Fake Financial Advisor Scam Detector

Logo
fruity friends architecture
Fetch comments design
AI detector design
Algo detector design
Chief Architect
Chief Engineers and subsequent software systems running on independent machines

Inspiration

We browse through all types of social media for financial help, from Youtube, TikTok, Instagram, and Facebook. We looked at YouTube videos specifically, and many financial help videos include malicious comment threads to social engineer people into visiting fraudulent fake financial advisor websites and contacting them. We decided to take the fight back against these recently introduced scams.

What it does

Fake Financial Advisor Scam Detector is an advanced tool designed to automatically detect and flag suspicious comments on YouTube videos. The system scans for known scam keywords, identifies patterns like fake names, and detects red flags such as all-uppercase text and multiple capitalized words. It then calculates a suspicion score and provides an analysis of whether the comment thread is likely linked to a scam, helping users avoid fraudulent financial advisors.

How we built it

We built the tool using Python, gRPC for client-server communication, and MongoDB for storing and retrieving scam data. We integrated the nameparser library to detect potential fake names in comments, and implemented a pattern-matching algorithm to flag scam indicators such as keywords, suspicious names, and formatting anomalies. The detector also communicates with an API to cross-reference the identified names with known scam websites. We also used machine learning techniques to improve accuracy in identifying malicious content based on flagged patterns.

Challenges we ran into

One major challenge we encountered was accurately distinguishing between legitimate financial advisors and scammers using similar language. Another hurdle was efficiently processing and analyzing large volumes of comments without overwhelming the system, especially when working with YouTube's API. Additionally, ensuring the detector could recognize names and phrases in various languages and formats required careful testing and adjustments.

Accomplishments that we're proud of

We’re proud of creating a functional, scalable tool that not only helps everyday users in identifying and avoiding scams but also actively reports detected scams to relevant authorities. We’re making a real difference by taking action and addressing the problem head-on. Our solution has processed over 600 comments and fed 25+ videos into the database. It detects scam comments with a high confidence level, alerts users in real-time, and ensures malicious actors are reported. With a 91% accuracy rate in our algorithm-based detector and an 83% accuracy AI model, we’ve developed a robust system. We are proud to be not only detecting fraud but actively doing something about it by reporting 20 websites we strongly believe are linked to malicious origins

What we learned

Throughout the development process, we gained extensive experience with various technologies and methodologies that greatly enhanced the functionality of our tool. By utilizing gRPC (Google Remote Procedure Call), we were able to create a fast, efficient, and scalable communication framework between our server and clients. This allowed real-time processing of comments and enabled us to report detected scams instantly. Protocol Buffers (protobuf) were instrumental in serializing structured data efficiently, ensuring smooth and compact data transmission between services.

We also ventured deep into the world of Artificial Intelligence (AI), leveraging natural language processing (NLP) to analyze and detect fraudulent comments. This involved creating sophisticated algorithms that could sift through large volumes of comments, identify patterns indicative of scams, and flag malicious content with a high level of accuracy. Training our AI to understand nuances, like recognizing scam keywords or detecting suspicious three-part names, required careful data curation and model tuning.

In the backend, we focused on creating and managing databases to store vast amounts of scam data, enabling the system to learn from previous scam reports and improve its detection capabilities. We employed MongoDB as our database solution, which provided the flexibility and scalability needed to handle unstructured data, such as scam reports and comment threads, and to efficiently store and retrieve information for real-time analysis.

Additionally, we developed a robust algorithm to calculate confidence levels for scam detection based on various signals such as suspicious names, keywords, and comment patterns. This algorithm dynamically adjusts based on new data, making it adaptable to the evolving tactics of scammers.

What's next for Fake Financial Advisor Scam Detector

Moving forward, we plan to expand the detector to work on more social media platforms like TikTok, Instagram, and Facebook. We aim to refine the detection algorithm to improve accuracy, especially for international scams. Additionally, we hope to partner with financial institutions and social media platforms to integrate the tool directly, providing users with real-time scam alerts as they browse online.

https://www.canva.com/design/DAGTainfYuc/x7P_x6DDKXq-HtuCQcVZPA/view?utm_content=DAGTainfYuc&utm_campaign=designshare&utm_medium=link&utm_source=editor