Social listening AI to identify potential scams
-
We're interested in potential ways to use AI to scan social media (e.g., X/ Twitter, Reddit, news articles and other various sources) and identifying potential scams and schemes so clients can mobilize accordingly. Does anyone have experience with implementing something like this to scan social media? Or have perspectives on how this can be done?
-
Procuring the data in a legal and cost-effective manner is likely the biggest hurdle. Many APIs for social media are limited (e.g. Reddit) or costly (e.g. X/ Twitter), and using tools to scrape their websites is usually against their Terms of Service. News aggregators that include smaller news sources may be useful, like OW's Factiva Sentiment Signals product
If you have access to the data, you can use NLP methods or LLMs to filter for posts that describe scams, and use LLMs to find themes and extract entities (e.g. if scammer is posing as a certain company), and ultimate then classify posts to generate trend reports.
There are quite a few examples at OW where NLP / LLM have been used for similar purposes (e.g. identifying new themes and trends in customer complaints) – you’ll probably hear from them directly (if you don’t, let me know and I can try to find those who have this experience).
- Note: This is not identifying whether a post is a scam (as classifying new types of scam is a much tougher task), but rather identifying posts/ news describing or complaining about potential scams
-
There are two steps:
- (1) raw social data collection; and
- (2) analysis with NLP or GenAI
(1) is the difficult part. You would need to know where the discussions are happening for your client or its industry – is it in the news, Facebook, Instagram, a forum? Once you identified the data source, you would need to find a way to collect the data, and the legal and cost question comes in to play.
I recommend you to buy data from a third party (there are many of them). They can either provide you only the raw data feed (e.g. give me all the Facebook discussions about my brand), or they can provide you with a dashboard with standard analytics. If you only need one off data analysis for maybe a proposal, getting raw data and crunch it yourself is a cheaper, better way forward