The Yelp Collaborative Knowledge Graph

This is the The Yelp Collaborative Knowledge Graph (YCKG) - a transformation of the Yelp Open Dataset into RDF format using Y2KG. <strong>Paper Abstract</strong> The Yelp Open Dataset (YOD) contains data about businesses, reviews, and users from the Yelp website and is available for research purposes. This dataset has been widely used to develop and test Recommender Systems (RS), especially those using Knowledge Graphs (KGs), e.g., integrating taxonomies, product categories, business locations, and social network information. Unfortunately, researchers applied naive or wrong mappings while converting YOD in KGs, consequently obtaining unrealistic results. Among the various issues, the conversion processes usually do not follow state-of-the-art methodologies, fail to properly link to other KGs and reuse existing vocabularies. In this work, we overcome these issues by introducing Y2KG, a utility to convert the Yelp dataset into a KG. Y2KG consists of two components. The first is a dataset including (1) a vocabulary that extends Schema.org with properties to describe the concepts in YOD and (2) mappings between the Yelp entities and Wikidata. The second component is a set of scripts to transform YOD in RDF and obtain the Yelp Collaborative Knowledge Graph (YCKG). The design of Y2KG was driven by 16 core competency questions. YCKG includes 150k businesses and 16.9M reviews from 1.9M distinct real users, resulting in over 244 million triples (with 144 distinct predicates) for about 72 million resources, with an average in-degree and out-degree of 3.3 and 12.2, respectively. <strong>Links</strong> Latest GitHub release: https://github.com/MadsCorfixen/The-Yelp-Collaborative-Knowledge-Graph/releases/latest PURL domain: https://purl.archive.org/domain/yckg <strong>Files</strong> Graph Data Triple Files One sample file for each of the Yelp domains (Businesses, Users, Reviews, Tips and Checkins), each containing 20 entities. <code>yelp_schema_mappings.nt.gz</code> containing the mappings from Yelp categories to Schema things. <code>schema_hierarchy.nt.gz</code> containing the full hierarchy of the mapped Schema things. <code>yelp_wiki_mappings.nt.gz</code> containing the mappings from Yelp categories to Wikidata entities. <code>wikidata_location_mappings.nt.gz</code> containing the mappings from Yelp locations to Wikidata entities. Graph Metadata Triple Files <code>yelp_categories.ttl</code> contains metadata for all Yelp categories. <code>yelp_entities.ttl</code> contains metadata regarding the dataset <code>yelp_vocabulary.ttl</code> contains metadata on the created Yelp vocabulary and properties. Utility Files <code>yelp_category_schema_mappings.csv</code>. This file contains the 310 mappings from Yelp categories to Schema types. These mappings have been manually verified to be correct. <code>yelp_predicate_schema_mappings.csv</code>. This file contains the 14 mappings from Yelp attributes to Schema properties. These mappings are manually found. <code>ground_truth_yelp_category_schema_mappings.csv</code>. This file contains the ground truth, based on 200 manually verified mappings from Yelp categories to Schema things. The ground truth mappings were used to calculate precision and recall for the semantic mappings. <code>manually_split_categories.csv</code>. This file contains all Yelp categories containing either a &amp; or /, and their manually split versions. The split versions have been used in the semantic mappings to Schema things.

本数据集为Yelp协同知识图谱（Yelp Collaborative Knowledge Graph, YCKG）—— 即通过Y2KG工具将Yelp开放数据集转换为资源描述框架（Resource Description Framework, RDF）格式的产物。

## 论文摘要
Yelp开放数据集（Yelp Open Dataset, YOD）收录了来自Yelp平台的商家、评论与用户相关数据，仅面向研究用途开放。该数据集已被广泛应用于开发与测试推荐系统（Recommender Systems, RS），尤其是结合知识图谱（Knowledge Graphs, KGs）的推荐系统，例如整合分类体系、商品类别、商家地理位置与社交网络信息的推荐系统。

遗憾的是，既往研究者在将YOD转换为知识图谱的过程中，多采用粗糙或错误的映射方案，最终得到了与实际不符的结果。其中典型问题包括：转换流程未遵循前沿研究方法，未能正确关联至外部知识图谱并复用现有词汇表。

本研究针对此类缺陷，提出了Y2KG工具，用于将Yelp数据集转换为标准化知识图谱。Y2KG包含两大核心组件：其一为附属数据集，包含（1）一套扩展自Schema.org的词汇表，用于精准描述YOD中的各类概念；（2）Yelp实体与维基数据（Wikidata）之间的映射关系。其二为一组转换脚本，用于将YOD转换为RDF格式并生成Yelp协同知识图谱（YCKG）。

Y2KG的设计以16个核心能力问题为导向。YCKG共收录15万家商家、1690万条评论，数据来自190万独立真实用户，整体包含约7200万条资源，累计超过2.44亿条三元组（涉及144种不同谓词），其平均入度与出度分别为3.3与12.2。

## 链接
最新GitHub发布版：https://github.com/MadsCorfixen/The-Yelp-Collaborative-Knowledge-Graph/releases/latest
PURL域名：https://purl.archive.org/domain/yckg

## 文件
### 图数据三元组文件
针对Yelp各业务域（商家、用户、评论、用户提示与商家签到）各提供一个样本文件，每个文件包含20个实体。
<code>yelp_schema_mappings.nt.gz</code> 包含Yelp类别至Schema类别的映射关系。
<code>schema_hierarchy.nt.gz</code> 包含已映射Schema类别的完整层级结构。
<code>yelp_wiki_mappings.nt.gz</code> 包含Yelp类别至维基数据实体的映射关系。
<code>wikidata_location_mappings.nt.gz</code> 包含Yelp地理位置至维基数据实体的映射关系。

### 图元数据三元组文件
<code>yelp_categories.ttl</code> 收录所有Yelp类别的元数据。
<code>yelp_entities.ttl</code> 收录该数据集的相关元数据。
<code>yelp_vocabulary.ttl</code> 收录所构建的Yelp词汇表与属性的元数据。

### 工具文件
<code>yelp_category_schema_mappings.csv</code> 包含310条Yelp类别至Schema类别的映射关系，所有映射均经过人工验证确保准确无误。
<code>yelp_predicate_schema_mappings.csv</code> 包含14条Yelp属性至Schema属性的映射关系，均为人工梳理所得。
<code>ground_truth_yelp_category_schema_mappings.csv</code> 包含基于200条人工验证的Yelp类别至Schema类别的映射关系所构建的基准真值（ground truth），该基准真值曾用于计算语义映射的精确率与召回率。
<code>manually_split_categories.csv</code> 收录所有包含&或/的Yelp类别及其人工拆分后的版本，拆分后的类别已被用于语义映射至Schema类别。