bangdedadi/understanding-the-effect-of-opinion-polarization-in-short-video-browsing

Project Overview

This project encompasses a comprehensive dataset and associated code, aiming to analyze and predict exposure to opinion polarization using EEG data, behavior signals, and questionnaire responses.

Data Folder Overview

This folder contains three key datasets: behavio_signals, EEG_data, and questionnaire.

Behavio_Signals Dataset Overview

• Content: Data on likes and viewing duration for each video by participants during both pre-study and post-study phases.
• Privacy: Uses "member_id" to represent users for privacy protection.
• Data Columns:
  • Like Column: A checkmark (✓) for a like, and a blank space for no like.
  • Viewing Time Column: Time format is hours:minutes:seconds.
• Video Title Information: Includes "LAB2" for post-study, "LAB1" for pre-study, and terms like "neutral," "fear," "happy," and "sad" for video types from SEED-IV dataset.

Questionnaire Data Overview

• Content: Responses from participants in a Likert scale format.
• Structure:
  • 20 Questions: On sentiments and familiarity with 10 different personages.
  • Distractor Questions: Unrelated to the personages.
• Consistency and Aim: Reflects changes in sentiments towards personages after the field study.
• Bias Elimination: Three different randomized questionnaire setups (A, B, and C).

EEG Data Overview

• Content: EEG data of 23 participants from both pre-study and post-study phases.
• Data Size and Accessibility: Total size nearly 40GB; one participant's data provided as an example.
• Future Access: Post-review, a cloud drive link with the full dataset will be added to GitHub.

The Annotation Guideline

Please rate the character's opinion in the video on a scale from 1 to 5, reflecting its polarity. A score of 1 corresponds to a pessimistic evaluation, while a score of 5 signifies an optimistic evaluation. Scores 2 and 4 represent moderately negative and positive evaluations, respectively, and a score of 3 denotes neutrality.

Code Documentation

The code is provided in Python and Jupyter Notebook formats.

Preprocessing Code

• Purpose: Aligns EEG data with web-collected timestamps and extracts data for events.

DE_Feature Code

• Purpose: Extracts Differential Entropy (DE) features.
• Frequency Bands: Includes delta, theta, alpha, beta, and gamma.

Draw_Topomap Code

• Purpose: Calculates correlation between exposure to Opinion Polarization (OP) and EEG signals.
• Visualization: Plots an electroencephalogram to visualize correlation.

OP_Detection_Model Code

• Purpose: Uses EEG data, behavior signals, and questionnaire data for training a model.
• Application: Predicting exposure to opinion polarization.