Use Case
Unlock insights of your – Retail Partner Management – email campaigns! Our recent analysis delved into sentiment and content clusters, revealing a majority of neutral tones with a significant positive undertone in subject lines. By clustering, we identified key themes: Branding, Offers, Encouragement, Urgency, and Sales.
Instructions for Custom GPT
This specialized GPT is tailored to support digital content marketing analytics tasks, offering user-friendly features while ensuring the use of publicly available data for privacy concerns. It leverages popular libraries such as TensorFlow, PyTorch, and Hugging Face Transformers when applicable and accessible. Here’s how it performs the specified tasks:
Content Upload: Users can conveniently upload content in various formats, including CSV, XLSX, XML, PPTX, DOCX, or PDF.
Keyword Clustering: To enhance content analysis, the system employs TF-IDF and KMeans algorithms to cluster keywords effectively.
Keyword Extraction and Aggregation: It extracts and aggregates keywords from all clusters, allowing users to gain valuable insights from their content.
Mapping to Themes: The GPT maps content clusters to themes based on the keywords present, providing a clear understanding of the underlying topics.
Text Preprocessing: Non-alphabetic characters are removed, and content is split into words while excluding stopwords, ensuring data quality.
BERT Embeddings: The system generates BERT embeddings for content analysis, enhancing the accuracy of results.
Data Visualization: For clear data presentation, the GPT utilizes data visualization libraries like Matplotlib and Seaborn, making insights easily accessible.
Downloadable Data: Users can download all data and insights in appropriate formats, ensuring accessibility and transparency.
- Step 1: Take a data set of around 800 email subjects of a leading global fashion and lifestyle company in the premium segment, extraction timeframe between 2019 and 2020.
- Step 2: Create a Custom GPT and provide detailed instructions.
- Step 3: Save and confirm Custom GPT, explore, analyze, and visualize data set.
Summary of prompt conversation
Dataset Parsing Issue and Resolution: Initially, there was a problem loading the dataset due to a parsing error. This issue was resolved by identifying and using a semicolon (;) as the correct delimiter, revealing a dataset with columns for Date, Received From, and Subject.
Sentiment Analysis: The email subjects underwent sentiment analysis, classifying them as positive, negative, or neutral based on their sentiment polarity. The results indicated a predominance of neutral sentiments, followed by positive, with a few negative sentiments.
Sentiment Distribution Visualization and Dataset Update: The sentiment distribution was visualized, and the dataset was updated to include columns for Sentiment_Weight and Sentiment_Category. The visualizations provided a clear understanding of the overall emotional tone of the email subjects.
Emails Sent by Weekday Analysis: A bar chart was created to show the distribution of emails sent across different days of the week, highlighting the most and least active days for email communications.
Content Clustering: Utilizing a simplified text preprocessing approach and KMeans clustering, the email subjects were grouped into content clusters based on thematic similarities. The clusters were defined by their predominant themes, such as General Branding and Seasonal, Exclusive Offers, Purchase Encouragement, Urgency and Interest, and Discounts and Sales.
Visualization of Content Clusters: The distribution of email subjects across the content clusters was visualized through bar charts, both in absolute numbers and percentages, providing insights into the focus areas of the analyzed email subjects.
1
