Skip to content

Latest commit

 

History

History
164 lines (108 loc) · 4.83 KB

tutorial.md

File metadata and controls

164 lines (108 loc) · 4.83 KB

🚢 More Tutorials

👨‍👨‍👧‍👦 User Generation

🟠 Reddit User Profile Generation

We utilized the findings from a survey on the distribution of real Reddit users (https://explodingtopics.com/blog/reddit-users) as a reference to randomly generate demographic information for our users. Leveraging this data, we then employed GPT to create more detailed descriptions.

  • Step 1:

Modify your OpenAI API key at the top of the generator\reddit\user_generate.py file.

client = OpenAI(api_key='sk-xxx')
  • Step 2:

Change the number of users you want to generate and the path where the user data will be saved at the bottom of the generator\reddit\user_generate.py file.

if __name__ == "__main__":
    N = 10000  # Target user number
    user_data = generate_user_data(N)
    output_path = 'user_data_10000.json'
    save_user_data(user_data, output_path)
    print(f"Generated {N} user profiles and saved to {output_path}")
  • Step 3:

Run the Script:

python generator/reddit/user_generate.py

🔵 Twitter User Profile Generation

The Twitter users are further enhanced based on the information we have crawled. The steps to run are as follows:

pip install -r generator/twitter/requirement.txt

python generator/twitter/gen.py

python generator/twitter/network.py

Since the repository does not contain information from the actual Twitter in complete_user_char.csv and new_stars.csv, it might not run directly and is for reference only now.

📊 Data Visualization

🟠 Reddit Sore Analysis

  • Step 1:

After running python scripts/reddit_simulation_align_with_human/reddit_simulation_align_with_human.py, a database file and a JSON file with the same name will be generated. Modify visualization/reddit_simulation_align_with_human/code/analysis_all.py here to their respective paths and the common filename (excluding the extension).

if __name__ == "__main__":
    folder_path = ("visualization/reddit_simulation_align_with_human"
                   "/experiment_results")
    exp_name = "business_3600"
    db_path = folder_path + f"/{exp_name}.db"
    exp_info_file_path = folder_path + f"/{exp_name}.json"
    analysis_score.main(exp_info_file_path, db_path, exp_name, folder_path)
  • Step 2:
pip install matplotlib
  • Step 3:

Run the Script:

python visualization/reddit_simulation_align_with_human/code/analysis_all.py

Then, you will see the scores of the three groups—down-treated, control, up-treated at the end of the experiment, like this

🟠 Reddit Counterfactual Content Analysis

  • Step 1:

First, you need to add your OpenAI API key to the system's environment variables.

  • Step 2:

After running python scripts/reddit_simulation_counterfactual/reddit_simulation_counterfactual.py, 3 database files will be generated. Modify visualization/reddit_simulation_counterfactual/code/analysis_couterfact.py here to their respective paths.

db_files = [
    'couterfact_up_100.db',
    'couterfact_cnotrol_100.db',
    'couterfact_down_100.db'
]
  • Step 3:
pip install aiohttp
  • Step 4:

Run the Script:

python visualization/reddit_simulation_counterfactual/code/analysis_couterfact.py

Then, you will see the disagree scores of the three groups—down-treated, control, up-treated in each timestep of the experiment, like this

🟣 Dynamic Follow Network Visualization

  • Step 1:

Visit https://neo4j.com/ and register for an account, then create a free instance. Obtain NEO4J_URI, NEO4J_USERNAME, and NEO4J_PASSWORD and save them in your environment variables. Open the instance and connect.

  • Step 2:
pip install neo4j
  • Step 3:

After running python the simulation script, a database file will be generated. Modify visualization/dynamic_follow_network/code/vis_neo4j_reddit.py or visualization/dynamic_follow_network/code/vis_neo4j_twitter.py here to their respective paths.

if __name__ == "__main__":
    sqlite_db_path = "all_360_follow.db"  # Replace with your SQLite database path
    main(sqlite_db_path)

And run the Script.

python visualization/dynamic_follow_network/code/vis_neo4j_reddit.py
# or
python visualization/dynamic_follow_network/code/vis_neo4j_twitter.py
  • Step 4:

Use the explore page of the https://console.neo4j.io/ dashboard. In the search bar, select user-follow-user. For the slicer, choose follow-timestamp to visualize changes in follow relationships. Like this,

network.mp4