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

Abstract

Place name extraction refers to the task of detecting precise location information in texts like microblogs. It is a vital task to assist disaster response, revealing where the damages are, where people need assistance, and where help can be found. All current approaches for extracting the place names from microblogs face crucial problems: rule-based methods do not generalize, gazetteer-based methods do not detect unknown multi-word place names, and machine learning methods lack sufficient data, which is costly to annotate on scale. We propose a hybrid method that avoids these problems, named GazPNE, which fuses rules, gazetteers, and deep learning methods to achieve state-of-the-art-performance without requiring any manually annotated data.Specifically, we utilize C-LSTM, a fusion of Convolutional and Long Short-Term Memory Neural Networks, to decide if an n-gram in a microblog text is a place name or not. The C-LSTM is trained on 4.6 million positive examples extracted from OpenStreetMap and GeoNames and 220 million negative examples synthesized by rules and evaluated on 4,500 disaster-related tweets, including 9,026 place names from three floods: 2016 in Louisiana (US), 2016 in Houston (US), and 2015 in Chennai (India). Our method improves the previous state-of-the-art by 6%, achieving an F1 of 0.86.

ChineseTR: A weakly supervised toponym recognition architecture based on automatic training data generator and deep neural network

“…This process offers a new perspective of combining deep learning and gazetteers to automate the generation of sufficiently large training sets while minimizing the interference caused by noise. Generated sentences may be pseudo‐sentences (Hu et al., in press; Huang, Qu, Jia, & Zhao, 2019; Rolnick, Veit, Belongie, & Shavit, 2017).…”

“…With the development of machine learning (i.e., deep learning), a body of research efforts for extracting toponyms based on statistical learning models has been developed using unstructured text (Das & Purves, 2020; Hu et al., in press; Qi et al., 2020; Wang et al., 2020). Based on a large amount of annotated data, the statistical learning‐based approach can recognize toponyms based on contextual cues and intrinsic features of the toponyms, achieving higher annotation accuracy than the rule‐based approach.…”

“…Therefore, accurate extraction and recognition of toponyms from the natural language texts is necessary, and it is regarded as an extractive problem rather than a retrieval problem. This task (i.e., geoparsing) aims to recognize a set of toponyms from natural language texts and identify their relevant spatial footprints (Gelernter & Balaji, 2013; Gritta, Pilehvar, & Collier, 2018; Hu et al., in press; Wang et al., 2020). As a study theme, it has been extensively researched in the domain of geographical information retrieval (Jones & Purves, 2008; Purves, Clough, Jones, Hall, & Murdock, 2018; Wang et al., 2020).…”

“…Very recently, three approaches toward the extraction and recognition of toponyms from natural language texts have been applied. The first method is a rules‐based approach that applies pattern and regex matching, and depends heavily on a dataset of words or orthographic feature‐based handcrafted rules (Hu et al., in press; Malmasi & Dras, 2015; Weissenbacher, Magge, O'Connor, Scotch, & Gonzalez, 2019). Considering the enormous variation in writing styles and the large number of grammatical errors found in different texts (e.g., microblogs), hand‐designed rules are often fragile and cannot produce a universal place‐name extractor (Hu et al., in press; Ritter, Clark, & Etzioni, 2011).…”

“…The first method is a rules‐based approach that applies pattern and regex matching, and depends heavily on a dataset of words or orthographic feature‐based handcrafted rules (Hu et al., in press; Malmasi & Dras, 2015; Weissenbacher, Magge, O'Connor, Scotch, & Gonzalez, 2019). Considering the enormous variation in writing styles and the large number of grammatical errors found in different texts (e.g., microblogs), hand‐designed rules are often fragile and cannot produce a universal place‐name extractor (Hu et al., in press; Ritter, Clark, & Etzioni, 2011). The second method is a gazetteer‐based method that recognizes toponyms by matching them with information in the gazetteer (Al‐Olimat, Thirunarayan, Shalin, & Sheth, 2018; Dutt et al., 2018).…”