Wenxue Hong

A new approach of rules extraction for word sense disambiguation by features of attributes

A new approach of rule extraction by features of attributes is proposed.Simple class exclusive attributes and composite class exclusive attributes are calculated.The attributes are used in rule extraction for WSD of English preposition on.The accuracy of WSD is improved. Comparative result shows the new approach has a few advantages over the well-formed SPOAD approach. Classification is an important issue in data mining and knowledge discovery. It is a significant issue to develop effective and easy approach of rule extraction for classification. A new approach of rule extraction by features of attributes is proposed in this article for word sense disambiguation (WSD). English preposition on is taken as a target word of WSD, a data set of 600 samples is randomly selected from a 350,000 words corpus. Semantic and syntactic features are extracted from the context, and the corresponding formal context is generated. The rules for WSD of English preposition on are extracted based on the theoretical descriptions and calculation of the simple class exclusive attributes and composite class exclusive attributes. The extracted rules are used in the WSD of English preposition on, and the accuracy reaches 93.2%. The results of the comparative analysis show that the proposed feature of attribute approach is simpler, more effective and easier to use than the existing well-formed structural partial ordered attribute diagram approach.

Feature Extraction and Analysis of Ovarian Cancer Proteomic Mass Spectra

The use of mass spectrometry(MS) as a analytical tool in proteomics is poised to revolutionize early cancer detection and biomarker identification. Although proteomic mass spectra has shown the promising potential of finding disease-related protein patterns, key challenges remain in the processing of them especially for the curse of dimensionality. In the present study, an alternative approach to feature extraction from MS data of ovarian cancer is proposed. The proteomic mass spectrum data after preprocessing are first wrapped into information images that are accordingly mapped to binary images under adaptive threshold. The energy curves of binary images are the result of dimensionality reduction that make up of the alternative biomarker patterns that can be used to classify cancer samples from non-cancer ones using similarity. Applying the procedure to mass spectra of proteomic analysis of serum from ovarian cancer patients and serum from cancer-free individuals in the Food and Drug Administration/National Cancer Institute Clinical Proteomics Database, a sensitivity of 98%, a specificity of 95% and a positive predictive value of 95.15% is obtained.

Research on the Radar Chart Theory Applied to the Indoor Environmental Comfort Level Evaluation

We endeavour to provide a novel theory to evaluate environmental comfort level. A method that fuses multidimensional environmental parameters based on radar chart representation theory is provided in this paper. The value of the parameters is transformed into the linguistic description according to linear and nonlinear membership creation method. Consequently, an integrated and objective description about environmental comfort level is given. A measuring and testing system of indoor environmental comfort level is designed here by MSP430 MCU. The design is given and better performance is achieved. It is our belief that this method can be used in many other areas, where the quantitative and qualitative information transform is needed

Deep First Formal Concept Search

The calculation of formal concepts is a very important part in the theory of formal concept analysis (FCA); however, within the framework of FCA, computing all formal concepts is the main challenge because of its exponential complexity and difficulty in visualizing the calculating process. With the basic idea of Depth First Search, this paper presents a visualization algorithm by the attribute topology of formal context. Limited by the constraints and calculation rules, all concepts are achieved by the visualization global formal concepts searching, based on the topology degenerated with the fixed start and end points, without repetition and omission. This method makes the calculation of formal concepts precise and easy to operate and reflects the integrity of the algorithm, which enables it to be suitable for visualization analysis.

The New Graphical Features of Star Plot for K Nearest Neighbor Classifier

The graphical representation or graphical analysis for multi-dimensional data in multivariate analysis is a very useful method. But it rarely is used to the pattern recognition field. The paper we use the stat plot to represent one observation or sample with multi variances and extract the new graphical features of star plot: sub-area features and sub-barycentre features. The new features are used for the K nearest neighbor classifier (KNN) with leave one out cross validation. Experiments with several standard benchmark data sets show the effectiveness of the new graphical features.

Parallel Filter: A Visual Classifier Based on Parallel Coordinates and Multivariate Data Analysis

Multivariate visualization techniques are often used as assistant tools for classification tasks up to now. However, few classification systems fully utilize the capability of multivariate visualization and integrate them with multivariate analysis algorithms into a compact system. We propose an interactive visual classification model based on some multivariate graphical presentation in this paper. As an example of it, a visual classifier based on parallel coordinates plot is developed. The multivariate data is first mapped to the parallel coordinates plot, and then an optimizer based on linear discriminant analysis optimizes it into the visualization more fit for classification tasks. This optimized visualization then can be processed by decision tree algorithm and attain classification rules. It has the merit of making the invisible visible and users can steer the classification process, consequently favor the understanding and knowledge discovery of original data.

Parallel Dual Visualization of Multidimensional Multivariate Data

A visualization technique based on the ideas of scatter plot matrix, star diagrams and parallel coordinates plot is developed in this paper. The principles and algorithms of these visualizations are analyzed and compared from a geometrical perspective. It shows that the three coordinate-bused geometrical visualizations have a unified mathematical foundation and presentation ways thus can he combined into a single visualization that we call parallel dual plot. The new visualization is created by firstly transforming a scatter plot into a star glyph, and then the star glyph is presented by parallel coordinates. Thus, this approach provides a point-to-point mapping and effectively overcomes the over-plotting problem of parallel coordinates. Moreover, this technique has merits of simple algorithm and easy interpretation. Example of application is demonstrated at the end of this paper.

A New Method for Dimensionality Reduction based on Multivariate Feature Fusion

Dimensionality reduction is the process of mapping high-dimension patterns to a lower dimension subspace. When done prior to classification, estimates obtained in the lower dimension subspace are more reliable. We propose a novel method based on graphical multivariate feature fusion and use it to offer a visual representation of high dimensional data. The graphical processing method we propose, relies on using a multilayered structure of feature fusion which produces as output of the lower dimensional representation. We implement feature fusion by combining method of feature selection and feature extraction. Experiments on the data set of machine learning database indicate the novel method we propose provides better representation than Fishers linear discriminant (FLD) and some other nonlinear methods of dimensionality reduction that are often used.

Visual Pattern Recognition Method Based on Optimized Parallel Coordinates

A visual pattern recognition method based on optimized parallel coordinates is proposed in this paper. We first introduce the traditional theory of parallel coordinates and indicate that parallel coordinates has a potential for classification tasks due to its projective transformation interpretation. Nevertheless, some optimization is needed. The main aim of optimization is to hide the valueless information and reveal the most valuable information for classification. Three interaction operations are proposed to do this work. We demonstrate by several examples how to solve some pattern recognition problems graphically, we also point out that some start-of-the-art classification techniques such as Fisher linear discriminant analysis, decision trees and support vector machines are related to and can be visualized by this method. By cooperation and union of this method with some machine learning and human-machine interaction techniques into a system, the black boxes of some pattern recognition techniques are expected to be opened up, the original data, the process of classification and final results will be more transparency and understandable.

Classification of Vegetable Oils Based on Graphical Presentation and Bivariate Discriminant Node Model

A novel method for classification of vegetable oil samples by fatty acid composition is proposed. It is based on classification trees with bivariate linear discriminant node model, which can simultaneously reduce tree size, improve class prediction, and enhance data visualization. Exploratory analysis by parallel coordinate plot indicates outliers exist in oil samples. Linear discriminant analysis (LDA) is sensitive to outliers; consequently when it is applied to 96 samples of known vegetable oil classes, three oil samples are misclassified. This paper considers a nonparametric alternative to LDA: classification trees with bivariate linear discriminant node model. The classification tree method is found to be a good model for the classification of different edible vegetable oils. Compared with other methods such as counterpropagation neural network, this method has merits of robust to outliers, easy interpretation and enhanced interactive data visualizing