基于形式概念分析的推荐算法研究及应用

发布时间：2018-05-05 09:21

本文选题：推荐算法 + 协同过滤　；参考：《郑州大学》2017年硕士论文

【摘要】：做为处理信息过载的有效手段,推荐系统在近些年得到了广泛的研究与发展,推荐系统在各领域应用的成功案例也不断涌现,但是依然面临着很多问题亟待解决。形式概念分析(Formal Concept Analysis,FCA)的核心数据结构—概念格(Concept lattices),是一种数据分析与规则提取的有效工具。外延与内涵做为概念的组成部分使得形式概念展现出了聚类的特性。概念之间存在的偏序关系也揭示了其泛化与特化的本质。随着其研究的不断深入,形式概念分析开始逐步应用于数据挖掘、信息检索等领域。协同过滤(Collaborative Filtering,CF)推荐作为应用最广的推荐策略之一,其中经典的基于邻域的协同过滤算法通常只考虑用户间或项目间的相似关系,而忽略了不同对象之间的内在联系。此外越来越多的研究人员发现,推荐系统往往面对的是无法直观反映用户喜好程度的隐式数据,并且随着产品种类的剧增,用户与项目间产生的隐式数据也会变得极为稀疏。所以由于稀疏数据环境下信息的缺失,协同过滤算法往往获取不到充足的邻域信息,从而直接影响了最终的推荐效果。针对以上问题,本文提出了一种面向隐式数据的基于概念邻域的协同过滤推荐算法(Conceptual Neighborhood-based Collaborative Filtering,CNCF)。该算法针对Top-N推荐问题,以概念格为载体进行推荐问题求解。首先在用户与项目的关系数据转化而成形式背景的基础上进行概念格的构造,将用户与产品分别以对象与属性的形式聚集在概念中,并基于概念格生成相应的起始概念索引,借助索引结构高效地对对象的起始概念进行定位。之后利用概念之间的偏序关系,以对象(用户)的起始概念为起点探索其近邻概念并获取候选项集。最后结合所提出的全局偏好度与邻域偏好度筛选出用户可能感兴趣的推荐列表。通过对CNCF算法的实现,并在两个真实数据集上进行实验验证,相较于传统基于邻域的协同过滤推荐算法,CNCF算法在可以在保持较好的推荐效果同时,更适用于数据稀疏环境下的推荐。
[Abstract]:As an effective means to deal with information overload, recommendation system has been widely studied and developed in recent years. The successful cases of recommendation system in various fields are emerging, but there are still many problems to be solved. Formal Concept Analysis (FCA) is an effective tool for data analysis and rule extraction. Extension and connotation as part of the concept make the formal concept show the characteristics of clustering. The existence of partial ordering between concepts also reveals the essence of generalization and specialization. With the deepening of its research, formal concept analysis has been gradually applied to data mining, information retrieval and other fields. Collaborative filtering (CF) recommendation is one of the most widely used recommendation strategies, in which the classical neighborhood based collaborative filtering algorithms usually only consider the similarity between users or items, but ignore the internal relations between different objects. In addition, more and more researchers find that recommendation systems often face implicit data that can not directly reflect the degree of user preferences, and with the sharp increase in product types, the implicit data generated between users and projects will become extremely sparse. Therefore, because of the lack of information in sparse data environment, collaborative filtering algorithm often can not obtain sufficient neighborhood information, which directly affects the final recommendation effect. To solve the above problems, this paper proposes a Conceptual Neighborhood-based Collaborative filtering algorithm based on Conceptual Neighborhood-based Collaborative filtering for implicit data. The algorithm solves the Top-N recommendation problem with concept lattice as the carrier. Firstly, the concept lattice is constructed on the basis of transforming the relational data between the user and the project into a formal background. The user and the product are gathered in the concept in the form of objects and attributes respectively, and the corresponding initial concept index is generated based on the concept lattice. With the help of index structure, the starting concept of object is located efficiently. Then, by using the partial order relation between the concepts, starting from the initial concept of object (user), we explore the concept of nearest neighbor and obtain the set of candidate items. Finally, the proposed global preference degree and neighborhood preference degree are combined to filter out the list of recommendations that the user may be interested in. Through the implementation of CNCF algorithm and experimental verification on two real data sets, compared with the traditional collaborative filtering recommendation algorithm based on neighborhood, it can keep a good recommendation effect at the same time. More suitable for data sparse environment recommendation.
【学位授予单位】：郑州大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】