基于图神经网络和BiGRU的商品推荐模型

Commodity Recommendation Model Combining GNN and BiGRU

个性化推荐是互联网经济的核心竞争力。为了解决推荐系统中数据稀疏性问题,提出基于图神经网络和BiGRU的商品推荐模型。该模型先利用来自变换器的双向编码器表征量(Bidirectional Encoder Representations from Transformers,BERT)模型进行预训练,再结合BiGRU与注意力机制提取评论文本的特征,并利用图神经网络提取用户与商品的高阶交互关系,最后将两种特征向量进行拼接以实现推荐预测。在多个亚马逊公开数据集上进行实验,使用均方误差(Mean Square Error,MSE)作为评价指标。实验结果表明,与已有的优秀基准模型相比,该模型有效提高了预测精度。

Personalized recommendation is the core competitiveness of the internet economy. In order to alleviate the problem of data sparsity in recommender system,this paper proposes a commodity recommendation model based on graph neural network and BiGRU. The model is pre-trained with the Bidirectional Encoder Representations from Transformers(BERT) model, then combing the BiGRU and the attention mechanism to extract the comment text, and uses graph neural network to extract the high-order interactive relationship between users and commodities. Finally, the two feature vectors are fused to achieve recommendation prediction. Experiments were conducted on several Amazon public data sets, Mean Square Error(MSE) used as the evaluation index. The experimental results show that model effectively improves the prediction accuracy compared to the existing excellent benchmark models.