个性化推荐系统算法研究

发布时间：2019-03-15 08:17

【摘要】：计算机技术的发展使得我们的社会进入了信息时代,信息时代使我们的生活发生了翻天地覆的变化,我们随时随地都可以在互联网中查找自己需要的信息。信息时代在为我们带来方便的同时,也带来了一些问题,那就是所谓的“信息过载”问题。“信息过载”是信息化发展过程中所产生的负面影响之一,是指在信息化建设的过程中,由于网络中信息量的指数级增长,导致信息在网络中存在大量的冗余,无法被人们充分利用。为了解决这一问题,研究人员提出了很多方法,其中最具代表性的就是推荐系统了。推荐系统通过对用户的历史数据和行为信息进行科学的运算、处理、分析,建立用户的兴趣模型,并通过兴趣模型对用户推荐可能喜欢的内容。虽然推荐系统可以有效地解决“信息过载”,但也不可避免地面临很多问题(如冷启动,推荐精度和用户兴趣时变问题等)。因此,本文主要研究如何提高推荐系统的性能,解决推荐系统的冷启动和用户兴趣时变问题。针对时间对用户兴趣变化的影响,本文通过分析用户在网络活动中的整体行为对推荐系统的影响,提出了标签活跃周期的概念,标签活跃周期可以很好的反映用户的行为对推荐系统的影响。同时分析用户添加标签时间对整体推荐的影响,从而提出了标签时间加权因子。结合基于网络结构的推荐技术特点,应用时间加权因子对网络结构推荐算法进行改进,提出了一种新的基于时间权值的个性化推荐算法。并将该算法与一些经典的算法进行比较分析,结果显示该算法能够在Delicious和Movielens数据集得到令人满意的结果,有效的提高了推荐系统的精度、多样性。在进一步的实验中发现,基于时间权值的个性化推荐算法在两个数据集中,当资源对象的权值越小时,该算法表现越好,结果也证明了本文提出的算法可以很好地解决“冷启动”问题。
[Abstract]:With the development of computer technology, our society has entered the information age, and the information age has changed our lives all over the world. We can find the information we need in the Internet anytime and anywhere. The information age not only brings us convenience, but also brings some problems, that is, the so-called "information overload" problem. "Information overload" is one of the negative effects in the process of information development. It means that in the process of information construction, due to the exponential growth of information in the network, there is a large number of redundancy of information in the network. People can't make full use of it. To solve this problem, researchers have proposed many methods, the most representative of which is the recommendation system. The recommendation system carries on the scientific operation, the processing, the analysis, the establishment user's interest model through carries on the scientific operation to the user's historical data and the behavior information, and recommends to the user the content which the user may like through the interest model. Although the recommendation system can effectively solve the "information overload", it is inevitable to face many problems (such as cold start, recommendation accuracy and user interest time-varying problems, etc.). Therefore, this paper mainly studies how to improve the performance of recommendation system and solve the problem of cold start-up and time-varying interest of users. In view of the influence of time on the change of user's interest, this paper analyzes the influence of user's overall behavior in network activities on recommendation system, and puts forward the concept of label active cycle. Label active cycle can well reflect the impact of user behavior on the recommendation system. At the same time, the influence of user tagging time on the overall recommendation is analyzed, and then the label time weighting factor is proposed. Combined with the characteristics of recommendation technology based on network structure, a new personalized recommendation algorithm based on time weight is proposed by using time weighting factor to improve the network structure recommendation algorithm. The algorithm is compared with some classical algorithms, and the results show that the algorithm can get satisfactory results in Delicious and Movielens data sets, and improve the accuracy and diversity of the recommendation system effectively. In further experiments, it is found that the personalized recommendation algorithm based on time weights is in two data sets, and the smaller the weight of the resource object is, the better the performance of the algorithm is. The results also show that the algorithm proposed in this paper can solve the problem of "cold start" well.
【学位授予单位】：哈尔滨理工大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【参考文献】