上下文感知推荐技术研究

发布时间：2018-04-05 13:04

本文选题：上下文感知推荐　切入点：复杂分割　出处：《江西理工大学》2016年硕士论文

【摘要】：个性化推荐系统的目的是解决信息过载问题,目前已被广泛应用于互联网的各个领域。传统的推荐系统只通过分析用户-项目之间的二元关系来为用户提供推荐,而忽略了上下文信息对用户决策的影响。随着上下文感知技术以及智能移动终端技术的快速发展,将上下文感知技术融入推荐过程的上下文感知推荐系统研究愈演愈烈。该研究在信息检索、移动互联网、物联网、电子商务、智能家居/办公/交通等诸多工业领域具有广泛的应用前景。目前该领域的研究在上下文信息挖掘与检测、用户建模与行为分析、上下文用户偏好提取、上下文感知推荐算法等方面都存在许多问题亟待解决。为了进一步提高推荐准确度和效率,本文针对上下文建模方法、推荐生成方法等关键问题进行研究,主要取得了以下成果:(1)为了得到更加专一化的数据以进一步提高推荐结果的准确性,本文提出基于离散二进制粒子群算法的上下文复杂分割方法,将历史数据中处于不同上下文环境下的同一个用户(或项目)分割成两个不同的用户(或项目)。该方法主要过程为首先利用离散二进制粒子群算法对最佳分割上下文条件组合进行优化,然后根据最佳分割组合中的这些上下文条件对项目或用户进行分割,便能得到更加专一化的评分数据,最后将这些数据输入到推荐算法中获得更加准确的推荐结果。采用真实电影评分数据集进行实验,得出的结果验证了提出算法的有效性和可靠性。(2)针对现有相关研究存在同等对待所有上下文而忽略各上下文对用户评分影响力强弱的问题,本文提出基于贝叶斯方法与聚类的上下文用户兴趣建模方法。首先采用特征聚类方法对项目进行聚类,然后利用贝叶斯公式计算单个上下文条件下一个用户喜欢某类项目的概率,再通过复合概率公式求得多个上下文条件下用户喜欢一类项目的联合概率。最后根据喜欢同一类项目的用户之间相似度更高这一认识,将所求的联合概率融入到传统协同过滤算法中用户相似度计算过程以提高相似度精度。采用真实电影评分数据集进行对比实验,实验结果表明该方法与传统协同过滤方法相比能够有效利用上下文信息提高推荐准确度。
[Abstract]:The purpose of personalized recommendation system is to solve the problem of information overload, which has been widely used in various fields of the Internet.Traditional recommendation systems only analyze the binary relationship between users and items to provide recommendations for users, but ignore the impact of context information on user decisions.With the rapid development of context-aware technology and intelligent mobile terminal technology, the research of context-aware recommendation system which integrates context-aware technology into the recommendation process is becoming more and more serious.This research has a wide range of applications in information retrieval, mobile Internet, Internet of things, e-commerce, smart home / office / transportation and many other industrial fields.At present, there are many problems in this field, such as context information mining and detection, user modeling and behavior analysis, context user preference extraction, context-aware recommendation algorithm and so on.In order to further improve the accuracy and efficiency of recommendation, this paper focuses on some key problems, such as context modeling method, recommendation generation method and so on.In order to obtain more specific data to further improve the accuracy of the recommended results, this paper proposes a context complex segmentation method based on discrete binary particle swarm optimization (Dbinary Particle Swarm Optimization).The same user (or item) in a different context in the historical data is split into two different users (or items).The main process of this method is to first optimize the optimal segmentation context conditions by using discrete binary particle swarm optimization algorithm, and then segment the items or users according to these context conditions in the optimal segmentation combination.Then we can get more specialized scoring data and input these data into the recommendation algorithm to obtain more accurate recommendation results.The experimental results show that the proposed algorithm is effective and reliable. (2) there is a problem that all contexts are treated equally and the influence of each context is ignored.This paper presents a contextual user interest modeling method based on Bayesian method and clustering.Firstly, the feature clustering method is used to cluster the items, and then the Bayesian formula is used to calculate the probability of a user liking a certain type of item under a single context.Then the joint probability of the user like a class of items under multiple contexts is obtained by the compound probability formula.Finally, according to the higher similarity between users who like the same type of items, the proposed joint probability is integrated into the traditional collaborative filtering algorithm to calculate the user similarity to improve the accuracy of similarity.The experimental results show that the proposed method can effectively improve the accuracy of recommendation by using context information compared with the traditional collaborative filtering method.
【学位授予单位】：江西理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.3

【参考文献】