Research And Application Of Personalized Search Algorithm Based On Levenberg-Marquardt Neural Network

For decades,driven by the rapid development of Internet,the e-commerce grows vigorously,because of which various e-commerce platforms came into being accordingly.In China,in order to contend for more users,these e-commerce platforms,led by Tmall and JD,keep exploring new fields of their business.In addition,the number of commodities soared in an exponential way with the increase of merchants.For the same item,customers will get dozens or even hundreds of records that presents diverse sales volume,reviews and qualities after searching it,not to mention congeneric items or even a fuzzy key words.In such case,consumers are easily get confused.At present,e-commerce platforms have launched their own personalized services to solve the problem of overload information,but some problems,such as inaccurate recommendation and sparse data,still exists.To build the user's preference model,this paper employs algorithm based on LM-BP Neural Network to establish a feature attribute matrix first.Next,in view of the the large number of acquired features on basis of users' historical data,the paper reduces the dimensionality of data,after which the algorithm based on LM-BP neural network is used to build a user-item rating matrix.Furthermore,for items that have not been scored,it is possible to accurately reflect the user's interests and preferences through these experiments.Nevertheless,there are always new users who register for e-commerce websites.The relatively lack of information,incompleteness and uncertainty for new users make the users' preference model built fail to be applied to them,to satisfy the actual intelligent needs of users and to discover users' preferences effectively.For the problem of data sparseness,this paper,by using the collaborative filtering algorithm based on grey relational clustering,calculate the similarity among users and derive the new user's preference model through weighted average.The building of user's preference model aims to provide different users with varied and targeted search results when searching.To prioritize the products that are of interest to users,a search engine,which not only responds quickly but also combines with the built user model,is built to rank the searched items according to the user-item rating matrix and to present them to the users.In this paper,the user-item rating matrix,generated by the built user preference model,is stored in the high-performance redis cache system.The ElasticSearch search server is applied to construct an efficient search system,and a self-defined rating plug-in is used to obtain the user product ratings in redis,which are assigned to the searched items,so that the rating effect produced by each user's different interest model is certainly different.Finally,the aim that various search results are presented to different users can be achieved.The system that has been tested proves to be quick-responded,and it reflects the user's interests and preferences well.Based on the Manjiwang e-commerce platform in Chongqing,this issue tries providing personalized services for it.After being tested,personalized search list can be displayed while the user is searching,achieving the desired effect.