Based On Rough Set Classification Knowledge Discovery And Information Push

In information dissemination system based on grid, information handled is enormous. If improperly handled, it will result the flooding of information and data rubbish. Therefore, users will be unable to gain the information advantage. Information intelligent push in this thesis is proposed to this problem. It is an important component in information dissemination system and its main objective is to ensure to push valid information to users who need them indeed.User's interest discovery is a key component in information intelligent push. It can find the potential interest of different users to different kinds of information by studying historical visiting information behaviour of user. In fact, user's interest discovery can be belonged to category of classification knowledge discovery which can divide user information need into two groups as interested group and uninterested group by classification forecast analysis of user information need which reflects historical visiting information behaviour of user. Classification knowledge discovery model can be built of different classification knowledge technologies. However, some models based on classification knowledge discovery technologies are more complicated or need certain field knowledge which is subjective. As a kind of strong mathematic tool, rough set has a lot of advantages for solving classification problem compared with other classification knowledge discovery technologies. Better rules or knowledge can be mined by it in a fewer of data and these rules are easy to understand.Considering the above analysis, this thesis studies classification knowledge discovery based on rough sets and put an emphasis on the important components including attribute reduction, rules reduction, classification forecast. As for attribute reduction, combining attribute reduction basic concepts, the absolute core computing algorithm is put forward. Applying information entropy theory, attribute importance degree is refined which is more reasonable. Attribute reduction algorithm is improved on the basis of the absolute core and the attribute importance degree which increases computing efficiency. As for rules reduction, kernel value is refined which is suitable to inconsistent decision table. Combining the refined kernel value and attribute importance degree classification rules reduction algorithm is proposed which is suitable to not only consistent decision table but also inconsistent decision table. As for classification forecast, classification forecast algorithm based on similarity degree is applied to solve the problem which no rules match the new object.The above algorithms proposed in the thesis are also applied to information intelligent push. On one hand, functions of information intelligent push are analyzed and designed on the basis of the architecture of information dissemination system. On the other hand, the application of classification knowledge discovery based on rough sets in information intelligent push is researched to implements user's interest discovery.