Danuta Zakrzewska

Using Fuzzy Logic for Recommending Groups in E-Learning Systems

Performance of Web-based learning environment depends on the degree it is adjusted into needs of virtual learning community members. Creating groups of students with similar needs enables to differentiate appropriately the environment features. Each new student, who joins the community, should obtain the recommendation of the group of colleagues with similar characteristics. In the paper, it is considered using fuzzy logic for modeling student clusters. As the representation of each group, we assume fuzzy numbers connected with learner attributes ranked according to their cardinality. Recommendations for new students are determined taking into account similarity of their dominant features and the highest ranked attributes of groups. The presented approach is investigated, taking into considerations learning style dimensions as student attributes. The method is evaluated on the basis of experimental results obtained for data of different groups of real students.

Effective Multi-label Classification Method for Multidimensional Datasets

Multi-label classification, contrarily to the traditional single-label one, aims at predicting more than one predefined class label for data instances. Multi-label classification problems very often concern multidimensional datasets where number of attributes significantly exceeds relatively small number of instances. In the paper, new effective problem transformation method which deals with such cases is introduced. The proposed Labels Chain (LC) algorithm is based on relationship between labels, and consecutively uses result labels as new attributes in the following classification process. Experiments conducted on several multidimensional datasets showed the good performance of the presented method, taking into account predictive accuracy and computation time. The obtained results are compared with those obtained by the most popular Binary Relevance (BR) and Label Power-set (LP) algorithms.

Improving Children Diagnostics by Efficient Multi-label Classification Method

Using intelligent computational methods may support children diagnostics process. As in many cases patients are affected by multiple illnesses, multi-perspective view on patient data is necessary to improve medical decision making. In the paper, multi-label classification method—Labels Chain is considered. It performs well when the number of attributes significantly exceeds the number of instances. The effectiveness of the method is checked by experiments conducted on real data. The obtained results are evaluated by using two metrics: Classification Accuracy and Hamming Loss, and compared to the effects of the most popular techniques: Binary Relevance and Label Power-set.

Improving Multi-label Medical Text Classification by Feature Selection

Multi-label text classification plays a significant role in information retrieval area. The effectiveness of the techniques is especially important in the case of medical documents. In the paper, application of feature selection methods for improving multi-label medical text classification is discussed. We examine combining problem transformation methods with different approaches to feature selection techniques including the hybrid ones. We check the performance of the considered methods by experiments conducted on the dataset of free medical text reports. There are considered cases of different number of labels and data instances. The obtained results are evaluated and compared by using two metrics: Classification Accuracy and Hamming Loss.

Feature selection for classification incorporating less meaningful attributes in medical diagnostics

In medical diagnostics there is a constant need of searching for new methods of attribute acquiring, but it is difficult to asses if these new features can support the existing ones and can be useful in medical inference. In the paper the methodology of discovering features which are less informative while considering independently, however meaningful for diagnosis making, is investigated. The proposed methodology can contribute to better use of attributes, which have not been considered in the diagnostics process so far. The experimental study, which concerns arterial hypertension as one of the civilization diseases demanding early detection and improved treatment is presented. The experiments confirmed that additional attributes enable obtaining the diagnostic results comparable to the ones received by using the most obvious features.

Using frequent pattern mining algorithms in text analysis

In text mining, effectiveness of methods depends on d cument representations. The ones based on frequent word sequences are used in uch tasks as categorization, clustering and topic modelling. In the paper a comp arison of different algorithms for finding frequent word sequences is presented. There are considered techniques dedicated for market basket analysis such as GSP an d PrefixSpan as well as a method based on a suffix array. The investigated te chniques are compared with the new approach of searching maximum frequent word seq uences in document sets. Performance of the algorithms is examined taking in to account execution times for the considered test collections.

Unsupervised feature selection using reversed correlation for improved medical diagnosis

Statistical inference has been usually used for medical data analysis, however in many cases it appears not to be efficient enough. Cluster analysis enables finding out groups of similar instances, for which statistical models can be built more effectively. In the paper a feature selection method for finding clustering attributes, which are supposed to improve performance of statistical analysis, is proposed. The method consists in selecting reversed correlated features as attributes of cluster analysis. The proposed technique has been evaluated by experiments done on real data sets of cardiovascular cases. Experiment results showed that the presented approach stimulates efficacy of statistical inference applied to medical diagnosis.

Topic Modeling Based on Frequent Sequences Graphs

Huge amount of documents in the digital libraries requires automatic and efficient techniques for their management. Topic modeling is considered as one of the most effective method of automatic document categorization. In the paper, contrarily to using “bag of words”, phrase based topic modeling is considered. We propose a methodology, which consists in building frequent sequences graph and finding significant word sequences. Graph structure makes possible selecting sequences of words which are characteristics for different topics. The methodology is evaluated on experiments performed on real document collections. The results are compared with the ones received by using LDA algorithm.

Effective Outlier Detection Technique with Adaptive Choice of Input Parameters

Detection of outliers can identify defects, remove impurities in the data and what is the most important it supports the decision-making processes. In the paper an outlier detection method based on simultaneous indication of outliers by a group of algorithms is proposed. Three well known algorithms: DBSCAN, CLARANS and COF are considered. They are used simultaneously with iteratively chosen input parameters, which finally guarantee stabilization of the number of detected outliers. The research is based on data retrieved from the Internet service allegro.pl, where comments in online auctions are considered as outliers.

Building Contextual Student Group Recommendations with Fuzzy Logic

Groups of learners of similar features are often created in order to diversify the environment accordingly. However student preferences may differ depending on the context of the system usage. Each new student, who intends to join the community, should obtain context-aware recommendation of the group of colleagues matching his needs. In the paper, using fuzzy logic for modeling student groups is considered. We propose to build the possibility-based representation of each group. We assume that context can be modeled by a vector of weights. Then recommendations for new students are determined taking into account a degree of possibility of matching together with the respective context parameters. We examine the presented approach by taking into account learning style dimensions as attributes which characterize student preferences. The method is evaluated on the basis of experimental results obtained for data of different groups of real students.