在本論文中,我們應用了一連串類神經網路的技術在網際網路探勘的領域上，使能夠解決在網路資料分類的問題。它首先利用特徵權值探測的方法，從複雜且大量的訓練資料中找出可靠的特徵，再利用這些特徵配合比例式學習向量量化網路的方法找出最適合的群中心點。最後，將群中心點應用於徑向機網路以分類測試資料庫。在實驗中我們將資料依據Session Length來預作分割，再利用這些分割的資料來做實驗，由最後的實驗我們發現做適當的資料分割能夠獲得較好的正確率。 In this thesis, we apply widely-used data mining techniques, neural network, in the user’s characteristics of classification on WWW. In this proposed method, we first utilize feature weight detector networks to discover the reliable features from mass and complex training data on WWW. Secondly, we use the proportional learning vector quantization network to learn the appropriate centroid of each cluster. Finally, we apply radial basis function network associated with the centroid of clusters to classify the test data. We in advance partition the data set into several data sections that are used in experiments according to session length. Experimental result show that it has better classification result than ones using overall data set.
