Extracting Person Names from User Generated Text: Named-Entity Recognition for Combating Human Trafficking

Findings of the Association for Computational Linguistics: ACL 2022, pages 2854 - 2868 May 22-27, 2022 c⃝2022 Association for Computational Linguistics Extracting Person Names from User Generated Text: Named-Entity Recognition for Combating Human Trafficking Yifei Li Pratheeksha Nair Kellin Pelrine Reihaneh Rabbany School of Computer Science McGill University Mila – Quebec AI Institute Abstract Online escort advertisement websites are widely used for advertising victims of human trafficking. Domain experts agree that advertis- ing multiple people in the same ad is a strong indicator of trafficking. Thus, extracting person names from the text of these ads can provide valuable clues for further analysis. However, Named-Entity Recognition (NER) on escort ads is challenging because the text can be noisy, colloquial and often lacking proper grammar and punctuation. Most existing state-of-the-art NER models fail to demonstrate satisfactory performance in this task. In this paper, we pro- pose NEAT (Name Extraction Against Traf- ficking) for extracting person names. It effec- tively combines classic rule-based and dictio- nary extractors with a contextualized language model to capture ambiguous names (e.g penny, hazel) and adapts to adversarial changes in the text by expanding its dictionary. NEAT shows 19% improvement on average in the F1 classifi- cation score for name extraction compared to previous state-of-the-art in two domain-specific datasets. 1 Introduction There are approximately 4.8 million people be- ing trafficked around the world for commercial sex, a global industry estimated to be worth $99 bil- lion USD (Office, 2017). Technology has been a critical tool for traffickers to recruit, advertise and exploit victims (on Drugs and Crime, 2020) and the majority of human trafficking (HT) victims are advertised on online escort websites (Rhodes and Rhodes, 2016). Recently, there have been multi- ple efforts to leverage AI techniques for analyz- ing the online advertising market and providing law enforcement with actionable intelligence to counter human trafficking (Tong et al., 2017; Rab- bany et al., 2018; Lee et al., 2021). One of the key Correspondence to: pratheeksha.nair@mila.quebec tasks in this domain is extracting information from the online escort ads e.g. names, phone numbers, locations, prices and activity types, which are crit- ical for higher level analysis and modus operandi detection. This information needs to be extracted from the text of the ads, and it plays a vital role in tasks such as identifying groups of related ads and finding links between them. However, this is a challenging task since in order to avoid detection, this data is made: • noisy and obscured e.g. using ‘rose’ symbols as a proxy for dollar sign and spelling variations such as ‘Cathy’ and ‘Kathy.’ • evolving and adversarial e.g. traffickers are ac- tively introducing new patterns and intentional misspellings to avoid detection, e.g. adapting new phrases to indicate underage victims such as ‘amber alert’ or intentionally misspelling ‘miss’ as ‘mizz.’ Names have a particular importance. Traffick- ing is an organized activity (Lee et al., 2021) and multiple people being involved in an escort ad is a strong indicator of human trafficking. Also, the more …

Content truncated to 500 words