Graph-Based Geolocation
Dirk and Matt discuss the challenges of applying graph-based methods to geolocate tweets, emphasizing the importance of managing neighborhood sizes for computational efficiency. By restricting word connections based on occurrence frequency, they were able to streamline the process, showcasing how precomputing shared neighborhood information can significantly speed up computations in NLP tasks.In this clip
From this podcast
NLP Highlights
76 - Increasing In-Class Similarity by Retrofitting Embeddings with Demographics, with Dirk Hovy
Related Questions