organize a list of icons using an LLM

icon_categorizer.py

# Importing necessary libraries
# To install the dependencies, use the following commands:
# pip install openai
# pip install pandas

import json  # Built-in library, no installation required
import openai  # For interacting with OpenAI's API
import pandas as pd  # For data manipulation and analysis

client = openai.OpenAI()
# client.api_key = "" # uncomment and add key if you dont have OPENAI_API_KEY set in your environment

# optional set of seed categories to influence category generation
categories = {"Weather", "Connectivity"}

# load icons.txt and split by comma
with open("icons.txt", "r") as file:
    icons = file.read().split(",")
    print(icons)


def get_icon_category(icon_name):
    system_prompt = """
You are an expert icon categorizer. You are given an icon name and you need to categorize it. You can categorize it as one of the following categories: {categories}

If no category is suitable, create a new category using the guidelines below:
- It should be a category that is related to the icon.
- It should be a category that is not already in the categories list.
- It should be a category that is not too specific.

Respond with a JSON object with the following format: {{"category": "<category>"}}
""".format(categories=", ".join(categories)).strip()

    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": system_prompt},
            # add few shot examples here for steering the model
            # {"role": "user", "content": "icon: Tree"},
            # {"role": "assistant", "content": "{\"category\": \"PLANT\"}"},
            # {"role": "user", "content": "icon: Car"},
            # {"role": "assistant", "content": "{\"category\": \"TRANSPORTATION\"}"},
            # {"role": "user", "content": "icon: Phone"},
            # {"role": "assistant", "content": "{\"category\": \"COMMUNICATION\"}"},
            {"role": "user", "content": f"icon: {icon_name}"}
        ],
        temperature=0,
        response_format={"type": "json_object"}
    )
    # load the json response
    response = json.loads(response.choices[0].message.content)
    return response["category"]


def get_icon_synonyms(icon_name):
    system_prompt = """
You are an expert writer. You are given an icon name and you need to create a list of keywords for it. Keywords are words that are related to the icon, and can be synonyms.

Respond with a JSON object with the following format: {"keywords": ["<keyword>", "<keyword>", ...]}
""".strip()
    response = client.chat.completions.create(
        model="gpt-4o",
        messages=[
            {"role": "system", "content": system_prompt},
            # add few shot examples here for steering the model
            # {"role": "user", "content": "icon: Tree"},
            # {"role": "assistant", "content": "{\"keywords\": [\"tree\", \"plant\"]}"},
            # {"role": "user", "content": "icon: Car"},
            # {"role": "assistant", "content": "{\"keywords\": [\"automobile\", \"transportation\"]}"},
            # {"role": "user", "content": "icon: Phone"},
            # {"role": "assistant", "content": "{\"keywords\": [\"telephone\", \"communication\"]}"},
            {"role": "user", "content": f"icon: {icon_name}"}
        ],
        temperature=0,
        response_format={"type": "json_object"}
    )
    response = json.loads(response.choices[0].message.content)
    return response["keywords"]


limit = 10  # limit the number of icons to process
# limit = len(icons) # uncomment to process all icons

data = []
for icon in icons[0:limit]:
    print(f"Processing icon: {icon}")

    category = get_icon_category(icon)
    print(f"Identified category for {icon}: {category}")

    synonyms = get_icon_synonyms(icon)
    print(f"Generated synonyms for {icon}: {synonyms}")

    # Add the new category to the set
    if category not in categories:
        print(f"Adding new category to the set: {category}")
        categories.add(category)

    # Append the data for the DataFrame
    data.append({"icon_name": icon,
                "category": category,
                "keywords": ", ".join(synonyms)
                })

# Create a DataFrame and write to a CSV
df = pd.DataFrame(data)
df.to_csv("icon_categories.csv", index=False)