基于游记文本内容的旅游场景知识图谱的构建

Construction of Tourism Scene Knowledge Graph Based on the Content of Travelogs

摘要: [目的/意义]：传统以旅游本体概念为基础的知识表示形式，侧重对静态特征的描述，缺乏对动态特征的捕捉，如若考虑融入在线UGC中的一些动态信息，如：旅游活动事件发生的顺承关系，群体旅游情绪等特征，可进一步增强旅游知识图谱的可解释性，从而为当今体验式旅游的智能应用提供更加具有可推理性的知识库。[方法/过程]：因此，本文提出一种基于贝叶斯网络的旅游场景知识图谱表示模型，该模型以旅游活动事件作为单位实体节点，从游记文本中获取人们的旅游情感信息扩充实体的动态属性，并以人们在目的地之间的转移概率和情感差值作为边的动态属性，丰富图谱中边的信息，从而提升知识图谱的可解释性。[结论]：最终实验表明，基于本文所构建的旅游场景知识图谱设计的LTTE推荐模型，在旅游目的地推荐实验中表现的效果明显优于基准算法TF-IDF。

Abstract: The traditional knowledge representation based on the concept of tourism ontology, which is focus-ing on the description of static features and lacks dynamic features. If some dynamic information hidden in the online UGC is fully utilized, such as the inheritance relationship of tourism events, group tourism Features such as emotions, can further enhance the interpretability of the tourism knowledge graph, thereby providing a more inferred knowledge graph for intelligent applications of experiential tourism. Therefore, this paper proposes a tourism scene knowledge graph (TSKG) representation model based on Bayesian network. The model uses tourism events as unit entity to obtain people’s travel emotional information from the travelogs. The sentiment difference and transition probability between destinations is used as the dynamic attribute of edges to enrich the information of edges in the graph. The experimental result shows that the LTTE recommendation model based on the TSKG performs significantly better than the benchmark algorithm TF-IDF in the tourism destination recommendation experiment.