A natural language processing and geospatial clustering framework for harvesting local place names from geotagged housing advertisements

摘要: Local place names are frequently used by residents living in a geographic region. Such may not be recorded existing gazetteers, due to their vernacular nature, relative insignificance gazetteer covering large area (e.g., the entire world), recent establishment name of newly-opened shopping center), or other reasons. While always recorded, local play important roles many applications, from supporting public participation urban planning locating victims disaster response. In this paper, we propose computational framework for harvesting geotagged housing advertisements. We make use those advertisements posted on local-oriented websites, such as Craigslist, where often mentioned. The proposed consists two stages: natural language processing (NLP) and geospatial clustering. NLP stage examines textual content advertisements, extracts candidates. focuses coordinates associated with extracted candidates, performs multi-scale clustering filter out non-place names. evaluate our comparing its performance six baselines. also compare result four gazetteers demonstrate not-yet-recorded discovered framework.