To scale Airbnb’s ability to answer travel queries, Airbnb decided to build a knowledge graph to store and serve high-quality information about entities (e.g. cities, landmarks, events, etc.) and the relationships between them (e.g. the most popular landmark in a city, the best neighborhood for tacos, etc.). In this post, Xiaoya Wei, a software engineer at Airbnb, dives into what went under the hood of Airbnb's knowledge graph infrastructure.
To begin with, the knowledge graph development team (team) at Airbnb built a graph storage module. The team adopted an in-house relational data store as the underlying database. Further, the team implemented a node store and edge store and assigned a global unique identifier (GUID) to each node or edge. Next, to support their product use case, the team implemented a graph query endpoint in addition to create, read, update, and delete (CRUD) endpoints for nodes and edges in the knowledge graph API module.
The team then built a storage mutator to facilitate the import of data to the graph storage and a mutation publisher to propagate the mutations downstream. Subsequently, they built a rich taxonomy in the knowledge graph and used it to categorize the inventories at Airbnb. With this initiative, Airbnb aimed at providing its users with a deep understanding of every single home on the Airbnb platform, enabling them to find the single best home for their trip.
To enrich semantics in the knowledge graph, the team built a taxonomy as a part of the ontology used to describe Airbnb’s inventory and the world around it. The taxonomy is in a hierarchical structure, which represents concepts in different levels of granularity, such that higher-level concepts can be mapped to very specific ones. Further, every experience, home, or restaurant was tagged with the relevant nodes in the taxonomy to categorize Airbnb’s inventory.
The team set out to use a knowledge graph to provide a consistent interface to clean, current, and complete structured data about Airbnb’s inventory and the world of travel. By serving connected and high-quality data via the knowledge graph, the goal was to improve guest and host experiences at Airbnb. This graph knowledge initiative has already helped Airbnb enhance and personalize searching, supply groupings, and content delivery.
Click here to read the article.
Please give your feedback on this article or share a similar story for publishing by clicking here.
