Uncertainty Mining Based On Implicit Feedback In User Consumption Behavior

With the development of information technology,people are facing increasingly serious information overload,in order to help people quickly and effectively obtain the desired information,the recommended system came into being,as an effective means to solve information overload,recommended system has become one of the research hotspots of the industry circles and academiccommunity,itrecommended information and products to users that they may prefer base on their preferences characteristics and needs which learned by their explicit or implicit feedback information.The recommendation system that based on explicit feedback information is the mainstream,most of the recommended systems ignore the massive implicit feedback information,and the generalization of implicit feedback information makes the recommendation system based on such information have better extension,However,the implicit feedback information can not reflect the user's preferencesdirectly,so it requires more research about how to effectively use the implicit feedback information.This paper explores the implicit feedback in the user behavior log from the indecisiveness perspective of the user's consumer behavior and build the recommended system model.the consumers' indecisivenessrefers to the inability to make quick and assertive decisionswhen they choose among competing product options.Although the user'sindecisivenessthat among the consumption process has been investigated a lot of research in many any fields,such as psychology,economics.However,most of these studies are based on subjective consumer questionnaires-researchers develop questions that are answered by consumers,and questionnaires are usually subjective,and conclusions may not be accurate,andthis paper presents a complete data-driven way in the absence of human intervention,automatically mining the indecisiveness through the consumer online behavior log,make the results more accurate.This papermainly has the following contents:Firstly,it introduces the background and significance of this paper,and the research status of recommendation system that based on implicit feedback.This paper summarizes the key issues in the current implicit feedback and the main recommendation techniques for implicit feedback,including recommendations based on single class collaborative filtering,recommendation of introduction of auxiliary information,and recommendation based on sorting.Secondly,it proposes and improve the method of automatically quantifying the indecisiveness index of each action section without human intervention and the basic framework of calculate theindecisiveness index of users and commodities at the same time.This framework is not limited to electrons Business areas,the same applies to other areas of similar problems.The experimental data also validate therationality of quantitative method proposed in this paper and its rationality after improvement.Furthermore,based on the ambiguity of the observed quantification,as for the drawback that the eigenvector distribution of the sparse users in the IMF model tends to be the same as the priori averages,which leads to the expectation that the model will tend to the global averageof the IMF,it proposes CIMF modelthat can be used to study the indecisiveness index of consumers and products.The model introduces the additional constraint matrix for the user,which makes the forecast of the sparse user more accurate and solved the sparseness of the implicit feedback data to a certain extent.Finally,it uses the Kaggle public data set which derived from the real shopping site of 2.75 million user operations records for model validation.The results show that CIMF reduces the error of the IMF model by 16% and promotes the prediction accuracy of the new sample by 8%,which indicates that the CIMF model is superior to the IMF model in the case of extremely sparse data(0.003% nonzero value).This paper analyzes the user operation data after preprocessing,including the distribution of the number of operations,the time distribution on each commodity,the distribution of operation and so on.It is helpful to understand the behavior of consumers and the decision-making process of consumers.The analysis shows that the higher degree of indecisiveness of consumers,the lower the purchase rate,so the recommended system should help users make quick decisions.Choosing Uncertainty Mining has a lot of potential applications,such as the detection of competitive goods,data-driven access to the information of competitive goods,it can help retailers improve their products and specify competitive strategies,and personalized commodity recommendations,based on this model Recommended system can greatly ease theindecisiveness of consumer;The analysis of consumer "hesitancy" in this article helps to understand consumer preferences while at the same time imprecisely digs into predicting consumer behavior,providing better referral systems to consumers and online retailers,and providing better Personalized services are very helpful.