顾及地址语义和地理空间特征的多源POI位置融合

Multi-source POI location fusion considering address semantics and geospatial features

多源POI位置融合是实现地理空间数据匹配融合的关键技术之一。然而,由于不同POI数据源之间位置编码的差异及定位误差,导致位置融合更加困难。本文提出了一种顾及地址语义和地理空间特征的多源POI位置融合方法。首先,通过TextRCNN和图注意力网络提取地址属性的语义特征;然后,使用多层感知机提取位置属性的地理空间特征;最后,基于自注意力机制通过特征聚合实现多源POI位置融合,并对成都市百度地图、腾讯地图和高德地图的POI数据进行试验验证。结果表明,该方法显著优于现有方法,平均位置融合精度优于12 m。

Multi-source POI location fusion is one of the key technologies for geospatial data matching and fusion.However,due to the difference of location coding and location error between different POI data sources,location fusion becomes more difficult.Multi-source POI location fusion considering address semantics and geospatial feature is proposed.Firstly,semantic features of address attributes are extracted by TextRCNN and graph attention network.Then,Multi-layer perceptron is used to extract geospatial features of location attributes.Finally,multi-source POI location fusion is realized by feature aggregation based on self-attention mechanism.We conduct experimental verification on the POI data of Baidu map,Tencent map and Amap in Chengdu.The results show that this method is significantly superior to the existing methods,and the average location fusion accuracy is better than 12 m.