这两天还是纠结于分类模型的准确率。因为对从网上随机摘录的文本进行分类时,结果总是不甚理想,不像使用cross-validation得到的结果那么好。
于是决定使用独立测试集(含1402个实例)进行评估。训练集实例9804个,特征9302个,没有使用特征选择。准确率大约78%,其中“历史”和“艺术”有点分不清。结果如下:
-------------------------------------------------------------------------
weka.filters.unsupervised.attribute.StringToWordVector in:9804
Number of instances: 9804
Number of attributes: 9302
loading test data in:test_segmented......
weka.filters.unsupervised.attribute.StringToWordVector in:1402
weka.filters.unsupervised.attribute.ReplaceMissingValues in:9804
weka.filters.unsupervised.attribute.Normalize in:9804
evaluating.........
=== Detailed Accuracy By Class ===
TP Rate FP Rate Precision Recall F-Measure ROC Area Class
0.91 0.008 0.901 0.91 0.905 0.993 C11-Space
0.455 0.001 0.938 0.455 0.612 0.928 C15-Energy
0.464 0 1 0.464 0.634 0.974 C16-Electronics
0.556 0.001 0.938 0.556 0.698 0.989 C17-Communication
0.98 0.031 0.705 0.98 0.82 0.985 C19-Computer
0.588 0.003 0.833 0.588 0.69 0.96 C23-Mine<