Chatbot Data Analytics

Analysis of 2.8M Twitter chatbot interactions from customer service conversations.

Overview

This project analyzes chatbot data to identify communication patterns between customers and support bots. Built during an internship with Open Avenues and Businessolver (Sept-Oct 2022).

Key Findings

Volume Growth Tweet volume increased by 4 orders of magnitude between Q4 2016 and Q4 2017.

Top Contributors

• AmazonHelp: 169,840 tweets (most active bot)
• AppleSupport: 106,860 tweets
• Uber_Support: 56,270 tweets
• 700,000 unique tweeters total

Temporal Patterns

• 90% of tweets occurred in Q4 2017
• Weekdays dominated in 2017 (vs Saturdays in 2016)
• Both years showed spikes in early November

Content Reduction Text cleaning removed 40% of original content (590MB → 358MB) by eliminating URLs, punctuation, stop words, and emojis.

Technologies

• Python
• Pandas (data manipulation)
• Matplotlib (visualization)
• Jupyter Notebook

Data Structure

Dataset contains 7 columns:

• tweet_id: Unique identifier
• author_id: Sender address
• inbound: Boolean (True = to bot, False = from bot)
• created_at: Timestamp
• text: Tweet content
• response_tweet_id: Original tweet ID
• in_response_to_tweet_id: Conversation starter ID

Analysis Pipeline

1. Load CSV data
2. Clean text (lowercase, remove URLs/punctuation/stop words/emojis)
3. Extract temporal patterns
4. Identify top contributors
5. Compare year-over-year trends
6. Visualize findings

Author

Steve Eckardt
September 2022 - October 2022
Analysis and evaluation of chatbot data, Open Avenues and Businessolver Internship
This project is for educational and portfolio purposes.