面向农产品的协同过滤推荐算法研究

发布时间：2018-05-09 01:29

本文选题：协同过滤推荐算法 + 农产品电子商务　；参考：《东北农业大学》2017年硕士论文

【摘要】：近年来,电子商务以方便信息沟通、收付方式灵活便捷等优点,促进社会信息化发展。但是,随着互联网的发展,很多用户深受“信息过载”的困扰,影响购物体验。于是,能够帮助消费者从海量数据中挖掘出其感兴趣的商品的推荐系统应运而生。推荐系统通过分析消费者的个人特征和行为特征,确定用户兴趣,进而向用户进行推荐。对经营者而言,能够提高交易量,达到精准营销的目的;对消费者而言,能够节约时间,快速找到其有意愿购买的商品。农产品电商是整个电商平台中尤为重要的一部分,农产品推荐系统具有增加农民收入、提高第一产业竞争力、促进农业信息化发展等重大意义,然而,现阶段关于农产品电子商务个性化推荐系统的研究较少。本文的研究目标是设计一种面向农产品的推荐算法,该算法以协同过滤推荐算法为基础,以快速推荐,效果良好为目标。本文的研究旨在抛砖引玉,以激发更多面向农产品的推荐算法的进一步、更深入的研究,本文主要工作及创新点如下:(1)选取面向农产品的推荐算法的基础算法。对农产品电子交易区别于其他商品交易的特点进行梳理,同时对协同过滤推荐算法进行总结和分类,并指出算法面临的问题。综合农产品线上交易的特点和各种协同过滤算法的特点,选取基于项目的协同过滤算法作为本文研究的面向农产品的推荐算法的基础算法。(2)针对冷启动问题,对算法做出改进。传统的协同过滤算法在冷启动情况下,推荐效果不佳,针对这一问题,本文提出IPSS项目相似性度量方法。此方法有两个主要部分:评分相似度和结构相似度,其中,评分相似度主要考虑两个项目评分之间的评分差、项目评分与评分中值之差,以及项目评分与其他评分平均值之差;结构相似度部分定义了共同评分项目占所有项目比重并惩罚活跃用户的逆项目频率(Inverse Item Frequence,IIF)系数。(3)针对可扩展性问题,对算法做出改进。传统的协同过滤推荐算法在海量数据情况下,会受到可扩展性问题的影响,针对这一问题,本文提出融合项目谱聚类的协同过滤算法。算法分为离线和在线两个步骤:离线时对项目聚类;在线时,首先确定目标项目从属的类别,然后在类中搜索近邻,最后求未评分项目的预测评分并据此为目标用户产生推荐。(4)对算法在真实数据上进行实验。简要介绍推荐引擎整体框架,并根据现有数据和算法使用MATLAB2009b在Movie Lens 100k和Grecs两个数据集对优化算法进行实验。首先需要按提示输入需要为其进行推荐的目标用户的编号,在手动输入编号后,程序会自动计算未评分项目的预测评分并根据计算结果产生推荐项目列表,展示给用户。本文提出的面向农产品的协同过滤推荐算法(记作SC-ICF)能够有效实现农产品推荐功能,此外,在冷启动、大规模数据集情况下,推荐效果优于传统算法。
[Abstract]:In recent years, electronic commerce has the advantages of convenient information communication and flexible and convenient receipt and payment, which promotes the development of social information. However, with the development of the Internet, many users are deeply troubled by "information overload" and affect the shopping experience. Therefore, it can help consumers to excavate the recommended system of the goods they are interested in from the mass data. By analyzing the personal characteristics and behavioral characteristics of consumers, the system can determine the interests of the users and recommend them to the users. For the operators, they can improve the volume of transactions to achieve the goal of accurate marketing; for the consumers, they can save time and find the goods they wish to buy. The electric business of agricultural products is the whole business. An important part of the platform is that the agricultural product recommendation system is of great significance to increase the income of farmers, improve the competitiveness of the first industry, and promote the development of agricultural information. However, there are few studies on the personalized recommendation system of agricultural products at the present stage. The research aim of this paper is to design a recommendation algorithm for agricultural products. The algorithm is based on collaborative filtering recommendation algorithm, with the aim of quick recommendation and good effect. This study aims to stimulate more agricultural products and further research. The main work and innovation of this paper are as follows: (1) the basic algorithm of the recommendation algorithm for agricultural products. The electronic transaction is distinguished from the characteristics of other commodity trading, at the same time, it summarizes and classifies the collaborative filtering recommendation algorithm, and points out the problems faced by the algorithm. The characteristics of the integrated agricultural products online transaction and the characteristics of various collaborative filtering algorithms are selected. The collaborative filtering algorithm based on the project is selected as the research oriented agricultural product in this paper. The basic algorithm of the recommendation algorithm. (2) to improve the algorithm for cold start problem. The traditional collaborative filtering algorithm has poor results in the cold start situation. In this paper, the IPSS project similarity measurement method is proposed in this paper. This method has two main parts: evaluation phase similarity and structural similarity, among which, score similarity is the main method. Considering the difference between the scores of two projects, the difference between the score and the median of the score, and the difference between the project score and the average value of the other scores; the structural similarity part defines the proportion of the common scoring items and the inverse project frequency (Inverse Item Frequence, IIF) coefficient of the active users. (3) for the scalability problem, To improve the algorithm, the traditional collaborative filtering recommendation algorithm will be affected by the scalability problem in the case of massive data. Aiming at this problem, this paper proposes a collaborative filtering algorithm, which is divided into two steps: offline and online: off-line clustering; when online, the target item is subordinate first. Category, and then search the nearest neighbor in the class, and finally seek the prediction score of the non scoring project and propose a recommendation for the target user. (4) the algorithm is experimentation on the real data. A brief introduction of the overall framework of the recommendation engine is introduced, and the optimization algorithm is used by the MATLAB2009b in the Movie Lens 100k and Grecs data sets according to the existing data and algorithms. Experiment. First, you need to enter the number of the target users that need to be recommended according to the prompt. After manual input number, the program will automatically calculate the prediction score of the non rated items and produce a list of recommended items based on the results of the calculation. The proposed cooperative filtering recommendation algorithm for agricultural products (SC-ICF) can be used in this paper. It is effective to realize the recommendation function of agricultural products. Moreover, in the case of cold start and large scale data sets, the recommendation effect is better than the traditional algorithm.

【学位授予单位】：东北农业大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3;F323.7;F724.6

【参考文献】