基于物品嵌入向量的会话型推荐算法  

Session-Based Recommendation Algorithm with Item2Vec

作　　者：陈恩华[1] 方宝富[1] CHEN Enhua;FANG Baofu(School of Computer and Information,Hefei University of Technology,Hefei 230601,China)

机构地区：[1]合肥工业大学计算机与信息学院,合肥230601

出　　处：《计算机工程》2021年第7期74-80,共7页Computer Engineering

基　　金：安徽省自然科学基金(1708085MF146)。

摘　　要：传统基于会话的推荐算法主要利用点击物品的时序信息进行建模,忽略了挖掘物品的特征信息,且未利用物品之间的相似性。为提升推荐效果,提出一种新的基于会话的推荐算法SR-I2V。通过Skip-gram模型和层次softmax优化方法学习物品的嵌入向量,由意图递进公式对已发生的物品点击提取出意图特征向量,并根据特征向量相似度计算出每个候选项的推荐分数。实验结果表明,与I2I、Po P和S-POP等传统基于会话的推荐算法相比,该算法在Yoochoose和Diginetica两个数据集上的推荐召回率分别提高了至少4.67个百分点和3.97个百分点,平均倒数排名指标也有相应提高。The existing session-based recommendation algorithms mainly leverage the temporal information of the items while ignoring their feature information and the similarity between the items.In order to improve the recommendation effect,this paper proposes a novel session-based recommendation algorithm named Session-based Recommendation with Item2Vec(SR-I2V).The algorithm learns the embedding vectors of the items through the skipgram model and the hierarchical softmax optimization method.Then it uses the intent progression formula to extract the intent feature vectors from the click events that have occurred,and calculates the recommendation score of each candidate item based on the similarity between the feature vectors.Experimental results show that compared with traditional session-based recommendation algorithms such as I2I,PoP,and S-POP,the proposed algorithm increases the recommendation recall rate by at least 4.67 percentage points on Yoochoose and 3.97 percentage points on Diginetica.In addition,it improves the mean reciprocal rank.

关 键 词：推荐算法 循环神经网络 嵌入向量 层次softmax 意图递进 

分 类 号：TP391.3[自动化与计算机技术—计算机应用技术]