基于全局有向图的商品会话序列推荐算法

针对商品会话序列推荐中传统推荐算法过分依赖临近点击状况, 在一定程度上丢失整体商品访问趋势的问题, 提出一种新的基于全局有向图的商品会话序列推荐算法. 首先, 构建商品会话序列全局有向图, 图中节点为商品, 节点间的弧表示点击次序, 并用图数据库存储该有向图; 其次, 给出在有向图上的全局偏好传播策略, 同时考虑点击时间因素对推荐的重要影响; 最后, 获得待推荐商品的评分. 在Diginetica和Yoochoose标准数据集上, 该算法根据P@20标准, 比传统Item-KNN方法推荐准确率分别提升了6.12%和30.25%; 根据MRR@20标准, 则分别提升了15.04%和33.88%. 实验结果表明, 该全局有向图搜索和评分策略有效.

Recommendation Algorithm of Commodity Session Sequence Based on Global Directed Graph

Aiming at the problem that the traditional recommendation algorithm in commodity session sequence recommendation relied too much on adjacent clicks and lost the overall commodity access trend to a certain extent, we proposed a new commodity session sequence recommendation algorithm based on global directed graph. Firstly, the global directed graph of commodity conversation sequence was constructed. The nodes in the graph were commodities, the arcs between nodes represented the click order, and the directed graph was stored in the graph database. Secondly, the global preference propagation strategy on the directed graph was given, and the important influence of click time on recommendation was considered. Finally, the score of the commodity to be recommended was obtained. On Diginetica and Yoochoose standard data sets, the recommended accuracy of the algorithm was improved by 6.12% and 30.25% respectively compared with the traditional Item-KNN method according to the P@20 standard, according to the MRR@20 standard, the recommended accuracy was improved by 15.04% and 33.88% respectively. The experimental results show that the proposed global directed graph search and scoring strategy is effective.