基于时间段图模型的用户行为分段与兴趣建模研究

发布时间：2018-05-20 17:46

本文选题：推荐系统 + 时间段图模型　；参考：《浙江大学》2017年硕士论文

【摘要】：在当前社会,推荐系统已经取得了广泛的应用,从小型社区到大型电子商务网站,推荐系统无疑在扮演着十分重要的角色。如何计算用户对一个物品的感兴趣程度在推荐系统领域是一个至关重要的问题。对用户兴趣的预测越准确,推荐算法的准确度就越高。然而用户的兴趣是动态变化的,在不同的时间,用户的兴趣会有不同,这给推荐系统带来了极大的挑战。因此,在对用户的兴趣进行建模时,如何把时间因素考虑在内显得尤为重要。时间段图模型通过在用户物品二分图中增加一个新的结点类型,即时间段结点来引入时间信息。但是时间段图模型中采用了固定时间窗口进行时间段划分的方式,很难对所有用户都划分出在时间上比较聚集的行为记录,且时间段图模型没有考虑到兴趣会随着时间衰减。基于此,本文提出了基于贪心策略的时间段划分方法,并且提出了基于物品时间段二分图的兴趣衰减模型,本文的主要研究工作如下:1)研究了时间信息在图模型中的表示,提出并实现了一个基于贪心策略的时间段划分方法,相比于采用固定时间窗口进行时间段划分的方法,该方法能够更好地适应不同用户具有不同的行为密度的情况。2)研究了图模型中的用户兴趣建模与推荐,提出并实现了基于物品时间段二分图的兴趣衰减模型IDSSTG和GIDSSTG。相比于时间段图模型,该模型考虑到了兴趣随时间衰减的情况,且减少了图的结点数和边的个数,提高了推荐算法的时间效率。3)在前面两步的基础上,实现了物品时间段二分图上的兴趣衰减模型、基于用户的协同过滤和时间段图模型等多种推荐方法,并在公开的数据集MovieLens和豆瓣电影数据集上进行测试和比较,验证了本文所作工作的有效性。
[Abstract]:In the current society, recommendation system has been widely used, from small community to large e-commerce website, recommendation system undoubtedly plays a very important role. How to calculate a user's interest in an item is a critical issue in the field of recommendation systems. The more accurate the prediction of user interest, the higher the accuracy of recommendation algorithm. However, the user's interest is dynamic. At different times, the user's interest will be different, which brings great challenges to the recommendation system. Therefore, it is very important to take time into account when modeling user interest. Time graph model introduces time information by adding a new node type to the bipartite graph of user items, that is, time period node. But in the time graph model, the fixed time window is used to divide the time period, it is difficult to divide all the users into the behavior records which are relatively aggregated in time, and the time period graph model does not consider that interest will decay with time. Based on this, this paper proposes a time division method based on greedy strategy and an interest attenuation model based on the bipartite graph of item time. The main research work of this paper is as follows: 1) the representation of time information in the graph model is studied. A time division method based on greedy strategy is proposed and implemented, which is compared with the time division method based on fixed time window. This method can better adapt to different users with different behavior density. (2) this paper studies user interest modeling and recommendation in graph model, and proposes and implements interest attenuation model IDSSTG and GIDSSTG based on item time interval bipartite graph. Compared with the time period graph model, the model takes into account the decay of interest with time, and reduces the number of nodes and edges of the graph, and improves the time efficiency of the recommended algorithm. 3) on the basis of the first two steps, The interest attenuation model on the bipartite graph of item time is implemented, and several recommended methods based on the user's collaborative filtering and time graph model are implemented, and tested and compared on the published data set MovieLens and Douban movie data set. The validity of the work done in this paper is verified.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TP391.3

【相似文献】