Locality-adapted kernel densities of term co-occurrences for location prediction of tweets

•We estimate tweet locations using spatial co-occurrence patterns and kernel density estimators.•Our method combines probability distributions of texts based on locality-adapted KDE settings.•Our method does not use any language-specific setting, parameter tuning or data other than tweet text.•We compare our method with neural networks, generative models and discriminative models.•Evaluations using three different tweet sets indicate significant improvement in accuracy.