基于用户需求深度驱动的个性化推荐算法研究

发布时间：2018-12-25 15:08

【摘要】：随着互联网时代尤其是移动互联网时代的到来,当代的我们都处在一个“信息爆炸”的时代,由于个性化需求的增强,导致用户需要个人去过滤掉大量的无效信息,无形中仍然没有根本性的解决信息量过多的难题。于是,个性化推荐算法出现并成为了大家关注的热门话题。个性化推荐系统自出现以来就获得了广泛的关注,众多的专家学者都提出了各自关于个性化推荐系统的研究方法。当前主流的推荐算法主要是基于内容的推荐算法、基于图结构的推荐算法、协同过来推荐算法以及混合推荐算法。但是当前的推荐算法由于过分依赖数据的显性评分而存在冷启动、数据稀疏性以及推荐滞后等影响推荐精度的问题。本文对个性化推荐算法的优化问题进行了研究,重点对如何充分利用用户的隐性行为和行业领域知识为用户进行更加精准的个性化推荐进行了深入研究。提出了一种基于用户需求深度驱动的个性化推荐算法。算法主要针对当前推荐算法存在的冷启动、数据稀疏性、推荐滞后等问题提出了自己的改进方案,在进行用户聚类时加入了用户的隐性行为分析,综合利用用户隐性行为信息和用户的属性信息来实现为用户进行聚类。同时在为用户生成推荐列表时,加入了行业的领域知识,根据大数据来生成行业链,从相似产品的横向推荐和关联产品的纵向推荐并行推荐实现对用户进行引导消费,帮助用户明确潜在需求,产生可观的经济效益和社会效益。最后将推荐系统设计为闭环控制系统,由于需求会经常发变化,因此生成推荐列表后系统会在特定时窗检测推荐精度,可以及时检测推荐的精确性便于及时调整推荐列表。本文在进行算法的实验时采用的是淘宝网举办的天池大赛的数据,通过数据集进行了算法的对比验证,将本文提出的算法与先前的用户聚类算法与二部图推荐算法进行了精确性的比对,实验证明,本文提出的算法在聚类精度和推荐精度方面均有明显的提升。并通过推荐系统的闭环设计,有效的保证了推荐精度的稳定性,避免了推荐滞后。
[Abstract]:With the advent of the Internet era, especially the mobile Internet era, we are all in the era of "information explosion". Because of the enhancement of individualized demand, users need individuals to filter out a large number of invalid information. There is still no fundamental solution to the problem of too much information. As a result, personalized recommendation algorithm appeared and became a hot topic. Since the emergence of personalized recommendation system, many experts and scholars have proposed their own research methods of personalized recommendation system. The current mainstream recommendation algorithms are mainly content-based recommendation algorithms, graph structure-based recommendation algorithms, collaborative recommendation algorithms and hybrid recommendation algorithms. However, the current recommendation algorithms have some problems such as cold start, data sparsity and recommendation lag, which affect the recommendation accuracy due to over-reliance on the dominant score of the data. In this paper, the optimization of personalized recommendation algorithm is studied, with emphasis on how to make full use of the implicit behavior of users and industry domain knowledge for users to carry out more accurate personalized recommendation in-depth research. In this paper, a personalized recommendation algorithm based on user's demand depth is proposed. Aiming at the problems of cold start, data sparsity, recommendation lag and so on, the algorithm puts forward its own improvement scheme, and adds the hidden behavior analysis of users in the process of user clustering. The user's hidden behavior information and user's attribute information are used to cluster the user. At the same time, when generating the recommendation list for users, we add the domain knowledge of the industry, according to big data to generate the industry chain, from the horizontal recommendation of similar products and vertical recommendation of related products, we can guide the consumption of users. Help users to identify potential needs, generating considerable economic and social benefits. Finally, the recommendation system is designed as a closed-loop control system. Because the requirements will often change, the system will detect the recommendation accuracy in a specific time window after generating the recommendation list, which can timely detect the recommendation accuracy and facilitate the timely adjustment of recommendation list. This paper uses the data of the Tianchi contest held by Taobao in the experiment of the algorithm, and carries on the comparison and verification of the algorithm through the data set. The proposed algorithm is compared with the previous user clustering algorithm and bipartite graph recommendation algorithm. The experimental results show that the proposed algorithm improves the clustering accuracy and recommendation accuracy obviously. Through the closed-loop design of recommendation system, the stability of recommendation accuracy is effectively guaranteed and the recommendation lag is avoided.
【学位授予单位】：山东师范大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F274

【参考文献】