This is a template repo for your project to help you organise and document your code better. Please use this structure for your project and document the installation, usage and structure as below.
In this project, we aim to explore strategies to enhance the popularity of a YouTube channel using the Youniverse dataset. Our analysis focuses on two key aspects: the impact of video category diversity and the optimal video duration. Specifically, we investigate how these factors correlate with key engagement metrics, including the number of views, likes, dislikes, comments, the growth number of subscribers and views. To better capture video popularity, we also examine and propose methods for quantifying the degree of popularity. This comprehensive approach seeks to provide actionable insights for content creators to optimize their video strategies and achieve greater audience engagement.
- How does video category diversity impact a YouTube channel? Will a more diverse video category be better, or will a more focused category be more popular?
- What is the optimal video duration for a YouTube video? How does video duration correlate with key engagement metrics? Does the optimal duration vary across different video categories?
- How can we quantify the degree of popularity of a YouTube video? What are the key factors that contribute to video popularity? How can we measure if the degree of the popularity is precise?
-
Sample Data: Considering the limitations of computing resources. We select a subset of YouTube channels from df_channels_en.tsv. The sample dataset includes channel_id with varying video category diversity and have the same proportion of video categories as the original dataset. Then, we use these channel_ids to filter data from df_timeseries_en.tsv, num_comments.tsv and yt_metadata_helper.feather.
-
Data Preprocessing: We identified missing or NaN values in each data frame and addressed these by imputing values. Specifically, missing values were filled using the mean computed per 'channel_id' to maintain consistency across related entries.
-
Statistical Analysis:
a) Mean, median and standard deviation of different inputs are calculated for further comparison and analysis.
b) The Shannon Entropy Formula is used to calculate the diversity of a channel's content based on the categories of its videos. A higher entropy value indicates that the content is more evenly distributed across categories. A lower entropy value means the content is more concentrated in a few categories, indicating less diversity.
c) Analysis of Variance : To explore the relationships between diversity and performance within different categories, relevant hypotheses are proposed, and then the corresponding F-values and p-values are calculated to categorize whether content strategy (Diversified, Mixed, Specialized) has a noticeable impact on metrics such as view count, engagement rate, and subscriber growth.
-
Metrics Calculation:
a) The Engagement Rate is calculated as the ratio of like_count to view_count.
b) The Popularity Score is calculated based on the weights of different normalized factors to reflect overall video performance.
-
Visualization:
a) Scatter Plot: The plots explore the relationship between content diversity and average view_count for each channel, the relationship between content diversity and engagement rate, the relationship between content diversity and weekly subscriber growth and the relationship between content diversity and weekly view growth. Logarithmic Scale is used for better visualization.
b) Bar chart: Bar charts are used to show the mean view count, mean weekly subscriber growth, mean weekly view growth and mean engagement rate of different content strategies.
c) Line plot: Optimal video length for different categories are shown in line plot by different metrics. Also, the average popularity score over time is plotted for each video duration category.
d) Box plot: Several box plots are generated to visualize how different the video performances vary by video duration categories.
-
16.11.2024-30.11.2024
Diversity Outlier Investigation.
Analyze highly successful diversified channels, study their unique strategies and characteristics.
-
30.11.2024-06.12.2024
Further analyze the data, especially in conjunction with the differentiation of audience preferences for videos of different durations from 2016 to 2019.
Find external reasons that lead to YouTube viewers' preference for duration.
-
07.12.2024-13.12.2024
Build a basic framework to tell the data story.
Relate the content strategies with the analysis of duration preferences for better storytelling perspective.
-
14.12.2024-20.12.2024
Select relevant data and visualization tools, complete and improve the final story presentation.
- Shengze: Diversity Outlier Investigation.
- Shuhua: Study the impact of events on preference duration
- Xuanrui: Frequency domain analysis, time series fitting and forecasting
- Xinran: Relate the content strategies with the analysis of duration preferences
- Xinyue: Build a framework to tell the data story
# clone project
git clone <[email protected]:epfl-ada/ada-2024-project-laughtodeath666.git>
cd <project repo>
# [OPTIONAL] create conda environment
conda create -n <env_name> python=3.9
conda activate <env_name>
# install requirements
pip install -r pip_requirements.txtDownload the dataset from this following link and add the data files to the directory 'data'.
The directory structure of this project looks like this:
├── data <- Project data files
│
├── src/ <- Source code
│ ├── data/ <- Data loading and preprocessing
│ │ ├── sample_data.ipynb <- Sample data generation
│ │ ├── loader.py <- Data loading functions
│ │ └── preprocessor.py <- Data preprocessing functions
│ │
│ ├── utils/ <- Analysis utilities
│ │ ├── content.py <- Content diversity strategy analysis
│ │ └── duration.py <- Video duration analysis
│
├── results.ipynb <- a well-structured notebook showing the results
│
├── .gitignore <- List of files ignored by git
├── pip_requirements.txt <- File for installing python dependencies
└── README.md