基于图书语义特征的推荐模型

Recommendation Model Based on Book Semantic Features

为了充分利用图书丰富的文本信息,更准确地表达图书属性,为读者提供更精准的图书推荐服务,提出了一种基于图书语义特征的深度学习推荐模型。该模型将预训练模型(BERT)与文本卷积神经网络(TextCNN)相结合提取图书语义特征。首先利用BERT网络对图书书名、内容摘要等图书文本信息生成向量表示;然后将获得的字向量通过TextCNN模型抽取文本局部特征,再与句向量一起输入神经网络进行训练,得到图书向量;最后将提取的图书特征与读者年龄、性别、专业等人口属性特征拼接后输入多层神经网络进行模型训练,获得预测结果。实验结果表明:所提出的模型对比其他模型推荐效果有较大提升。

To provide readers with more accurate book recommendation services,the paper proposes a recommendation model based on book semantic features(deep semantics mining for book recommendation)for making full use of the rich text information of books to express the book attributes more accurately.This model combines bidirectional encoder representations from transformers(BERT)networks with text convolutional neural networks(TextCNN)to extract the book semantic features.Firstly,we used the BERT networks to generate the vector representation of the book text information such as book title and abstract.Then,the local features of the obtained word vectors were extracted through the TextCNN model,and were inputted together with the obtained sentence vectors into the neural networks for training to get the final book vectors.Finally,the book features and reader demographic features such as age,gender and specialty of readers were inputted into the neural networks for training to obtain expected results.Results show that the model proposed in this paper has significant improvement compared with other models.