The Research Of Knowledge Recommender Model Base On The Spreading Activation Process Of Human Memory

With the continuous development and the rapid spread use of the Internet, hugeamount of news event is spreading around the internet all the time. The carrier ofthese kinds of information mainly includes documents, pictures and videos. Becauseof the large quantity, it is difficult for human user to find the useful and requiredknowledge among these information. A lot of time and energy tend to be spentfiltering unwanted noise. The human’s information processing ability is limited whilethe processing ability of computer is comparative unlimited. For this reason, it ismeaningful to study how to use computer to recommend useful knowledge to userand help user to understand these knowledge.In the research of knowledge recommender model, the core problem ischoosing appropriate model to guide building the knowledge recommender model.The classic recommender systems rely on two kinds of data: the data of the contentof the item and the information of user history. These recommender systems use theinformation of the item or the information about the user to predict the utility orrelevance of the items to a particular user, thus providing personalizedrecommendation. The knowledge is different from the normal items. Users choosethe knowledge by the content of the knowledge. The content of the knowledgedetermines the value of the knowledge. So the knowledge recommender shouldfocus on the content of the knowledge, follow the rule of human cognition processand recommend the knowledge on user’s demand. Taking advantage of theprocessing ability of computer, it can save user’s time and effort. But most of thecurrent recommender systems depend on the user history which causes the cold startproblem and shilling problem. Moreover they are not consistent with the cognitiveprocess and lack of sufficient theoretical support.In order to establish the recommender model special for knowledge, takingadvantage of the current the computer representation model of text knowledge, thispaper build up a knowledge recommender model by drawing on the research resultsof human memory and brain science and combining the matrix theory, game theoryand probability theory. The details and contributions of the dissertation are listed asfollows:First of all, the usage of the different kinds of the Spreading Activation model inknowledge recommendation is analyzed.The Pure Spreading Activation model, Distance Constraint SpreadingActivation model, Path Constraint Spreading Activation model and Fan-outConstraint Spreading Activation of Human Cognition Process are discussed andanalyzed in the paper. These models guide us to build up the knowledgerecommender model of this paper. Secondly, establish the Query Expansion method based on the SpreadingActivation modelTaking advantage of the ALN and using the distance constraint SpreadingActivation, the Query Expansion method, which is not rely on the user search log, isestablished. And our Query Expansion method can offer user semantic relationexpansion and semantic community expansion. These kinds of expansion cannot beoffered by other Query Expansion method.Thirdly, establish the knowledge recommender model based on the SpreadingActivation modelOur knowledge recommender model is based on the fan-out constraintSpreading Activation model and path constraint Spreading Activation model. Andthe rehearse mechanism has been added to our model. Moreover, the attention gamemodel, which is combined the result of brain science and game theory, is establishedto make our model more conform to the custom of human cognition process, andbecomes more useful. The hebb learning process is used to modify the networkaccording to user’s demand. After the recommender the serial correlation terms, thedocuments, as the knowledge carrier, will be recommended to users according to theresults of the term recommendation. The knowledge of different granularity helpuser to better understand the knowledge.