Knowledge Graph Dataset used in DecKG

1. Research Hypothesis: The proposed method DecKG aims to improve location recommendation systems by integrating users' check-in behavior (user-POI interactions) with structured urban knowledge graphs (KGs). The hypothesis is that leveraging contextual information from KGs—such as POI categories, geographic proximity, and brand affiliations—will enhance the accuracy and relevance of personalized POI recommendations.

2. Data Description:
2.1 Datasets: Two real-world datasets, Beijing and Shanghai [1], are used. Each contains check-in histories of 10,000 users and a corresponding knowledge graph.
2.2 User-POI Interactions: These represent users' historical check-ins at Points of Interest (POIs), capturing preferences and behavior.
2.3 Knowledge Graph (KG): The KG is structured as triplets (head entity, relation, tail entity). Entities include POIs (e.g., a restaurant), their categories (e.g., "Chinese cuisine"), geographic segments (e.g., a district), and attributes like brand. Relations define contextual links between entities, such as "belongs to brand X" or "is near [region]."
2.4 Data Split: Each dataset is divided into training (80%), testing (10%), and validation (10%) sets to evaluate model performance.

3. Data Collection:
3.1 User check-in data was sourced from location-based services, reflecting real-world visitation patterns as introduced in [1].
3.2 The KG was constructed using POI metadata (e.g., categories, brands) and spatial relationships (e.g., proximity between regions) as introduced in [1].
3.3 The datasets were preprocessed to select 10,000 active users per city, ensuring balanced representation.

[1] Liu, C., Gao, C., Jin, D., Li, Y., 2021. Improving location recommendation with urban knowledge graph. arXiv preprint arXiv:2111.01013.

1. 研究假设：所提出的DecKG方法旨在通过将用户签到行为（用户-POI交互）与结构化城市知识图谱（KGs）相结合，优化位置推荐系统。本研究假设，利用知识图谱中的上下文信息——如POI类别、地理邻近性与品牌归属——可提升个性化POI推荐的准确性与相关性。
2. 数据说明：
2.1 数据集：本研究采用两个真实世界数据集[1]，分别为北京数据集与上海数据集。每个数据集包含10000名用户的签到历史，以及与之对应的知识图谱。
2.2 用户-POI交互：该数据表征用户在兴趣点（POI，Point of Interest）的历史签到记录，可捕捉用户偏好与行为模式。
2.3 知识图谱（KG）：该知识图谱以三元组（头实体、关系、尾实体）的结构形式组织。实体涵盖POI（如餐厅）、其类别（如“中餐”）、地理分区（如行政区）以及品牌等属性；关系则定义实体间的上下文关联，例如“隶属于品牌X”或“邻近[某区域]”。
2.4 数据划分：每个数据集均被划分为训练集（80%）、测试集（10%）与验证集（10%），用于模型性能评估。
3. 数据采集：
3.1 用户签到数据源自位置服务，反映了真实世界的到访模式，相关细节参见文献[1]。
3.2 知识图谱的构建基于POI元数据（如类别、品牌）与空间关联关系（如区域间邻近性），相关细节参见文献[1]。
3.3 数据集经过预处理，筛选出每个城市的10000名活跃用户，以保证数据分布的均衡性。
[1] Liu, C., Gao, C., Jin, D., Li, Y., 2021. Improving location recommendation with urban knowledge graph. arXiv preprint arXiv:2111.01013.