Google Research and a team from the University of Southern California have introduced Mobility-Embedded POIs, or ME-POIs, a framework designed to improve semantic understanding of places by combining aggregated human mobility data with text-based place vector representations. The aim is to capture how locations are actually used, rather than relying only on static textual descriptions.
According to the report, the model has about 53.7 million parameters and can be pre-trained on a single NVIDIA Tesla V100 GPU. The framework was tested across five map enhancement tasks using mobility data from Los Angeles and Houston.
After adding ME-POIs to existing text encoders, performance improved in 34 out of 35 model-task combinations in Los Angeles. The biggest relative gain came in visit-intent classification, where the F1 score increased by as much as 81.9%. The item was cited by Techub, with MarkTechPost named as the original report source.
Google Research and a team from the University of Southern California have released Mobility-Embedded POIs, or ME-POIs, a framework that blends aggregated human mobility data into text-based vector representations of places.
The framework is meant to capture how locations are used in practice, instead of relying only on static descriptions of those places.
Mobility data added to place representations
According to the report, ME-POIs improves semantic understanding of points of interest by feeding aggregated human movement data into text-based place embeddings. The approach is intended to reflect real-world usage patterns tied to locations.
Model size and training setup
The model contains about 53.7 million parameters and can be pre-trained on a single NVIDIA Tesla V100 GPU.
Results from map enhancement tests
In tests covering five map enhancement tasks with mobility data from Los Angeles and Houston, adding ME-POIs to existing text encoders improved performance in 34 of 35 model-task combinations in Los Angeles. The largest relative gain was in visit-intent classification, where the F1 score improved by as much as 81.9%.
Techub carried the item and cited MarkTechPost as the source.
This article was originally published by Bit.Fan. For more cryptocurrency news and market insights, visit www.bit.fan. Disclaimer:
The market information, project data, and third-party content displayed on this platform are for industry information sharing only and do not constitute any form of investment advice or return commitment.
Cryptocurrency trading carries high risks. Users should fully assess their risk tolerance and make independent decisions. All profits, losses, and legal responsibilities are borne by the users themselves.
