Research On Key Technology Of Technology Transfer Platform

With the deepening of industry,academia and research work,we have made many achievements and faced many problems.One of the most prominent is that,demand and technology does not match.Enterprises can not find technology,and technicians do not know where there are demanded.In order to solve this problem,the Government Science and Technology Bureau conducts school-enterprise communications and organizes offline activities on a regular basis.However,the effect is not very satisfactory.At present,the conversion rate of scientific and technological achievements in our country is only 30%,far lower than 70%of that in developed countries.After the accumulation of technology transfer in recent years,we have obtained a lot of information on Talent,resources and information.However,this information is often messy.How to improve the utilization rate of these information and strengthen the exchange of information among talents,enterprises and governments have become the key to the transfer of technology.Therefore,we need to sort out the resources,classification,which will be able to recommend the appropriate information to the corresponding user.The current classification mainly considers vector data.However,the data we get usually contains more information,including its source and position of words.The information can be directly expressed in tensor space,Structure information.In order to make full use of these data sources for classification,this paper proposes a method of tensor text classification,using the tensor model for textual representation and using incremental model to improve the classification efficiency.On the other hand,About recommendation system,the reality of the recommendation should be dynamic and efficient.While The commonly recommended method is often static,offline.In this paper,we propose a parallelizable recommender scheme based on the time factor and the local low rank hypothesis of the recommended matrix,and prove its accuracy and efficiency through experiments.The work of this paper is mainly reflected in the following aspects1.Complete data acquisition part,through the distributed crawler station access to information such as data,pictures and other information.2.Combined with the supportive tensor,a tensor classification model that can be trained incrementally is proposed,and a textual tensor model is built by using textual attribute information to classify it efficiently.3.Combined with the matrix decomposition recommendation system,a tensor decomposition recommendation system is proposed for dynamic recommendation,and its parallelism is verified.