Research On The Recommendation Effect Of Personalized Recommendation System Of E-commerce

With the increasingly prominent problem,data information overload,brought about by the development of big data,research about personalized recommendation has gradually become one of the research hotspots in this field.However,after reading the literature,it is found that current personalized recommendations generally have problems such as poor experience and inappropriate recommendations.What's more,few studies have paid attention to users' actual experiences of using the recommendation system,starting with users' actual perceptions to evaluate recommendation effect.And there are no scholars have paid attention to the ability of recommended products can satisfy users' initial wishes,and how attractive they are to users.So,the purpose of this study is to propose an evaluation model which can properly measure products' recommendation effect based on users' experiences and true feelings.According to the survey,Taobao's buyers can be divided into three types based on purchasing purpose:full stereotype,semi-stereotype and unfixed type.Among them,the semi-stereotype users contribute the most to Taobao's sales,they often have fuzzy preliminary purchase targets for the main product attributes before purchasing and make purchase decisions while comparing during the purchase process,they are also easily affected by recommended products and recommendation information such as product advertisements.In addition,although recommendation results are ultimately presented to users,users cannot know how the recommendation algorithm works,which allows users to only subjectively evaluate recommendation effects based on the received recommended product information,the degree of meeting users' needs and their own feelings,etc.Therefore,this paper focuses on the real experience process of Taobao semi-stereotyped users using Taobao recommendation platform to purchase the required products,based on recommended products' data information which users can get and their perceptions,and then proposes recommendation effect evaluation model integrated by multiple methods such as triangular fuzzy number theory,fuzzy analytic hierarchy process,fuzzy optimization model,TOPSIS and super efficiency DEA.What's more,users usually refresh the recommendation list to obtain new recommendation results during using,so the first level of the evaluation model builds an index system about recommendation accuracy to evaluate recommendation accuracy for different refresh times of the same product.In order to measure whether the recommendation effect is different between different types of commodities,so as to verify the feasibility of the evaluation model from the negative side,so the second level of the evaluation model is to classify commodities from two dimensions of whether there are purchase records and the selectivity of product attributes,and then use the DEA model to evaluate the relative recommendation efficiency between different commodities.Finally,because this paper strives to truly restore the user 's actual using processes but Taobao has advanced anti-reptile technology,this article combines Python-picture recognition technology to manually collect 1,310 recommendation data about using processes of a semi-finalized user purchased 5 required products to verify the feasibility of the proposed evaluation model.Since few scholars currently build evaluation models from users' real feelings,this study is innovative and has positive theoretical and practical significance for expanding the research direction of personalized recommendation effect evaluation.