融合用户特征分类和协同过滤的餐饮位置服务推荐

发布时间：2018-09-10 11:52

【摘要】：随着推荐系统越来越追求个性化,传统的协同过滤推荐将物品的评分作为用户喜好的依据进行推荐,这种方式未能充分考虑用户自身的特征,对用户兴趣度量不准确,因此不能完全满足用户的个性化服务需求。自古以来,中国的饮食文化博大精深,而且具有很大的地域差异,每个人的饮食口味会受到出身地或者生活环境的影响,对餐厅的选择不仅局限于自身的消费能力,而是对餐厅的口味、环境、服务水平等都提出更高层次的要求。本文结合用户的特征,利用贝叶斯分类器构建了用户饮食兴趣模型,根据用户的饮食口味对周边范围内的餐厅进行协同过滤推荐。文中以朴素贝叶斯统计学方法论和基于物品的协同过滤为推荐的算法基础,在训练集上生成用户餐饮推荐模型,最后利用测试集进行了验证和评价。实验结果表明,与传统基于物品的协同过滤算法相比,本文提出的方法在推荐的准确度、冷启动和算法效率上有较大改进和提升。本文主要从以下几个方面展开了研究工作:(1)综合分析了个性化推荐的应用现状及其存在的问题,结合餐饮位置服务的推荐,明确提出了本文研究的内容和目标。(2)深入研究了协同过滤算法的分类及原理,对比分析现有算法的不足并总结了未来的研究趋势。(3)将贝叶斯决策论引入推荐算法中,以互信息和信息熵作为特征间影响因子,在朴素贝叶斯分类上进行改进,建立用户饮食兴趣模型。(4)结合大众点评网的评论数据进行实验设计。在用户饮食偏好模型的基础上,选取用户周边范围内符合用户口味的餐厅评分数据,训练生成饮食推荐系统,以评分较高的前Top-N个餐厅为推荐结果,将它们的位置信息推送给用户。(5)利用推荐系统的评价指标对实验结果的准确率、分类准确性以及算法的执行效率进行评价,分析其是否符合预期的目标以及存在的不足之处。
[Abstract]:With the increasing pursuit of individualization in recommendation systems, the traditional collaborative filtering recommendations recommend items as the basis of user preferences. This method fails to fully consider the characteristics of users, and the measurement of interest to users is not accurate. Therefore, it can not fully meet the user's personalized service needs. Since ancient times, Chinese food culture has been extensive and profound, and there are great regional differences. Everyone's diet taste will be affected by their origin or living environment. The choice of restaurants is not limited to their own consumption ability. But to the restaurant's taste, environment, service level and so on puts forward the higher level request. According to the characteristics of the users, this paper constructs a model of users' food interest by using Bayesian classifier, and makes collaborative filtering and recommendation according to the food taste of the users. Based on the naive Bayesian statistical methodology and the collaborative filtering based on articles, this paper generates the user dining recommendation model on the training set, and finally uses the test set to verify and evaluate it. The experimental results show that compared with the traditional collaborative filtering algorithm based on articles, the proposed method improves the accuracy, cold start and efficiency of the proposed algorithm. This paper mainly from the following several aspects of the research work: (1) the comprehensive analysis of the status quo of personalized recommendation and its existing problems, combined with the recommendation of catering location service, The contents and objectives of this paper are clearly put forward. (2) the classification and principle of collaborative filtering algorithms are deeply studied, the shortcomings of existing algorithms are compared and analyzed, and the future research trends are summarized. (3) Bayesian decision theory is introduced into recommendation algorithms. Taking mutual information and information entropy as the influencing factors between features, we improve the classification of naive Bayes and establish the model of users' dietary interest. (4) based on the comment data of Dianping net, the experimental design is carried out. On the basis of the food preference model, the food recommendation system was generated by selecting the restaurant rating data in the surrounding area of the user, and taking the former Top-N restaurants with higher score as the recommended results. Their location information is pushed to the user. (5) the accuracy of the experimental results, the accuracy of classification and the efficiency of the algorithm are evaluated by using the evaluation index of the recommendation system.
【学位授予单位】：武汉大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】