旅游知识图谱特征学习的景点推荐

基于知识图谱的推荐算法在多个领域取得了较好的效果,但仍然存在一些问题,如不能有效提取知识图谱中实体关系标签中的特征,推荐准确率会降低。因而提出将网络嵌入方法(network embedding)用于旅游知识图谱的特征提取,使得特征的提取更加充分。通过对旅游知识图谱中不同标签的属性子图独立建模,利用深度学习模型挖掘游客及景点等图节点语义特征,进而获得融合各个标签语义的游客和景点特征向量,最终通过计算游客和景点相关性生成景点推荐列表。通过在真实旅游知识图谱上的实验,验证了利用网络嵌入方法对知识图谱中数据建模后,可以有效提取节点的深层特征。

Tourism knowledge-graph feature learning for attraction recommendations

The recommendation algorithm based on knowledge graphs has achieved good results in several fields; however, it contains some problems as well. For example, this algorithm cannot effectively extract features from entity relationship tags in the knowledge graph, which reduces its recommendation accuracy. We propose a network embedding method that more fully extracts features from the tourism knowledge graph than the aforementioned method. By independently modeling different label-attribute subgraphs in the tourism knowledge graph and by using a deep learning model to mine node semantic features, such as tourists and scenic spots, we can obtain a feature vector of tourists and attractions that fuses the semantics of various tags. Finally, this method generates a recommended list of scenic spots by calculating the correlation between tourists and scenic spots. Experimental results on a real tourism knowledge graph verify that this network embedding method effectively extracts the deep features of knowledge graph nodes to model the data in the knowledge graph.