融合知识图谱与协同过滤的微地图推荐
牛雪磊(1994— ),男,内蒙古商都县人,博士生,主要从事深度学习、地理信息系统等研究。E-mail: NXLnxl186186@163.com |
收稿日期: 2022-08-09
修回日期: 2022-10-10
网络出版日期: 2024-05-11
基金资助
国家自然科学基金项目(42261067)
国家自然科学基金项目(61862039)
2021年度中央引导地方科技发展资金(2021-51)
兰州市人才创新创业项目(2020-RC-22)
兰州交通大学天佑创新团队(TY202002)
WeMap Recommendation by Fusion of Knowledge Graph and Collaborative Filtering
Received date: 2022-08-09
Revised date: 2022-10-10
Online published: 2024-05-11
Supported by
National Natural Science Foundation of China(42261067)
National Natural Science Foundation of China(61862039)
2021 Central Government Funds for Guiding Local Science and Technology Development(2021-51)
Talent Innovation and Entrepreneurship Project of Lanzhou City(2020-RC-22)
Tianyou Innovation Team of Lanzhou Jiaotong University(TY202002)
针对推荐系统中数据稀疏问题,传统的协同过滤算法无法捕捉辅助信息之间的相关性,从而降低了推荐的准确性。为此,本文提出融合知识图谱的协同过滤模型(Knowledge Graph Embedding Collaborative Filtering, KGCF),引入知识图谱作为辅助信息,利用知识图谱中多源结构性的数据来缓解数据稀疏问题。KGCF模型结合知识图谱的语义信息和协同过滤的偏好信息,能够挖掘出用户和微地图的隐语义交互信息,从而达到“千人千面”的推荐效果。① 融合知识图谱和协同过滤算法对微地图数据集进行采集训练; ② 通过皮尔逊相关系数计算出用户之间的相似矩阵,并对稀疏的评分矩阵进行隐语义矩阵分解,采用基准(Baseline)得到用户和微地图地名的偏好信息; ③ 通过知识图谱将微地图语义信息转化为低维向量,采用余弦相似度计算出微地图地名之间的相似矩阵; ④ 将用户和微地图地名结合为一个推荐结果集。通过在微地图数据集上实验,证明了本文提出的KGCF模型能有效解决数据稀疏,可准确为用户推荐感兴趣的微地图。
牛雪磊 , 杨军 , 闫浩文 . 融合知识图谱与协同过滤的微地图推荐[J]. 地球信息科学学报, 2024 , 26(4) : 967 -977 . DOI: 10.12082/dqxxkx.2024.220581
Based on sparse matrix, traditional collaborative filtering techniques usually have a low recommendation accuracy, since they cannot capture the correlations between auxiliary information from the sparse data. To fill the gap, this paper proposes a Knowledge Graph embedding Collaborative Filtering (KGCF) model to improve recommendation accuracy. In this model, the knowledge graph is introduced as auxiliary information, taking advantage of its multi-source structured data to alleviate the problem of data sparsity. By combining the semantic information of the knowledge graph and the preference information of collaborative filtering, the KGCF model can mine the interaction between users and WeMap to implement customized recommendations. Specifically, the knowledge graph and collaborative filtering algorithm are first combined to train the model on WeMap datasets. Secondly, the similarity matrix between users is calculated using the Pearson correlation coefficient, and the cryptic meaning matrix is decomposed through a sparse scoring matrix. In addition, the preference information of users and place names of WeMap is obtained using Baseline. Then, the semantic information of each object is transformed into a low dimension vector by the knowledge graph, and the similarity matrix between WeMap place names is calculated by cosine similarity. Finally, the users and place names of the WeMap are integrated into a recommendation result set. The experiments on WeMap datasets prove that the proposed KGCF model can effectively solve data sparsity and accurately recommend WeMaps of interest for users.
表1 WeMap数据集中获取的字段信息Tab. 1 Field information obtained from the WeMap data set |
字段id | 注释 | 含义 |
---|---|---|
Genes | 地图类型 | TourMap、SchoolMap、FoodMap、 DrivingMap |
Original_title | 地图名 | 仙居、云门山、张家界、五台山、惠山古镇 |
Event | 事件 | 对应的每个事件 |
表2 从WeMapRating数据集中获取的字段信息Tab. 2 The field information obtained from the WeMapRating dataset |
字段id | 注释 | 含义 |
---|---|---|
User_id | 评分用户id | 每条评分对应的评分用户id |
Map_id | 地图id | 每条评分对应的地图id |
Title | 地图名称 | 每条评分对应的地图名称 |
Ratings | 评分数 | 每条评分对应的评级分数 |
图3 4类专题微地图-评分力引导知识图谱Fig. 3 Knowledge graphs of force-directed relationship between four categories of thematic WeMaps and scores |
表3 所有地图主体关系Tab. 3 All map principal relationships |
实体名称 | 三元组内容 | 作用 |
---|---|---|
用户id(User_id) | 头实体 | 作为构建知识图谱的头实体 |
评分值(Rating)或事件(Event) | 关系 | 作为构建知识图谱的关系 |
微地图名称(cn_title) | 尾实体 | 作为知识图谱的尾实体 |
[1] |
|
[2] |
|
[3] |
|
[4] |
|
[5] |
|
[6] |
贾冲, 冯慧芳, 杨振娟. 基于出租车GPS轨迹和POI数据的商业选址推荐[J]. 计算机与现代化, 2020(2):21-25,30.
[
|
[7] |
闫浩文, 张黎明, 杜萍, 等. 自媒体时代的地图:微地图[J]. 测绘科学技术学报, 2016, 33(5):520-523.
[
|
[8] |
|
[9] |
|
[10] |
|
[11] |
|
[12] |
|
[13] |
|
[14] |
|
[15] |
|
[16] |
|
[17] |
|
[18] |
徐增林, 盛泳潘, 贺丽荣, 等. 知识图谱技术综述[J]. 电子科技大学学报, 2016, 45(4):589-606.
[
|
[19] |
蒋秉川, 万刚, 许剑, 等. 多源异构数据的大规模地理知识图谱构建[J]. 测绘学报, 2018, 47(8):1051-1061.
[
|
[20] |
|
[21] |
刘琼昕, 覃明帅. 基于知识表示学习的协同矩阵分解方法[J]. 北京理工大学学报, 2021, 41(7):752-757.
[
|
[22] |
刘峭, 李杨, 段宏, 等, 知识图谱构建技术综述[J]. 计算机研究与发展, 2016, 53(3):582-600.
[
|
[23] |
朱庆, 王所智, 丁雨淋, 等, 铁路隧道钻爆施工法智能管理的安全质量进度知识图谱构建方法[J]. 武汉大学学报·信息科学版, 2022, 47(8):1151-1164.
[
|
[24] |
张雪英, 张春菊, 吴明光, 等. 顾及时空特征的地理知识图谱构建方法[J]. 中国科学:信息科学, 2020, 50(7):1019-1032.
[
|
[25] |
|
[26] |
饶子昀, 张毅, 刘俊涛, 等. 应用知识图谱的推荐方法与系统[J]. 自动化学报, 2021, 47(9):2061-2077.
[
|
[27] |
|
[28] |
|
[29] |
|
[30] |
|
[31] |
|
[32] |
|
[33] |
|
/
〈 |
|
〉 |