Introduction

• The WHO has qualified stress as a 'world epidemic' due to its increasingly greater incidence on health.
   
• High levels of chronic stress have adverse effects on physical and mental health.
   
• If the stress response system remains active for a prolonged period, this can increase the likelihood of developing several health issues, such as anxiety, depression, and various physical ailments (McEwen, 2007), and add to the global burden of disease (GBD 2019 Mental Disorders Collaborators, 2022).

Introduction

• Measuring stress is crucial for timely intervention due to its impact on health and well-being. It helps in identifying vulnerable individuals and reducing the causes of stress (Cranwell-Ward, 2005).
   
• Despite its association with chronic diseases, stress is rarely assessed in primary care, highlighting the need for its routine measurement (Wulsin, 2022).
   
• Stress measurement also allows for the verification of intervention effectiveness and assessment of individual differences in stress reactivity (Sumińska, 2022).
   
• However, the lack of a precise definition of stress and a reliable measurement method remain significant challenges (Llobe, 2019).

Literature Review

• young adults in India face many stressors in their day-to-day lives and thus experience high levels of stress. A cross-sectional study with rural adolescent students in Maharashtra showed high prevalence rates of depression (53.9%), anxiety (59.7%), and stress (43.8%) measured with the Depression, Anxiety, and Stress Scale (DASS-21) and a strong correlation between stress, depression, and anxiety (Shaikh et al., 2018).
   
• A study with high school students from the south zone of Delhi indicated that stress notably affected adolescents’ mental health, leading to internalizing problems such as anxiety and withdrawal or externalizing problems such as rule-breaking and aggression (Mathew et al., 2015).

• A large-scale qualitative study was conducted with 191 young adults from Delhi and Goa, where 22 focus group discussions were conducted to identify the most common stressors experienced by Indian adolescents. The study revealed that academic pressure, romantic relationships, negotiating autonomy, and safety/victimization were the most frequently reported stressors.
   
• The students felt pressured to perform well academically, were stressed due to parental disapproval or failure of romantic relationships, faced restrictions of personal freedom, felt peer pressure to use substances, and experienced safety concerns such as bullying, corporal punishment, and gender discrimination (Parikh et al., 2019).
   
• Urban young adults in Mysore were found to face the following main stressors: family pressure, academic stress, peer-related stress, lack of financial security, and gender discrimination (Nagabharana et al., 2021).
   
• Girls, also face stress related to gender roles and sexual harassment (Parikh, 2019).

Literature Review

• A range of methods exist for measuring stress, including self-reporting scales, physiological and biochemical measures, and questionnaires (Downs, 1990; Reisman, 1997; Derevenco, 2000).
   
• These methods can assess stress factors, reactions, and relationships with the environment, and are particularly useful in diagnosing occupational stress (Derevenco, 2000).
   
• However, the diagnosis of stress is complex and subject to experimental error.

Literature Review

Literature Review

• A range of methods have been developed to identify and quantify stressors. Cohen (1995) provides a comprehensive overview, including check-list and interview measurements of stressful life events, as well as the measurement of stress hormones and immune response.
   

• Sharma (2012) focuses on non-invasive and unobtrusive sensors for measuring stress, and computational techniques for stress recognition and classification.
   

• Cooper (1983) reviews research on work stressors, such as shift work, job overload, and role conflicts.
   

• Aguiló (2015) presents a method to objectively quantify stress levels, based on the identification of stress types and indicators, and the use of psychometric tests and well-documented stressors.

   

Rationale

• Chronic stress in adolescence is a more pressing problem.
   
• Studies have shown that chronic stress in adolescence has deteriorating impact on several aspects of them including lowered academic performance, low self-esteem, high demands,  poor health, and insufficient sleep. (Schraml, 2012).
   
• Furthermore, The growing prevalence of stress, anxiety, and depression in adolescents is a concern, with potential impacts on brain development and treatment options (McKain, 2019).
   

Rationale

• The rise of viewership of short-form video content is an Interesting Phenomenon
   

• The Youth population in Particular is a large consumer of short-form video content, and its almost a second-hand habit to scroll through reels, and liking has become almost an unconscious effort one might take. 

Research Gap

• Several studies have studied the negative impacts of short-form video content on the youth population (Muda (2018)).
  
   
• Various other studies have looked into the possibility of using short-form video content as an intervention. (Yuting Yang,2023) For example, investigated the effect of a short video-based mental health intervention on depressive symptoms in Chinese adolescents.
   
• These efforts and studies suggest a strong tie between short-form video content and mental health, But No studies till now have tried to use short-form video content to estimate/quantify one's mental state like stress/ to identify stressors.
   
• We believe This type of measurement would be highly relevant, and a  natural means of measurement for the adolescent population. 

Objectives

• Hence our Interdisciplinary study aims to access the following:

Given A sequence of reels/shorts(Short form video Content) , and which reels/shorts a person likes (Liking Pattern), be used as an estimator for chronic stress, and possibly identify the stressors.

Objective

Given A sequence of reels/shorts(Short form video Content), and which reels/shorts a person likes (Liking Pattern), be used as an estimator for chronic stress, and possibly identify the stressors.

Objective

Given A sequence of reels/shorts(Short form video Content), and which reels/shorts a person likes (Liking Pattern), be used as an estimator for chronic stress, and possibly identify the stressors.

To This End, We need a resonable "representation" of Short Videos 

Objective

1) Given A sequence of reels/shorts(Short form video Content), and which reels/shorts a person likes (Liking Pattern), be used as an estimator for chronic stress, and possibly identify the stressors.

To This End, We need a resonable "representation" of Short Videos 

2) Learn a Reasonable Dense Representation on Short videos, that can be used for any downstream tasks

Methodology

• Recent times have seen rise of "Pre-Train and Adapt" paradigm in deep learning
• Scaling Laws Suggest Better Models Arise by simply scaling Model Size and Data
• Owing to this, we resort to extract embeddings using some pretrained models, Then adapt to our task at hand

Methodology

Methodology

V-JEPA

• Primarily motivated by
  the predictive feature principle, which states that representations of temporally adjacent sensory stimuli should be predictive of each other.
• Driven by this principle, the
  architecture aims to learn to predict adjacent sensory stimuli in latent space in a selfsupervised manner.

Methodology

V-JEPA

• Trained on the following objective

Methodology

Wav2Vec

Methodology

Indic- Wav2Vec

• Foundation Model By SPRING Labs, IITM
• Pretrained on 2000 hours of legally sourced and manually transcribed speech data spanning over 22 Indian languages

Methodology

Indic- Wav2Vec

Methodology

Insight face

RetinaFace-10GF

FaceDetection

Face Recognition

Methodology

Insight face

RetinaFace-10GF

Methodology

Dataset

3MASSIV

Methodology

Dataset

3MASSIV

• To the best of the authors’ knowledge, three datasets are available for portrait mode video recognition: S100-PM, 3Massiv PortraitMode-400
• We chose to utilize the 3Massiv Dataset due to its relevance to the Indian context and its multilingual nature.

Methodology

Dataset

3MASSIV

• According to the 3Massiv Report, the 3Massiv dataset comprises 50k labeled short videos from the Moj video sharing app.
• However, from the available links, only roughly half were functional, and we were able to down-
  load approximately 17k training samples, 5k test samples, and 5k validation samples.
• Although the videos were not evenly distributed across languages, the fraction of videos spanning
  various concepts and emotions across languages was fairly even.

Methodology

Dataset

3MASSIV

Methodology

Dataset

3MASSIV

Methodology

Dataset

3MASSIV

Methodology

Dataset

3MASSIV

Methodology

Dataset

3MASSIV

Methodology

Training Recipe

• The multi-task learning framework encompasses four tasks, namely, video categorization, topic identification, emotion recognition, and language detection. Each task is a simple classification task, and we employ a separate classification head for each task.
   
• To mitigate overfitting, we employ early stopping, where we monitor the validation loss for each task and terminate training when the validation loss for any task does not improve for a certain number of epochs
   
• we employed a learning rate decay factor of 0.1 and a decay interval of 10 epochs to adapt to the changing loss landscape during training.

Methodology

Training Recipe

Results

Results

Results

Academic Stress Prediction using
Short-Form Video Content

• we utilize the learned representation to explore whether there is a correlation between the type of content consumed and the type of stress experienced by students.

Academic Stress Prediction using
Short-Form Video Content

Data Collection

Academic Stress Prediction using
Short-Form Video Content

Data Collection

• All participants were recruited from Amrita Vishwa Vidyapeetham, Coimbatore campus, resulting in a total of 22 participants. 
•  The entire process was automated and hosted on Google Cloud and App Scripts.
• Once participants reached 50 video views, their session was terminated, and a Google Form was sent to them via email, containing psychometric questionnaires.

Academic Stress Prediction using
Short-Form Video Content

Data Collection

• We collected short-form video content through two distinct methods: a Chrome extension that monitored users’ short videos from YouTube or Instagram (Reels)
• custom-built app that presented users with a stratified sample of videos and randomly
  selected content for viewing.
• In both methods, the primary measure considered was whether the user ”liked” the content.

Academic Stress Prediction using
Short-Form Video Content

Data Collection

• We collected short-form video content through two distinct methods: a Chrome extension that monitored users’ short videos from YouTube or Instagram (Reels)
• custom-built app that presented users with a stratified sample of videos and randomly
  selected content for viewing.
• In both methods, the primary measure considered was whether the user ”liked” the content.

Academic Stress Prediction using
Short-Form Video Content

Academic Stress Prediction using
Short-Form Video Content

Materials

• We employed the materials proposed in Flynn et. al, which introduced the Academic Stressor Measure (ASM).
• Additionally, we utilized the 10-Item Personality Inventory (TIPI) .
• The ASM considered factors including academics, finances, important assignments, time management, and faculty/staff.
• The TIPI measured students’ big-five personality traits:conscientiousness, emotional stability, openness, extraversion, and agreeableness.

Academic Stress Prediction using
Short-Form Video Content

Materials

ASM

Academic Stress Prediction using
Short-Form Video Content

Materials

TIPI

I see myself as:

1. ___ Extraverted, enthusiastic.

2. ___ Critical, quarrelsome.

3. ___ Dependable, self-disciplined.

4. ___ Anxious, easily upset.

5. ___ Open to new experiences, complex.

6. ___ Reserved, quiet.

7. ___ Sympathetic, warm.

8. ___ Disorganized, careless.

9. ___ Calm, emotionally stable.

10. ___ Conventional, uncreative.

TIPI scale scoring (“R” denotes reverse-scored items): Extraversion: 1, 6R; Agreeableness: 2R, 7; Conscientiousness; 3, 8R; Emotional Stability: 4R, 9; Openness to Experiences: 5, 10R.

Academic Stress Prediction using
Short-Form Video Content

Materials

• With Help of the representation built in the previous section, the videos they watched were classified into different affective states:

affection, anger, confidence, confusion,

embarrassment,fear, happy, kindness, neutral, sad, surprise

The representations were classified using Neural Tangent Kernel (NTK)

Academic Stress Prediction using
Short-Form Video Content

Materials

The representations were classified using Neural Tangent Kernel (NTK)

Neural Tangent Kernel-based Kernel Regression

Academic Stress Prediction using
Short-Form Video Content

Materials

Now We have the following:

• Stressors and Personality traits measured by the questionnaires
• Number of Videos liked per affective state. 
• Now our goal is to find if there are some associations to the type of video consumed and stressors and personality traits

Academic Stress Prediction using
Short-Form Video Content

Results

Academic Stress Prediction using
Short-Form Video Content

There Might be Spurious Correlations. We try to do Causal Structure Learning

Results

Academic Stress Prediction using
Short-Form Video Content

There Might be Spurious Correlations. We try to do Causal Structure Learning

Results

Academic Stress Prediction using
Short-Form Video Content

interesting causal relationships are observed, including:

• Financial stress and conscientiousness appear to cause consumption of sad content.
• Academic stress and agreeableness seem to cause consumption of content with
embarrassment as an affective state.
• Academic stress, with positive emotional stability, appears to cause consumption
of content with fear as an affective state.

Results

Limitations

• Although we observe interesting associations and signs of relationships between content’s affective states and measured stressors, we note that our sample size is severely limited (n=22), which may not represent a clean sample representative of the population.
   
• There are also possibilities of other aspects of short videos that are associated with
  these stressors, and a multitude of other stressors should also be considered, such as
  family pressure, peer pressure, relationship issues, etc

Limitations

• In future studies, the graphical
  model might be driven by Causal Representation Learning,

Future Directions

• Larger and Representative Population.
• Causal Representation Learning
•  Focus on interventions to be taken once stressors are
  measured. For instance, recommending videos to alleviate specific stressors or build-
  ing an empathetic chatbot where individuals can express their emotions.

"Love is the only medicine that can heal the wounds of the world"

Mata Amritanandamayi

"Love is the only medicine that can heal the wounds of the world"

Mata Amritanandamayi

Build Technology. Empethetically.