Academia.eduAcademia.edu

Figure 3 – uploaded by Dia Eddin Abuzeina

See full PDFdownloadDownload figure
TABLE IV. THE ACCURACY OF EACH CLASS We repeated the experiments for different LDA dimensions in the transformed space with the following values: one, two, and three dimensions. The reported accuracies (with DF=65) are 46.0%, 63.2%, 74.4%, respectively. We then repeated the experiments using DF equal to 40. This DF gives 923 features; the accuracy scored 70.0% using both eigenvalue

Table 3 IV. THE ACCURACY OF EACH CLASS We repeated the experiments for different LDA dimensions in the transformed space with the following values: one, two, and three dimensions. The reported accuracies (with DF=65) are 46.0%, 63.2%, 74.4%, respectively. We then repeated the experiments using DF equal to 40. This DF gives 923 features; the accuracy scored 70.0% using both eigenvalue

Related Figures (6)

classifiers. In fact, utilizing dimensionality reductior techniques is extremely important especially for textual applications that have high-dimensional data which sometimes beyond the hardware capabilities. Text classification is generally about ten thousands dimensions in common. therefore, a dimensionality reduction technique can be though as data compression technique since it projects the original data into a smaller space. For example, It indicated in [1] that using dimensionality reduction significantly improve __ the performance. Reference [2] indicated that dimensionality reduction is the process of finding a suitable lower-dimensional space for several reasons such as: exploring high-dimensional data to discover structure that lead to the formation of statistical hypotheses, visualizing the data using 2-D or 3-D. and analyzing the data using statistical methods such as clustering or classification. Dimensionality reduction typically employed in very high-dimensional space domains such as date visualization, data mining, and information retrieval (IR) [3] Even though using dimension reduction is extremely important in many applications, however, [4] listed some of constrains such as the complexity of the algorithms for data reduction. predictive/descriptive accuracy (relevant features), and representation of the Data-Mining model (simplicity of representation). The concept of dimensionality reduction is something like what Fig. | presented. It shows a transformation process of textual features to generate a reduced dimensional space; the reduced feature vectors are arbitrary chosen to be three dimensions. Fig. 1. Atransformation process to generate a lower dimensitonal space.
classifiers. In fact, utilizing dimensionality reductior techniques is extremely important especially for textual applications that have high-dimensional data which sometimes beyond the hardware capabilities. Text classification is generally about ten thousands dimensions in common. therefore, a dimensionality reduction technique can be though as data compression technique since it projects the original data into a smaller space. For example, It indicated in [1] that using dimensionality reduction significantly improve __ the performance. Reference [2] indicated that dimensionality reduction is the process of finding a suitable lower-dimensional space for several reasons such as: exploring high-dimensional data to discover structure that lead to the formation of statistical hypotheses, visualizing the data using 2-D or 3-D. and analyzing the data using statistical methods such as clustering or classification. Dimensionality reduction typically employed in very high-dimensional space domains such as date visualization, data mining, and information retrieval (IR) [3] Even though using dimension reduction is extremely important in many applications, however, [4] listed some of constrains such as the complexity of the algorithms for data reduction. predictive/descriptive accuracy (relevant features), and representation of the Data-Mining model (simplicity of representation). The concept of dimensionality reduction is something like what Fig. | presented. It shows a transformation process of textual features to generate a reduced dimensional space; the reduced feature vectors are arbitrary chosen to be three dimensions. Fig. 1. Atransformation process to generate a lower dimensitonal space.
* Tt is not algebraic summation since the common words not counted. IL. THE CORPUS INFORMATION The preprocessing include deleting all characters that are out of the Arabic alphabetic characters. It also include deleting numbers, commas, full stops, and all other symbols. A normalization process also performed to change some Arabic characters such as (|!) and (|-!). The stoplist declared that contains all common words in the five categories of training documents. The proposed method summarized in the following algorithm: To investigate the proposed method, we prepared an Arabic text corpus that contains 1,750 documents for training and 250 documents for testing. The training set contains 929,205 words with 80,156 unique words. The collected documents belong to five categories as shown in Table II. The corpus was prepared with help by Alqabas newspaper in Kuwait [23]. Table II shows the statistical information of the corpus used.
* Tt is not algebraic summation since the common words not counted. IL. THE CORPUS INFORMATION The preprocessing include deleting all characters that are out of the Arabic alphabetic characters. It also include deleting numbers, commas, full stops, and all other symbols. A normalization process also performed to change some Arabic characters such as (|!) and (|-!). The stoplist declared that contains all common words in the five categories of training documents. The proposed method summarized in the following algorithm: To investigate the proposed method, we prepared an Arabic text corpus that contains 1,750 documents for training and 250 documents for testing. The training set contains 929,205 words with 80,156 unique words. The collected documents belong to five categories as shown in Table II. The corpus was prepared with help by Alqabas newspaper in Kuwait [23]. Table II shows the statistical information of the corpus used.
Fig. 6. The Data in the transformed space of three classes. Fig. 6 shows a chart that represent how LDA transforms 3- dimensional input feature vectors into a reduced 1-dimensional feature vectors with preserving the differences among the classes’ centroids. Even the points clearly separated into three different classes; however, some overlaps might appear in complex applications. In Fig. 6, the new transformed values used for both coordinates x, y, and z. For example, the value 9.91 is used for x, y, and z coordinates. Naturally, the new instance assigned to the closest centroid in the projected space.
Fig. 6. The Data in the transformed space of three classes. Fig. 6 shows a chart that represent how LDA transforms 3- dimensional input feature vectors into a reduced 1-dimensional feature vectors with preserving the differences among the classes’ centroids. Even the points clearly separated into three different classes; however, some overlaps might appear in complex applications. In Fig. 6, the new transformed values used for both coordinates x, y, and z. For example, the value 9.91 is used for x, y, and z coordinates. Naturally, the new instance assigned to the closest centroid in the projected space.
Fig. 5. Three class data set in the orginal space.
Fig. 5. Three class data set in the orginal space.
Fig. 7. Classification using Linear Classification Functions
Fig. 7. Classification using Linear Classification Functions
ABLE III. THE RESULTS USING EIGENVALUE DECOMPOSITION AND LINEAR CLASSIFICATION FUNCTIONS Table IV shows the accuracies of each class (DF=65) using both eigenvalue decomposition and linear classification functions. Four LDA dimensions used with the results presented in Table IV.
ABLE III. THE RESULTS USING EIGENVALUE DECOMPOSITION AND LINEAR CLASSIFICATION FUNCTIONS Table IV shows the accuracies of each class (DF=65) using both eigenvalue decomposition and linear classification functions. Four LDA dimensions used with the results presented in Table IV.
Related topics:
Computer ScienceArtificial IntelligenceLinear Discriminant AnalysisEigenvalues and Eigenvectors
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved