lyft2vec - Lyft的嵌入式系统



, and


An illustrative city map of an unidentified city showing highlighted streets and some topography, but with no labels.



Graph learning methods can reveal interesting insights that capture the underlying relational structures. Graph learning methods have many industry applications in areas such as product or content recommender systems and network analysis.


In this post, we discuss how we use graph learning methods at Lyft to generate embeddings — compact vector representation of high-dimensional information. We will share interesting rideshare insights uncovered by embeddings of riders, drivers, locations, and time. As the examples will show, trained embeddings from graphs can represent information and patterns that are hard to capture with traditional, straightforward features.


Lyft Data and Embeddings


At Lyft, we have semi-structured data capturing complex interactions between drivers, riders, locations, and time. We can construct graphs representing these interactions (e.g. a graph can be formed by connecting a rider with all the locations they have visited). From these graphs we can generate embeddings to succinctly express a rider’s or driver’s entire ride history. These embeddings allow us to efficiently summarize vast and varied information in a machine-friendly representation.


For example, there are over 9,000 Geohash-6 (Gh6) level locations around the San Francisco Bay Area. If we wanted to describe a driver’s ride history around the Bay Area without embeddings, we would need a histogram or vector of length over 9,000 to describe it precisely. The vector would contain the number of rides the driver has started in each Gh6, with lots of zeros if the driver has never been in some Gh6s.



首页 - Wiki
Copyright © 2011-2023 iteam. Current version is 2.121.1. UTC+08:00, 2023-12-01 17:38
浙ICP备14020137号-1 $访客地图$