conclusions We describe clinically and statistically significant diagnostic associations observed between UTIs and pyelonephritis presenting as a new problem in family practice, and all associated RfEs, and demonstrate that the significant diagnostic cues obtained are consistent with the literature. We conclude that it is possible to generate clinically meaningful diagnostic evidence from electronic sources of patient data. Electronic supplementary material The online version of this article (doi:10.1186/s12875-015-0271-4) contains supplementary material, which is available to authorized users.

The problem of finding optimal set of users for influencing others in the social network has been widely studied. Because it is NP-hard, some heuristics were proposed to find sub-optimal solutions. Still, one of the commonly used assumption is the one that seeds are chosen on the static network, not the dynamic one. This static approach is in fact far from the real-world networks, where new nodes may appear and old ones dynamically disappear in course of time. The main purpose of this paper is to analyse how the results of one of the typical models for spread of influence - linear threshold - differ depending on the strategy of building the social network used later for choosing seeds. To show the impact of network creation strategy on the final number of influenced nodes - outcome of spread of influence, the results for three approaches were studied: one static and two temporal with different granularities, i.e. various number of time windows. Social networks for each time window encapsulated dynamic changes in the network structure. Calculation of various node structural measures like degree or betweenness respected these changes by means of forgetting mechanism - more recent data had greater influence on node measure values. These measures were, in turn, used for node ranking and their selection for seeding. All concepts were applied to experimental verification on five real datasets. The results revealed that temporal approach is always better than static and the higher granularity in the temporal social network while seeding, the more finally influenced nodes. Additionally, outdegree measure with exponential forgetting typically outperformed other time-dependent structural measures, if used for seed candidate ranking.

Relational classification is a promising branch of machine learning techniques for classification in networked environments which does not fulfil the iid assumption (independent and identically distributed). During the past few years, researchers have proposed many relational classification methods. However, almost none of them was able to work efficiently with large amounts of data or sparsely labelled networks. It is introduced in this paper a new approach to relational classification based on competence region modelling. The approach aims at solving large relational data classification problems, as well as seems to be a reasonable solution for classification of sparsely labelled networks by decomposing the initial problem to subproblems (competence regions) and solve them independently. According to preliminary results obtained from experiments performed on real world datasets competence region modelling approach to relational classification results with more accurate classification than standard approach.

The web spam detection problem was considered in the paper. Based on interconnected spam and no-spam hosts a collective classification approach based on label propagation is aimed at discovering the spam hosts. Each host is represented as network node and links between hosts constitute network’s edges. The proposed method provides reasonable results and is able to compute large data as is settled in MapReduce programming model.

In the paper three learning and inference ordering approaches in the method for structured output classification are presented. As it was previously presented by authors, classification of single element in output structure can be performed by generalization of input attributes as well as already partially classified output elements [9]. The paper addresses crucial problem of how to order elements in the structured learning process to get greater final accuracy. The learning is performed by means of ensemble, boosting classification method adapted to structured prediction - AdaBoostSeq algorithm. Authors present several ordering heuristics for score function application in order to obtain better structured output classification accuracy.

The collective classification problem for big data sets using MapReduce programming model was considered in the paper. We introduced a proposal for implementation of label propagation algorithm in the network. The method was examined on real dataset in telecommunication domain. The results indicated that it can be used to classify nodes in order to propose new offerings or tariffs to customers.

Distance measures, especially between probability density functions, are essential in solving machine learning problems. Among classification and clustering, data reduction and selection are some of them. In the paper a new distance measure for comparing and selecting training datasets is described. The distance between two datasets is based on variance of entropy in groups obtained by clustering joint datasets. The proposed approach is examined in dataset selection during prediction of debt portfolio value. Finally, basic evaluation on prediction performance is conducted.

It is proposed in the paper a new method for structured output prediction using ensemble of classifiers composed on the basis of Error-Correcting Output Codes. It was presented that newly presented Multiple Classifier Method for Structured Output Prediction based on Error Correcting Output Codes requires comparable computation time in comparison to other accurate algorithms and simultaneously in the classification of more complex structures it provides much better results in the same computation time.

Node classification in Social Network is currently receiving raising attention in the Social Network Analysis research. The main objective of node classification is to assign the correct label to the unlabeled nodes from a set of all possible class labels. This classification task is performed using features extracted from a Social Network dataset. The success of proper feature extraction significantly influences classification accuracy, providing more discriminative description of the data. This paper describes label-dependent features extraction and examines the classification accuracy based on features extracted with this approach. The experiments on real-world data have shown that usage of label-dependent features can lead to significant improvement of classification accuracy.

Ensemble methods of incremental prediction for sequences refer to learning from new reference data that become available after the model has already been created from a previously available data set. The main obstacle in the prediction of sequential values in the real environment with huge amount of data is the integration of knowledge stored in the previously obtained models and the new knowledge derived from the incrementally acquired new increases of the data. In the paper, the new approach of the ensemble incremental learning for prediction of sequences was proposed as well as examined using real debt recovery data.

The concept of new approach for debt portfolio pattern recognition is presented in the paper. Aggregated prediction of sequential repayment values over time for a set of claims is performed by means of hybrid combination of various machine learning techniques, including clustering of references, model selection and enrichment of input variables with prediction outputs from preceding periods. Experimental studies on real data revealed usefulness of the proposed approach for claim appraisals. The average accuracy was over 93%, much higher than for simplifier methods.

The concept of the new hybrid method for debt portfolio repayment prediction has been presented and examined. The method provides functionality for repayment value prediction over time that describes the recovery profile of the debt portfolio. Experimental studies on hybrid combination of various data mining methods like clustering and decision trees into one complex process revealed usefulness of the proposed method for claim appraisals.

The relational label propagation problem for large data sets using MapReduce programming model was considered in the paper. The method we propose estimates class probability in relational domain in the networks. The method was examined on large real telecommunication data set. The results indicated that it could be used successfully to classify networks' nodes and, thanks to that, new offerings or tariffs might be proposed to customers who belong to other providers. Moreover, basic properties of relational label propagation were examined and reported.