Font Size: a A A

Personalized Recommendation Strategies And Application For E-Commerce User In Complex Context

Posted on:2017-04-20Degree:DoctorType:Dissertation
Country:ChinaCandidate:C H XuFull Text:PDF
GTID:1109330488471722Subject:Business management
Abstract/Summary:PDF Full Text Request
The emerging and widely utilizing of new technologies like E-commerce, along with the accumulation of enormous user data, provide an unprecedented developing space for exact-marketing and dynamic supply chain optimization. However, the existence of numerous information also occur problems such as resource-overload and information-mislead, which has become increasingly serious. For individual user, how to acquire useful content from information ocean quickly and accurately has become one prior issue. While for corporation user, how to dig customers’potential needs efficiently, enhance intelligence level of information searching and pushing, improve individualized service quality in this fierce competitive environment, has been put top in modify list in its E-commerce activities. To a certain extent, the creation of personalized recommendation technology has solved the dilemma between information diversity and customer needs specialization. Almost all the E-commerce platforms, such as Amazon, Taobao, Jd., has applied various kinds of recommendation system more or less. However, with the development of E-commerce and the increasingly complicated customer context, how to fulfill personalized needs has become a new trend in personalized recommendation service research.Complex user situations typically include contextual situation, social relationship situation and ontology situation, in the study of situational personalized recommendation strategy. Distinct from traditional research perspectives, this study incorporates vertical and horizontal methods. Previous studies have focused on the ternary elements in model construction, i.e. personalized recommender knowledge modeling based on various situations, contextualized analyses of user preferences, context-awared recommendation methods. Results may slackly satisfy some e-businesses’ requirements. In this paper, e-business customers and corporations are taken into account, firms are divided coarse-grained sets, and personalized recommendation strategy that meets the requirements of complex contexts is designed. The user complex context includes individual context, social relationships context. This paper divides E-commerce entriprises into three categories:single dimension context information entriprise, partial dimensions context information entriprise and rich dimensions context information entriprise. By in-depth analyzing typical personalized recommendation strategies and related models, this paper tries to find out the deficiencies of current researches and comes up with one personalized recommendation model satisfying various E-commerce platforms. This paper analyzes personalized recommendation model in complex context, focusing on personalized recommendation strategy of E-commerce users, applying data mining as modeling method, aiming at acquiring potential customer needs, improving customer satisfaction and maximizing benefits.Detailed innovations are as following:1. Personalized recommendation method based on resource heterogeneous diffusion in complex context. Personalized recommendation strategy of Collaborative Filtering is widely used because of its simple model, easy to apply and sufficient recommendation results. But it also has problems of data sparsity and cold-start. In order to release these two problems, improve recommendation quality while not adding informational dimension, this paper designs one personalized recommendation strategy of heterogeneous diffusion on bipartite network. As user-item nodes can get different resources depending on their different attraction levels, by computing these resources, this research gets its recommendation results. Results reflect that this model is totally different from similarity computation strategy between collaborative filtering users. Through resources diffusion and spread, this paper enriches the similarity computation database, improves recommendation accuracy and diversity, as well as overcomes data sparsity and cold-start problems to some extent. This recommendation strategy is especially efficient for the corporations with less users and lack access to user multi-dimension information.2. Personalized recommendation method based on individual context and trust in complex context. With the increasing complication of user context, traditional Collaborative Filtering recommendation strategy can no longer meet needs. In order to improve recommendation quality and provide better personalized recommendation service, this paper designs one personalized recommendation model based on Collaborative Filtering, which has involved customer individual context, taken into consideration user preference changed with user degree and affect of trust to similarity computation. The research finds that to cluster users based on individual context before user similarity computation and to add user trust during similarity computation can have positive influence on the improvement of recommendation accuracy and the increase of result diversity. Furthermore, this model can also relieve data sparsity and cold-start properly. To the E-commerce platforms which can acquire user individual context, adopting this modified Collaborative Filtering recommendation strategy by adding limited information dimension can have significant improvement in recommendation result quality.3. Personalized recommendation method based on social relationship in complex context. The merge of social network and E-commerce has led to enriched user information dimensions, resulting in a more accurate personalized recommendation result. This paper gets its recommendation result by designing an personalized recommendation model based on social network context, combining the effect of user preference and user social relationship, and adopting matrix factorization method. This research reveals that merging social network context can improve the accuracy of recommendation result and increase its diversity. Meanwhile, Matrix Factorization method proves to be a good solution to data sparsity and cold start. This personalized recommendation strategy can be applied to various situations and fields, such as friend recommendation with social network and product recommendation in E-commerce platform. It has most significant effect in corporations with user multi-dimension information.This analysis extents research idea of personalized recommendation strategy, enriches its method system. The three complementary strategies formed in this paper provide a systematic solution to corporation personalized recommendation strategy design, which can satisfy needs of various E-commerce platforms.
Keywords/Search Tags:Complex context, Heterogeneous diffusion on bipartite network, Collaborative filtering, Matrix factorization, Personalized recommendation Strategies
PDF Full Text Request
Related items