obtaining first portrait feature data of each POI in a plurality of POIs and second portrait feature data of each tag in a plurality of marked tags corresponding to the plurality of POIs;

inputting the first portrait feature data of each POI and the second portrait feature data of each tag to a feature extraction model to be mapped to a metric space to obtain a first feature vector of each POI and a second feature vector of each tag, comprising: training a siamese neural network based on first sample portrait feature data and second sample portrait feature data to obtain the feature extraction model, the feature extraction model comprising a first feature extraction sub-model and a second feature extraction sub-model, the first feature extraction sub-model being configured to map the first portrait feature data to the metric space, and the second feature extraction sub-model being configured to map the second portrait feature data to the metric space; inputting the first portrait feature data of each POI into the first feature extraction sub-model to obtain the first feature vector of each POI in the metric space; and inputting the second portrait feature data of each tag into the second feature extraction sub-model to obtain the second feature vector of each tag in the metric space, wherein a first number of dimensions of the first feature vector and a second number of dimensions of the second feature vector are adaptively determined during the training;

optimizing at least one marked tag corresponding to a target POI based on a vector similarity between a first feature vector of the target POI and a second feature vector of at least one tag; and

providing a service by a map application based on the at least one marked tag optimized,

wherein optimizing the at least one marked tag corresponding to the target POI based on the vector similarity between the first feature vector of the target POI and the second feature vector of the at least one tag comprises:

calculating a vector similarity between the first feature vector of the target POI and a second feature vector of a target tag;

when the vector similarity between the first feature vector of the target POI and the second feature vector of the target tag is greater than a first preset threshold, and the target tag is not a marked tag corresponding to the target POI, adding the target tag into the at least one marked tag corresponding to the target POI; and

when the vector similarity between the first feature vector of the target POI and the second feature vector of the target tag is lower than a second preset threshold, and the target tag is the marked tag corresponding to the target POI, deleting the target tag from the at least one marked tag corresponding to the target POI;

the second preset threshold being less than the first preset threshold,

wherein obtaining the first portrait feature data of each POI in the plurality of POIs and the second portrait feature data of each tagin the plurality of marked tags corresponding to the plurality of POIs comprises:

for each POI, determining historical accessing users and/or historical search users corresponding to the POI, and aggregating user portraits of the historical accessing users and/or the historical search users corresponding to the POI to obtain the first portrait feature data of the POI; and

for each tag, determining at least one POI corresponding to the tag, and generating the second portrait feature data of the tag based on first portrait feature data of the at least one POIs corresponding to the tag.