Research On OpenStreetMap Road Tag Recommendation Method Based On Spatial And Semantic Features

Open Street Map(OSM)is widely used in various industries as one of the most successful VGI.However,the data quality,especially the semantic quality,still needs to be further improved,due to the free mechanism and volunteers’ different comprehension of geographic objects.Therefore,this paper mainly takes the OSM road data as the research object and conducts relevant research about the tags in the data.The overall quality of OSM data by improving the semantic accuracy,consistency,and integrity of OSM road objects.The specific research contents mainly include:Recommendations based on OSM object type can greatly reduce the number of wrong tags and improve the semantic quality of data.In recent years,some researchers have extracted features through manual analysis and trained models.However,the manual extraction method lacks the consideration of the interaction between attributes.To further conform to the characteristics of OSM data and improve the accuracy of the model,this paper proposes road type prediction methods,which consider the interaction of low-order and high-order features of OSM data.In this method,firstly,the road data are classified by the OSM classification rules.Then,according to the characteristics of OSM data,the semantic and spatial attribute features are extracted and encoded.Finally,OSM road type prediction based on x Deep FM is realized by the low-order and high-order interactive learning mechanism among features.Experiments show that the proposed method can effectively predict road types,and it is superior to the Random Forest algorithm used in the relevant literature.At the same time,it is found that the unbalanced distribution of data types in different regions has a significant impact on the effect of type prediction.In order to further improve the semantic consistency of OSM data,on the basis of recommending road types for users,the experiment conducted deeper research on OSM’s element tags(such as highway=motorway).Therefore,this paper proposed an element tags recommendation method,which analyzes the differences of tags in different types,and the tagvalue matrix in the road type prediction method is modified adaptively.At the same time,the road name attribute with high integrity is added.Finally,the recommendation of element tags in the same type is realized.Comparative experiments show that the road element tag recommendation method can significantly improve the recommendation accuracy,and can rival the related literature methods without using the objects around OSM.Aiming at the non-element tags in OSM data,this paper proposes a non-element recommendation method based on machine learning.The method can recommend non-element tags for users to enrich the attribute information of objects and improve semantic integrity.In this method,firstly,the tag keys are separated from the data,and then the association rule algorithm is used to mine the co-occurrence rules among the tag keys.Finally,according to the actual situation that the tag keys are not reused in the OSM platform,the improved Markov process algorithm is adjusted to realize the recommendation of non-element tag keys,which provides a new idea for improving the semantic integrity of road objects.In this paper,the correctness and applicability of our proposed method are verified by combining the experimental results with official data.At the same time,this paper finds that there are differences in the cognition of geographical objects and the meanings of tag keys among contributors in different regions.