The Design And Implementation Of Product Recommendation System Based On "User Portrait"

Under the leading of big companies such as Taobao,Jingdong,Tianmao and so on,ecommerce is very prevalent in China,nowadays.As each company's business is growing exponentially,this has created massive amount of data that needs to be managed,and transferred to the user.But,as a user it would be very inconvenient if the list of items does not match with his/her requirements.This problem is particularly noticeable on mobile terminals because large number of undesired goods will not only take up the customer's time in searching for better ones,but will also consume a lot of traffic which will result in bad user experience.More importantly,this is the major reason for the loss of users.In order to solve the problem of product overload,this paper designs and realizes the product recommendation system based on the” user portrait”.This system combines the user's profile with automatic recommendation and helps the user in finding their desired goods quickly from a huge list of matching items.This kind of recommendation system will also help in delivering advertisements and suggestions which are in-line with the interests of the user.The main contents of this paper are as follows:Firstly,we will introduce the concept,composition and related theoretical knowledge of product recommendation system based on user profiles.The present state of domestic and international recommendation systems is studied deeply,and the technology needed is elaborated in detail.Secondly,the recommendation system based on user profiles is described in detail from the three aspects of demand,design and realization.This mainly consists of the following two parts:(1)User profile system: This system processes on the user's personal historical behaviour and builds the user interest model based on the ranking matrix.Then,this user interest model will create user label system using the label rules.Afterwards,the user profile can be accessed using data visualization technology ‘E-Charts'.Compared with the traditional systems,this merchandise recommendation system is user-oriented,takes the differences between the users into account,and provides personalized recommendation service according to each user's preference.(2)Product recommendation interface: This interface can return the list of goods depending on the user interest model mixed with a variety of rules and filters the products the user might not be interested in.The provision of data to different types of mobile terminals in the form of interfaces allows for the separation of data structures from transmission performance.This interface is written in the Thrift framework,which allows interaction between systems to have high performance,low latency,and support for synchronous and asynchronous communication.Finally,the test results are provided for the evaluation.The parameter optimization of the product recommendation system is carried out by NDCG algorithm.NDCG corresponds to Normalized Damage Cumulative Gain.The algorithm is one of the most popular choice for testing recommendation system nowadays.It relies on the user's feedback to quantify the user's satisfaction with the list of goods and adjust the system parameters accordingly.The results of the NDCG testing system show that the function and performance of this algorithm meets the requirements and can be used for online applications.