Research On Some Key Technologies Of Educational Resource Recommendation Service

With the development of network technology and educational informatization, educationalresource management system has been widely used in each stage of education. How toimprove the intelligent and the utilization efficiency of education resource system has beencommon concern.With the increase of educational resources in the network education system, finding theintrested resources has become more and more difficult. By adding the educational resourcerecommendation service in the network education system, the students can be free from thehuge information. It also make the center of network education system transform fromresources to students and develop to a higher level in network service. In this condition anddemand, the educational resource recommendation service technology has been developmentgradually.In this dissertation, we study on the key technologies of educational resourcerecommendation service, such as service model, resource features representation andrecommendation algorithm. The main work and contributions of this dissertation include:1. Based on the analysis of characteristics and disadvantages of the current networkeducation systems, a whole plan of network education system is designed. Theeducational resource recommendation service based on web mining technology is addedto the network education systems.The service model is given.The recommendationprocess can be divided into two stages: personalized information extraction andeducational resource recommendation.At personalized information extraction stage, apersonalized information extraction method base on web usage mining technology ispresented. At educational resource recommendation stage, a educational resourcerecommendation base on web content mning technology is presented.2. Focusing on the problem of resource features representation, use the content of textresources or the description of multimedia resources as the presentation of resourcecontent. A Chinese high-frequency word extraction algorithm based on tree structure andweighted entropy is presented. The algorithm can extract Chinese high-frequency wordsform the presentation of resource content without the support of the dictionary. Thesehigh-frequency words are use as resource features representation.3. Focusing on how to reduce training time, the manifold learning technology is used to reduce the dimensions of resource features representation. Focusing on how to improvethe efficiency of recommendation service, the active learning technology is used in therecommendation. The recommendation algorithm need a number of label informationwhich is accumulated in the using. The labelling time is reduced by the active learningtechnology, so the efficiency of recommendation service is improved.4. Focusing on the problem of cross-domain educational resource recommendation, atransfer learning algorithm combined with data timeliness and weight constraint ispresented. This algorithm can reflect the timeliness of data because a timeliness functionis added to the process of weight distribution. A operation of weight constraint is added,so this algorithm has more generalization capability.5. Focusing on the problem of large scale resource recommendation, a distributedrecommendation algorithm is presented. This algorithm adopts supervised Hebb learningrules. Simulation results show that this algorithm can solve the problems caused by largedataset, such as the large scale network and the long training time.The conclusions in this dissertation can provide the academic references for the research oneducational resource recommendation service and diversify the contents of research. It pushthe development of research on features representation and recommendation algorithm.