Inexpensive Detection of Substance Abuse Based on Social Media Data using Machine Learning

Abstract

Over the past few years, substance abuse has become one of the most severe public health problems in the United States. The annual cost of substance abuse aftereffects in the United States alone is approximately $3.73 Trillion. The societal costs of substance abuse include premature deaths, lost productivity, and increased crime rates. Unfortunately, many victims, especial¬ly in lower-income families, don't have access to early detection and early family intervention tools due to limited access to traditional diagnostic tools and rehab specialists. Currently, there is no complete diagnostic pipeline to inexpensively detect substance abuse and automatically inform family members or trusted contacts. To combat this, the experimenter developed the SOS 280 system, which utilizes machine learning techniques in a smartphone application. SOS 280 works through social media monitorin¬g and automatic notification using SMS and GPS location. The SOS280 algorithm primarily uses social media data, namely publicly available Twitter, and Instagram posts, to identify substance abuse-related activity. The experimenter collected and classified data by applying for the Twitter and Instagram Developer API Platforms, mining tweets and posts with specific drug keywords present. The investigator trained a Natural Language Processing (NLP) text classification model to analyze the sentiments on the tweets, then classifying them as positives (containing substance abuse-related keywords) and negatives. The master model is a Bidirectional Encoder Representations (BERT) derivative that uses a transformer-based architecture to detect emotions in sentences and conversations to classify substance abuse instances. In total, the researchers looked at 55,551 tweets and Instagram posts indicative of potentially alarming substance usage. Finally, the experimenter developed a smartphone application to capture trusted contact information and GPS location, send data to a remote server housing the neural network, output the network's detection, and send automated alerts to trusted contacts via SMS and GPS location. The experimenter further validated the system's effectiveness through a partnership with national nonprofit Faces and Voices of Recovery, which works with 23 million addiction recovery victims. SOS280 is an inexpensive, reliable, easy to use, and timely tool for families of young adults in predicting substance abuse.