Web数据挖掘的应用英文文献和中文翻译(3)
The above process is carried out for our entire testing data set. Partition of training and testing data sets are given in result analysis section. Fig. 3 depicts the entire structure of training and testing data set process with knowledge base.
4. Experimentation and Result analysis
In this section, extensive experimentation has been conducted on our proposed model and evaluated the obtained results using accuracy measures such as Precision, Recall and F-measure. For different evaluation purpose our data set has been split into three criteria like 70:30, 60:40 and 50:50 of Training: Testing dataset respectively.
4.1. Accuracy Analysis
Training dataset is the set of data that we use to train the system. It is basically used in various areas of information science. Testing dataset is the set of data used in various areas of information science to check the validation of the system which is trained based on the training dataset. Theoretically, 20% of the data is used for training the system and the rest of the 80% of data is used to test the validation of the system15. But, it is not a feasible fact in practical.
Hence, we have considered three categories of data viz., 70:30, 60:40 and 50:50. Where, 70, 60 and 50 refers to the percent of URLs we have considered to train the system and 30, 40 and 50 refer to the percent of URLs that we have used to test the validation of the trained system. The results obtained after training and testing processes is discussed in the following sections.
4.1.1. Results obtained from 70:30 dataset
XML
URLs True
Positive True
Negative False
Positive False
Negative Precision Recall F-Measure Accuracy
CODE 21 286 24 0 0.4665 1.0000 0.6300 92.7
HTML 150 161 0 20 1.0000 0.8800 0.9370 93.9
PURE 4 319 0 8 1.0000 0.33.. 0.5000 97.6
RSS 134 197 0 0 1.0000 1.0000 1.0000 100
Avg 0.7667 96.4%
Table 1 Results obtained for 70:30 dataset
In the Table 1, 70% of the data is considered as training set and the rest (30%) is used as testing data. With this set of data, we have achieved an average accuracy of 96.4% and average F-measure of 0.7667. Graph has been plotted for obtained F-Measure and Accuracy as shown in Fig. 4.
For few category our proposed algorithm achieves less recall and precision value(s) because of tag similarity with other category XML URLs miss classification occurs
Fig. 4 Accuracy analysis for 70:30
In Table 2, 60% of the data is considered as training set and the rest 40% is used as testing data. With this set of data we have achieved an average accuracy of 97.35% and an f-measure of 0.8731.
4.1.2. Results obtained from 60:40 Dataset
Table 2 Results obtained for 60:40 dataset.
XML
URLs True
Positive True
Negative False
Positive False
Negative Precision Recall F-Measure Accuracy
CODE 26 378 11 1 0.7021 0.9622 0.8120 97.11
下一篇：JSP投票系统英文文献和中文翻译