Effectiveness of Customer Support Interactions on Twitter.

• Tech Stack: Python, Machine Learning, Natural Language Processing, Latent Dirichlet Allocation, Social Computing
• GitHub URL: Project Link
• Report URL: Report Link

The central question of the research is, "How effective are customer support interactions on Twitter, and what patterns of communication contribute to successful resolutions?" This inquiry is novel in its comprehensive exploration of various dimensions of digital customer support on a major social media platform. Unlike traditional studies that primarily focus on generic customer support metrics, this project delves deep into the multifaceted nature of interactions on Twitter, a platform where public visibility adds a unique dimension to customer service.

The chosen dataset is from Kaggle, titled "Customer Support on Twitter." It contains over three million tweets and replies from some of the most prominent brands on Twitter. The dataset includes variables such as tweet ID, author ID, creation time, text of the tweet, and related response IDs. This dataset is suitable because it offers a comprehensive view of customer-brand interactions, allowing for a detailed analysis of support effectiveness, user engagement patterns, and communication styles.

Performed sentiment analysis and LDA on an extensive 3 million tweets Twitter dataset to dissect customer satisfaction, interaction patterns, and brand response times. Delivered strategic insights into customer behavior, enriching digital customer service strategies.

For the project, I tried different combinations of the three datasets provided by using them to train and evaluate different models like average perceptron, SVM, logistic regression, boosted perceptron and ensemble.