Text Mining from Migration Narratives

Abstract

The pervasive proliferation of textual information, combined with the swift advancement in data acquisition methods, has resulted in an overwhelming volume of data, making it challenging to uncover relevant patterns. Text mining is a crucial process for extracting noteworthy and non-trivial patterns, as well as valuable knowledge from extensive collections of textual data. In this paper, we present a step towards a text mining approach designed to harness migration narrative texts, those collected from interviews with migrants during their journeys in English and French. Our contributions can be summarized as follows: (1) We first collaborate with experts in Humanities and Social Sciences (HSS) to annotate the essential domain concepts, their related terms, and the locations mentioned in those narratives. (2) To automatically extract such related terms embedded in the narratives, we propose adapting a set expansion algorithm in a weakly supervised manner using a tiny set of annotated terms. We then evaluate the proposed algorithm by comparing its output terms to those annotated by experts. (3) We utilize some existing frameworks to automatically identify locations crossed by migrants, followed by a disambiguation model to precisely pinpoint them on a map. To evaluate the proposed systems, we conduct the experiments by comparing their recognized locations and disambiguated locations to those annotated by experts. (4) We design a tool to visualize the itineraries of those locations on a map, enabling the observation of migration routes. Our discussions with HSS experts reveal that our proposed approach assists their analyses by automatically retrieving pertinent terms and drawing itineraries of migrants on a map, enabling a comprehensive understanding of their construction.