Optimized Model Of Recommendation System For E-Commerce Website

The E-commerce environment includes all online activities and business operations achieved between multiple parties using electronic techniques.With the huge development of internet and E-commerce websites; when consumers choose their needs of items and commodities, they confront some serious problems of data overloading. Therefore; many website researches and projects have focused on recommendation system development, in order to provide users more individual recommendation services.Recommendation system has become serious business tools used by many of the largest commerce websites, in order to provide the users more effective and efficient way to find their interested products. The recommender systems work like salesman who provides users advices and services to help them find the commodities and items they are interested in.However, with the wide use of recommendation services, many common challenges and problems come out, such as real-time, sparsely of information, cold start problem and recommendation quality. The current recommender systems still meet some issues which affect the system work and advantages; such as Limited content analysis, Over-Specialization, Shilling attacks, Diversity with Long Tail and Scalability problem.So to solve these issues we have built a new model of recommender system which based on hybrid recommendation techniques and combined with data mining clustering technology to overcome the shortcoming points and provide the best recommendation results which meets all kind of users’ interests and needs.So the purpose of this work is to optimize the recommendation system by creating a new model of recommendation system with different services in a global e-commerce website.In this model the most effective data sources are integrated to increase the accuracy of recommendations system, which provides the client more intuitive browsing categories interface.The main services of our recommendation systems are five functions which can summarized as the following:Identifications based on search data:The recommender system generates the search engine data, search log history, and clicked URLs. Then use the model of search data with user profile info to provide the user a recommendation results according to these generated search data. We have proposed and applied an improved STC (Suffix Tree Clustering) algorithm combined with user interest’s profile to perform search recommendation service.Identifications based on ratings info:Generate the rated items, and use the model of rating info, items info and user profile data to provide recommendation results. We have used neighbor clustering method classified by Support Vector Machine (SVM) to classify sub-clusters of ratings data, in order to provide users recommendations based on their rating info and their similar users ratings data.New items recommendation:For new items which have no rating data, purchased information or even items features details, the system uses content-based model to analyze the input features of new items and similarity with rated items to provide recommendation, we have proposed a new techniques of clustering to solve the cold start issue which exist in the current recommendation system.New user’s recommendation:Two kinds of users in the system:users who just registered into website, users who did have an account without any rating or purchase data, and second kind is the active users who did have purchased, rated or even search logs. The recommender system uses new users profile data model to analyze and measure similarity with active users and provide new users a real-time recommendation.Recommendation based on location:The system analyze the user data including his network IP and user profile to determine his location, and then use the identification data of users locations to provide consumers most interested items according to same location user’s data.Hot releases by Admin of shop owner:The system provide some recommendation, such as hot releases items, discount promotion, wholesale prices, and...etc. According to the system algorithms, the admin or shop owner has the authority to manage and control this section of recommendation results; in order to improve the business strategy of e-commerce website.Our new model of recommender system belongs to a complete personalized recommender system using data mining clustering technologies which highly considering the recommendation quality, real-time recommendation, and proposed solutions for problems such as cold start and other issues. That makes our system an adaptive and scalable recommendation system. When the users browse the website, the data sources will automatically combine to incorporate the derived structure and associate items for each category into a new browsing recommendation interface.The advantages of this model will assist the users to discover their real interested items with flexibility and high efficiency; it also provides solutions for some serious problems and challenges that exist in the current recommendation services.Data mining technology and clustering algorithms have been proposed and applied to perform the idea of this work, in addition we have used multiple cross-validation technique methods to evaluate and measure the accuracy and quality of recommendation prediction results..NET Framework as the implementation environment for our application website; ORACLE Database System is used for database management.