Introduction
The Marathi Retail & E-Commerce Chat Dataset is a large-scale, high-quality collection of over 12,000 chat conversations between customers and call center agents, focused exclusively on Retail and E-Commerce domains. Designed to reflect real-world service interactions, this dataset supports the development of robust conversational AI and NLP models tailored for Marathi-speaking audiences.
Participant & Chat Overview
•
Contributors:
200 native Marathi speakers from the FutureBeeAI Crowd Community
•
Chat Length:
300–700 words per conversation
•
Turn Count:
50–150 dialogue turns across both participants
•
Chat Types:
Inbound and outbound
•
Sentiment Coverage:
Positive, neutral, and negative interaction outcomes
Topic Diversity
This dataset spans a wide range of Retail and E-Commerce conversation types:
•Inbound Chats (Customer-Initiated)
•Return or exchange requests
•Refunds and payment issues
•Membership or subscription queries
•Shipping, delivery, and more
•Outbound Chats (Agent-Initiated)
•Order confirmation and verification
•Cross-selling and upselling
•Loyalty program promotions
•Special offers and discounts
•Customer feedback and verification
This diversity enables training of models that handle varied intents, scenarios, and outcomes within customer service workflows.
Language Nuance & Realism
The dataset is rich in linguistic diversity and mirrors real conversational tone and structure used in Marathi-speaking regions:
•
Personal & Brand Names:
Culturally accurate naming conventions
•
Local Elements:
Realistic addresses, phone numbers, emails, currency references, and time/date formats
•
Slang & Idioms:
Local expressions, informal phrases, and customer service jargon
•
Cultural Specificity:
Region-aware vocabulary and tone
This linguistic authenticity ensures the development of culturally fluent AI models for Marathi Retail & E-Commerce use cases.
Conversational Structure & Flow
The conversations reflect natural dialogue dynamics and are organized into various types of interaction styles:
•Detailed problem-solving discussions
•Follow-ups and status updates
•Advisory and assistance sessions
Each conversation includes common dialogue stages such as:
This structured flow helps train models to manage real-world service interactions from start to finish.
Data Format & Structure
Available in TXT, CSV, and JSON formats, each conversation is structured with fields such as:
•Message timestamps (if needed)
This flexible formatting ensures compatibility with major NLP frameworks and workflows.
Applications
This dataset can be used across a wide range of commercial and research applications:
•Retail Chatbots & Voicebots
•Customer Service Automation
•Sentiment Analysis & Intent Detection
•NER & Information Extraction
•Text Generation & Prediction Models
•Multilingual Retail NLP Research
•Domain-specific Smart Assistants
Secure & Ethical Collection
•
Informed Consent:
All participants contributed with full awareness and written consent
•
Privacy Protected:
No personally identifiable information (PII) is included
•
Secure Pipeline:
All data was collected, processed, and stored within FutureBeeAI’s secure platform environment
Updates & Customization
This dataset is regularly updated with fresh conversations and offers extensive customization capabilities:
•
Annotation Options:
Add NER, sentiment, intent, or custom tags
•
Topic Expansion:
Custom chat collection for specific product categories or services
•
Language Flexibility:
Custom chat datasets available in Marathi or other languages upon request
Licensing
This dataset is developed and owned by FutureBeeAI and is available for commercial licensing. Flexible licensing options can be provided for enterprises, academic researchers, and solution providers.
