Pre-emptive Identification of online grooming using NLP Chatbots

Pre-emptive Identification of online grooming using NLP Chatbots

Using Chatbots to identify is not new [1-6], but more work is required given its estimated 1 out of 7 children online will receive a sexual solicitation [7] and the heinous nature of the crime. An multi-layered approach proposed below.

  1. Chatbots: We specifically recommend because they are more interactive, have benefited from recent algorithm improvements, and can include voice, images etc.
  2. Sentiment and Emotion Based Analysis: Use of Sentiment and emotions based analyses in Chatbots, but by combining newly identified features from all existing studies [1-6] and leveraging new discoveries in behavioural psychology in online grooming.
  3. Embedded Chatbots into Social Platforms: One approach not seen in my research is to proactively embed these Chatbots into social media sites frequented by young children. e.g., gaming forums, Facebook, children’s sites etc. Chatbots suspecting online grooming can alert administrators of the sites and police. This also allows active sharing of lessons learnt between platforms which improve their effectiveness (analogy of rising tide lifts all ships).
  4. Advice Feature: Another recommendation not seen in my research is to have the embedded Chatbot pre-emptively advise children if online grooming is suspected. Advantage is in alerting potential victims as early as possible and providing other avenues if children don’t want to involve parents. Advice can be:
    • Ask your parents for advice,
    • If child is not keen on approaching parents another adult, a hotline number can be provided in which they can report the incident and get help from experts.
  5. Remove Identifiable / Geo-location Data: Automatic removal of personal data embedded in photos shared e.g., school names/logs, geo-location information, i.e. removing identifiable information of victims.

In conclusion, more work is needed; a number of specific and practical recommendations were made. The key is to combine these recommendations with each other for maximum effectiveness.

References

[1] Maxime Meyer Machine learning to detect online Grooming http://uu.diva-portal.org/smash/get/diva2:846981/FULLTEXT01.pdf

[2] Dasha Bogdanova, Paolo Rosso and Thamar Solorio On the Impact of Sentiment and Emotion Based Features in Detecting Online Sexual Predators Proceedings of the 3rd Workshop on Computational Approaches to Subjectivity and Sentiment Analysis, 2012

[3] Nick Pendar. Toward spotting the pedophile: Telling victim from predator in text chats. In Proceedings of the International Conference on Semantic Computing, Irvine, California, 2007.

[4] Vincent Egan, James Hoskinson, and David Shewan. Perverted justice: A content analysis of the language used by offenders detected attempting to solicit children for sex. Antisocial Behavior: Causes, Correlations and Treatments, 2011.

[5] India McGhee, Jennifer Bayzick, April Kontostathis, Lynne Edwards, Alexandra McBride and Emma Jakubowski. Learning to identify Internet sexual predation. International Journal on Electronic Commerce 2011.

[6] AI chatbot poses as a young girl to trap pedophiles https://www.gmanetwork.com/news/scitech/technology/317923/ai-chatbot-poses-as-a-young-girl-to-trap-pedophiles/story/

[7] Predator statistics, http://www.internetsafety101.org/ predatorstatistics.htm

By: Mehdi Kiani

 

Leave a Reply

Your email address will not be published. Required fields are marked *