Skip to content

It is aimed at calculation of affective scores of videos using the audio visual feature and further analyzing the affective pattern generated from the YouTube watching history of an individual to predict his depression severity score.

Notifications You must be signed in to change notification settings

manandey/Detecting-Depression-through-YouTube-history

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

20 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Detecting-Depression-through-YouTube-history

Introducton

This repository maintains the code from our experiments where we used Long Short-Term Memory(LSTM) networks to understand the video content from the transcripts in order to identify if a video can trigger depression. Additionally, we propose a novel method of validating the results by analysing the CESD score of comments. You can find more details about this experiment in our paper / poster that we presented at NeurIPS 2019 AI for Social Good Workshop. Our major contributions in this work are:

  • We construct a classifier that can help understand the content of a video by classifying it as Depressive/NonDepressive.
  • To provide a real-life validation of the classification results in above step, we propose a methodology to evaluate the comments posted for a video and determine a potential score that would have been obtained on a CES-D scale and use it as a real life proxy to judge the accuracy of the classification.

Requirements

Data Collection

The process of data collection has been explained in detail in Section 2.1. The code for the same can be found here

Comments Evaluation

For evaluation of the comments, we introduce a scoring method (Section 3), called CES-D score, for each video to analyze how depressive the video is. The score is the density of the terms, derived from the various symptoms (Insomnia, Self-hate, Appetite, etc.) considered in the CES-D scale, present in a negative connotation within a given text. Code for the same can be found here

Comparison of CES-D score

Results

We compare the performance of various models on their ability to classify a video as depressive/non-depressive by processing its transcripts.

Results

Website

An extension of this project involves analysing audio-visual features of the video to predict the arousal-valense response it would generate. More details can be found on the project website

Collaborators

@manandey

About

It is aimed at calculation of affective scores of videos using the audio visual feature and further analyzing the affective pattern generated from the YouTube watching history of an individual to predict his depression severity score.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 100.0%