Personalized Emails Generation (OpenAI, LangChain)

This application leverages OpenAI and LangChain functionalities to extract structured data from unstructured text, specifically customer comments on received products. The processed data is then used to cluster comments into distinct segments, enabling the creation of personalized email campaigns. These targeted emails are designed to promote a new product, enhancing marketing effectiveness and customer engagement.

The current application is not to be used for direct integration into a production environment. Instead, it serves as a demonstration of how OpenAI models can be leveraged to customize communications based on historical events and data.

Purpose

In today's competitive market, effective communication between businesses and their customers is essential, especially when introducing new products or services. Tailoring these messages to reflect the unique relationship a customer has with a business can significantly enhance engagement. This project explores a sequence of steps that leverage customer comments to classify their acceptance and satisfaction levels with the company's products. By analyzing these insights, businesses can generate personalized promotional content when launching new products. This approach improves communication by aligning it more closely with customers' past experiences and perceptions, thereby fostering a more meaningful connection.

Process Overview

The generation of emails contents follows the below steps:

Features

• Source Data: The application uses as source data file the Amazon Fine Food Reviews (https://www.kaggle.com/datasets/snap/amazon-fine-food-reviews) This dataset comprises reviews of fine foods sourced from Amazon. The most recent 1,000 records have been selected for this analysis. The fields labeled Summary and Text have been combined into a new field titled Comment, which is to be utilized in the labeling process by the LLM.

• Labeling process: The execution of PersonalizedEmails_LABELING.py invokes OpenAI LLM model to extract from each Comment a required data structure:

  • product: as the product that the buyer is commenting on
  • sentiment: as the sentiment expressed in the text, categorized as: 'very negative,' 'negative,' 'neutral,' 'positive,' or 'very positive
  • aggressiveness: as the level of aggressiveness in the text, rated on a scale from 1 to 10
  • satisfaction: as the user's satisfaction with the product, rated on a scale from 1 to 5
  • confidence: as the confidence level in the extracted properties, rated on a scale from 1 to 5. The confidence level is used to filter the data: records with low confidence are excluded from the next steps of the process (Clustering).
    The process saves the results as intermediate_data/Amazon_Reviews_Labeled.csv

• Clustering: The execution of the PersonalizedEmails_CLUSTERING.py excludes records with customers' feedback discrepancies (i.e. when the Score given by a customer is quite different from the sentiment of satisfaction level addressed by the LLM). The kMeans model is used to cluster the data and the results are saved in the intermediate_data/Amazon_Reviews_Clustered.csv file. The Jupyter file PersonalizedEmails_CLUSTERING.ipynb is also available as it contains useful visualization outputs.

• Text Generation: The execution of the PersonalizedEmails_Text_Generation.py script processes the output from the clustering process by mapping the numerical values of each cluster, specifically the mean values, to corresponding categorical values. These categorical values are then utilized to create a prompt for each cluster, which is subsequently used to generate email content with the use of the LLM. The generated email content is then saved in the file results/AmazonReviews_WithEmails.csv. In the results directory there is also the file Sample of generated Personalized Emails.txt with the result of the Text Generation process.

• How to run the app:

• clone the repository: https://github.com/antdragiotis/Personalize-emails-LangChain-

• change current directory to cloned repository

• pip install -r requirements.txt

• it is assumed that your system has been configured with the OPENAI_API_KEY, otherwise you need to add the following statements to the python code files:

  • import os
  • os.environ["OPENAI_API_KEY"] = "your-openai-api-key"

• run the three Python files as the processes described above:

  • PersonalizedEmails_LABELING.py
  • PersonalizedEmails_CLUSTERING.py
  • PersonalizedEmails_Text_Generation.py

You get the results in the results/AmazonReviews_WithEmais.csv file

Project Structure

• *.py: main application code
• source_data: directory with the source data file
• intermediate_data: directory with intermediate data files that facilitate control and validations
• results: directory with the final output file
• README.md: project documentation