Résumé et développement de texte

Ingénierie des prompts avec l'API OpenAI

Fouad Trad

Machine Learning Engineer

Résumé de texte

Condense le texte dans un format plus court
Optimise les processus opérationnels
- Finance -> résume des rapports détaillés
- Marketing -> transforme les commentaires des clients en informations exploitables
Les LLM peuvent résumer des textes à l'aide de prompts efficaces

Image showing how summarization changes a pile of books into a document containing a summary.

Prompt inefficace

Indique uniquement le texte à résumer

text = "I recently purchased your XYZ Smart Watch and wanted to provide some feedback
based on my experience with the product. I must say that I'm impressed with the sleek design 
and build quality of the watch. It feels comfortable on the wrist and looks great with any 
outfit. The touchscreen is responsive and easy to navigate through the various features."


prompt = f"""Summarize the text delimited by triple backticks:
           ```{text}```"""
print(get_response(prompt))

The author purchased the XYZ Smart Watch and is impressed with its sleek design, build 
quality, and comfortable fit on the wrist. They find the touchscreen responsive and 
user-friendly for navigating the watch's features.

Amélioration du prompt

Limites de résultat
Structure de résultat
Accent sur la synthèse

Image showing a person moving from a step to a higher one as a sign of improvement.

Prompt efficace : limites de résultat

Indiquer le nombre de phrases, de mots et de caractères

text = "I recently purchased your XYZ Smart Watch and wanted to provide some feedback
based on my experience with the product. I must say that I'm impressed with the sleek design 
and build quality of the watch. It feels comfortable on the wrist and looks great with any 
outfit. The touchscreen is responsive and easy to navigate through the various features."


prompt = f"""Summarize the text delimited by triple backticks in one sentence:
           ```{text}```"""
print(get_response(prompt))

The customer is impressed with the sleek design, build quality, comfort, and responsiveness 
of the XYZ Smart Watch's touch screen.

Prompt efficace : structure du résultat

Préciser la structure du résultat

text = "I recently purchased your XYZ Smart Watch and wanted to provide some feedback
based on my experience with the product. I must say that I'm impressed with the sleek design 
and build quality of the watch. It feels comfortable on the wrist and looks great with any 
outfit. The touchscreen is responsive and easy to navigate through the various features."


prompt = f"""Summarize the text delimited by triple backticks, in at most three bullet points.
           ```{text}```"""
print(get_response(prompt))

- The XYZ Smart Watch has a sleek and impressive design with  excellent build quality.
- It feels comfortable on the wrist and complements any outfit.
- The touch screen is responsive and user-friendly for easy navigation through the features.

Prompt efficace : focus sur la synthèse

Demander au modèle de se concentrer sur des parties spécifiques du texte

text = "I recently purchased your XYZ Smart Watch and wanted to provide some feedback
based on my experience with the product. I must say that I'm impressed with the sleek design 
and build quality of the watch. It feels comfortable on the wrist and looks great with any 
outfit. The touchscreen is responsive and easy to navigate through the various features."


prompt = f"""Summarize the review delimited by triple backticks, in three sentences, 
focusing on the key features and user experience:
           ```{text}```"""
print(get_response(prompt))

Prompt efficace : focus sur la synthèse

The customer purchased the XYZ Smart Watch and was impressed with its sleek design 
and build quality. 
They found it comfortable to wear and versatile enough to match any outfit. 
The touch screen was responsive and user-friendly, making it easy to navigate 
through the watch's features.

Expansion de texte

Génère du texte à partir d'idées ou de listes à puces
Améliore l'efficacité et la productivité
Les LLM peuvent développer des textes à l'aide de prompts bien conçus

Image explaining that text expansion is about generating a full text based on some small ideas of requirements.

Prompts d'expansion de texte

Demander au modèle d'étendre le texte délimité
Mettre en évidence les aspects sur lesquels se concentrer
Indiquer les exigences relatives au résultat (ton, longueur, structure, public)

Icon showing a pencil writing on a notebook.

Développement de la description du service

service_description = """Service: Social XYZ
- Social Media Strategy Development
- Content Creation and Posting 
- Audience Engagement and Community Building
- Increased Brand Visibility
- Enhanced Customer Engagement
- Data-Driven Marketing Decisions"""

prompt = f"""Expand the description for the Social XYZ service delimited by triple 
backticks to provide an overview of its features and benefits, without bypassing 
the limit of two sentences. Use a professional tone.
```{service_description}```"""
print(get_response(prompt))

Développement de la description du service

Social XYZ is a comprehensive social media service that offers strategic 
development, content creation, and posting to help businesses effectively engage 
with their target audience and build a strong online community. 

With a focus on increasing brand visibility and enhancing customer engagement, 
Social XYZ enables businesses to make data-driven marketing decisions for optimal 
results.

Passons à la pratique !

Ingénierie des prompts avec l'API OpenAI