The Research And Application Of Recommendtion Technology Based On Enterprise Model

To solve the problem of "Information overload", the information servicer adopts recommendation system provide suggestions to customers, help customers get interested information. Recently, the research and application of the recommendation technology focused on two aspects:the user model and the recommendation algorithm. Especially, the user model is very important in the recommendation system, it is the basis of the whole recommendation system construction and determines the design of recommendation algorithm.As the coming of information age, the manufacturing enterprises are developing to the industrialization and informatization. In this process, enterprise needs to constantly obtain information product related. Therefore, we try to build an enterprise model from the product perspective. The main works of this paper as follows:Firstly, this paper proposes a method of enterprise model building. Through this method, we can express the information needs of enterprise by product related feature words and the relationship between features.Secondly, this paper proposes a method of assistant feature words acquisition and judge the relationship between feature words by the KNN classification algorithm. Firstly we get these candidates feature words from product related texts using CRF method. Then we build a formula which contains both positive and negative feature words set to get the score of current feature word. Last, we use active learning algorithm and stochastic gradient descent algorithm train parameter of other feature words. The MAP value and accurate value of the assistant feature words acquisition is0.713and0.754. Then we adopt two methods of feature selection to test the K value in the KNN classification algorithm and through experiment, we verified that when the k=25, the classification result is best and the value is0.871.Thirdly, this paper proposes a method of recommendation algorithm based on the enterprise model, assistant estimate the relation between feature words. This method considers the feature words, the parameter of feature, and the relationship between feature words. Using this recommendation algorithm retrieve documents in the2009patent text sets, the accurate value is0.8035.Finally, this paper implements the recommendation system based on the technology that we proposed and build the actual system for50products of30enterprises based on2010patent text sets.