NeuroTPR: A neuro‐net toponym recognition model for extracting locations from social media messages

Abstract

Social media messages, such as tweets, are frequently used by people during natural disasters to share real‐time information and to report incidents. Within these messages, geographic locations are often described. Accurate recognition and geolocation of these locations are critical for reaching those in need. This article focuses on the first part of this process, namely recognizing locations from social media messages. While general named entity recognition tools are often used to recognize locations, their performance is limited due to the various language irregularities associated with social media text, such as informal sentence structures, inconsistent letter cases, name abbreviations, and misspellings. We present NeuroTPR, which is a Neuro‐net ToPonym Recognition model designed specifically with these linguistic irregularities in mind. Our approach extends a general bidirectional recurrent neural network model with a number of features designed to address the task of location recognition in social media messages. We also propose an automatic workflow for generating annotated data sets from Wikipedia articles for training toponym recognition models. We demonstrate NeuroTPR by applying it to three test data sets, including a Twitter data set from Hurricane Harvey, and comparing its performance with those of six baseline models.

GazPNE: annotation-free deep learning for place name extraction from microblogs leveraging gazetteer and synthetic data by rules

“…Statistical learning: As the development of machine learning, specifically deep learning, using statistical learning algorithms to extracting place names is gaining more and more attentions (Wang, Hu, and Joseph 2020;Qi et al 2020;Gelernter and Mushegian 2011;Lingad, Karimi, and Yin 2013;Unankard, Li, and Sharaf 2015;Das and Purves 2019;Limsopatham and Collier 2016;Kumar and Singh 2019). Given abundant annotated data, statistical learning-based approaches can recognize the place names according to the context cues and the intrinsic features of place names, thus achieving a higher tagging accuracy than the rule-based approaches.…”

“…Using 10-fold crossvalidation, the model achieved an F1-score of 0.96. NeuroTPR (Wang, Hu, and Joseph 2020) extended a general Bi-LSTM architecture with several features to account for the linguistic irregularities in Twitter texts, such as the use of character embeddings to capture the morphological features of words, and POS tags and contextual embeddings to capture the semantics of tokens in tweets. The approach mitigates the need for a large training dataset by generating annotated data from Wikipedia articles for the task of location name extraction.…”

“…The incompleteness and inaccuracy of gazetteers (e.g., OpenStreetMap) are the two main drawbacks of the gazetteer-based approaches, leading to a low recall. The third approach is statistical learning (Finkel, Grenager, and Manning 2005;Kumar and Singh 2019), tagging the place names mainly according to the contextual features of the place names, such as Stanza (Qi et al 2020) and Neu-roTPR (Wang, Hu, and Joseph 2020). However, they require a considerable amount of annotated sentences which makes those approaches ineffective in most practical situations (Guerini et al 2018).…”

nLORE: A Linguistically Rich Deep-Learning System for Locative-Reference Extraction in Tweets

“…gazetteers of place names or lists of locative cues) together with supportive natural language processing (NLP) tasks, such as the part-of-speech (POS) tagging of n-grams [3,11,12]. Other systems are based on probabilistic frameworks, using Conditional Random Fields (CRF) or neural networks such as Convolutional Neural Networks (CNN) or Recurrent Neural Networks (RNN) with or without additional linguistic feature engineering [7,13]. Others present hybrid systems, which combine rule-based methods with machine-or deep-learning techniques and/or other named entity recognition (NER) tools [8,14].…”

“…A location extractor called NeuroTPR was built by [13], who used a bi-directional RNN with Long Short Term Memory (LSTM) enriched with linguistic features. Apart from building a tweet corpus from the 2017 Hurricane Harvey dataset, they also used GeoCorpora [4].…”