OpenAI JSON Text Transformer

This PHP library enables the transformation of texts using the OpenAI API in JSON mode. The library uses a multi-stage process to generate new, structured content from existing texts.

How It Works

The library performs the following steps:

1. Creating an Outline: Analyzes the input text and creates a structured outline.
2. Generating a Title: Creates a suitable title based on the outline.
3. Generating an Introduction: Creates an introductory summary.
4. Generating Sections: Creates each section and subsection based on the outline.
5. Compiling the Result: Combines all generated parts into a structured JSON document.

Requirements

• PHP 8.2 or higher
• cURL extension for PHP
• OpenAI API key

Installation

1. Clone or download this repository
2. Copy config.example.php to config.php and adjust the configuration
3. Make sure your OpenAI API key is entered in the configuration file

Configuration

The configuration file config.php contains the following sections:

API Settings

'api' => [
    'key' => 'YOUR_OPENAI_API_KEY_HERE', // OpenAI API key
    'model' => 'gpt-4o-mini', // Model for text generation
    'temperature' => 0.7, // Standard temperature (0.0 - 1.0)
    'presence_penalty' => 0.0, // Penalty for new topics (-2.0 - 2.0)
    'frequency_penalty' => 0.0, // Penalty for repetitions (-2.0 - 2.0)
    'max_retries' => 5, // Maximum number of retry attempts for errors
    'retry_delay' => 1, // Wait time in seconds between retry attempts
],

Specific Temperature Settings

'temperatures' => [
    'outline' => 0.7,  // Temperature for creating the outline
    'title' => 0.8,    // Temperature for titles (higher for more creative headlines)
    'introduction' => 0.7, // Temperature for introductions
    'section' => 0.6,   // Temperature for text sections (lower for more factual text)
],

You can adjust these temperatures as needed:

• Higher values (0.8-1.0) result in more creative, surprising texts
• Lower values (0.2-0.5) result in more focused, deterministic texts
• Medium values (0.6-0.7) provide a good balance

System Roles

'roles' => [
    'outline' => 'You are an experienced editor who creates clearly structured outlines.',
    'title' => 'You are an experienced editor who creates concise and appealing titles.',
    'introduction' => 'You are an experienced editor who writes captivating introductions.',
    'section' => 'You are an experienced editor who writes informative and well-structured sections.'
],

These system roles are provided to the AI model for each request and influence the style and tone of the generated texts:

• You can adapt the roles to your specific requirements
• For more technical texts, you could use "You are a technical writer who..." for example
• For more creative texts, "You are a creative writer who..." might be more appropriate

Logging Settings

'logging' => [
    'enabled' => true, // Enable/disable logging
    'path' => __DIR__ . '/logs/api.log', // Path to log file
    'level' => 'DEBUG', // Log level: DEBUG, INFO, WARNING, ERROR
],

Prompts

The prompts determine how OpenAI rewrites the texts. You can adjust them to suit your needs:

• outline: Prompt for creating the outline
• title: Prompt for generating the title
• introduction: Prompt for generating the introduction
• section: Prompt for generating the sections

Usage

The following example shows how to use the library:

// Load the required classes
use OpenAIJsonTransformer\Utils\ConfigManager;
use OpenAIJsonTransformer\Utils\Logger;
use OpenAIJsonTransformer\OpenAI\OpenAIClient;
use OpenAIJsonTransformer\OpenAI\TextTransformer;

// Initialize the components
$configManager = new ConfigManager('config.php');
$logger = new Logger($configManager->getLoggingConfig());
$openaiClient = new OpenAIClient($configManager->getApiConfig(), $logger);
$textTransformer = new TextTransformer($openaiClient, $configManager, $logger);

// Load articles from a JSON file
$articlesJson = file_get_contents('texte.json');
$articles = json_decode($articlesJson, true);

// Transform the articles
$result = $textTransformer->transformArticles($articles);

// Save the result
file_put_contents('output.json', json_encode($result, JSON_PRETTY_PRINT));

Example Script

In the examples folder you will find a complete example script:

php examples/transform_article.php

Input Format

The library expects articles in the following JSON format:

{
  "articles": [
    {
      "title": "Article Title",
      "text": "Article Text..."
    },
    {
      "title": "Another Article Title",
      "text": "Another Article Text..."
    }
  ]
}

Output Format

The library produces JSON in the following format:

{
  "title": "Generated Main Title",
  "introduction": "Generated Introduction...",
  "sections": [
    {
      "heading": "Section Heading",
      "content": "Section Content..."
    },
    {
      "heading": "Section Heading with Subsections",
      "content": "Section Content...",
      "subsections": [
        {
          "heading": "Subsection Heading",
          "content": "Subsection Content..."
        }
      ]
    }
  ]
}

Error Handling

The library implements robust error handling with automatic retry attempts for API errors. All errors are logged in the log file.

Extensions

The library is designed to be easily extended:

• Additional LLM Providers: The structure allows for the integration of other providers such as Anthropic.
• Additional Transformations: The transformation system can be extended with additional text processing functions.

License

This project is licensed under the MIT License.

Language

A German version of this documentation is available in LIESMICH.md.