User Feedback and Station Management with AI: Sentiment Analysis and Topic Modeling for the Munich Stammstrecke Network

About the Project

This repository contains the resources, datasets, and scripts used in the thesis titled "User Feedback and Station Management with AI: Sentiment Analysis and Topic Modeling for the Munich Stammstrecke Network". The study explores the integration of AI tools into public transportation infrastructure management, focusing on Munich's Stammstrecke train stations.

Table of Contents

• Overview
• Datasets
• Methodology
• Tools and Techniques
• Results
• How to Use
• License

Overview

Efficient station management is critical for improving user satisfaction and meeting urban mobility goals. This project applies sentiment analysis and topic modeling on user-generated feedback collected from Google Maps reviews to:

• Quantify user sentiment
• Identify key themes
• Propose actionable recommendations

Datasets

Data Sources

• Primary Source: User reviews from Google Maps for 10 Munich Stammstrecke stations (2019–2024).
  • Review dataset
  • Sentiment dataset
  • Topic modelling dataset
  • Interview questionare
• Ethics: Data is publicly available; no personally identifiable information was collected.

Data Details

• Total Reviews: 7,043 collected, 3,592 validated.
• Features: Review text, star rating, date, station name, etc.

Methodology

1. Data Preprocessing: Cleaning, tokenization, stopword removal, stemming, and lemmatization.
2. Sentiment Analysis: Using RoBERTa and IBM Watson Natural Language Understanding to classify sentiment and detect emotions.
3. Topic Modeling: Using BERTopic and Llama 2 to identify themes in feedback.
4. Urgency Classification: Categorizing feedback into immediate, short-term, and long-term priorities.

Tools and Techniques

• Sentiment Analysis: RoBERTa (Hugging Face) and IBM Watson NLU.
• Topic Modeling: BERTopic and Llama 2.
• Data Preprocessing: Python libraries (NLTK, SpaCy, Pandas).
• Visualization: Tableau

Results

• Sentiment Distribution:
  • Positive: 25.8%
  • Neutral: 51.9%
  • Negative: 22.3%
• Key Topics: Cleanliness, infrastructure, crowding, customer service, accessibility.
• Conclussions and Recommendations:
  • User-generated feedback, when systematically analyzed, offers a powerful resource for identifying recurring issues, prioritizing actions, and tailoring interventions to real-world needs.
  • The integration of AI tools like sentiment analysis and topic modeling demonstrates significant potential for transforming public transportation management, providing actionable insights into user satisfaction and system inefficiencies.
  • Station-specific analysis reveals critical differences in user experiences, emphasizing the need for localized solutions rather than one-size-fits-all strategies.
  • This research underlines the role of AI-driven tools in advancing smart city initiatives, fostering user-centered urban mobility systems that align with sustainability and inclusivity goals.

How to Use

Installation

1. Clone the repository:

   git clone https://github.com/your-username/station-feedback-ai.git
   cd station-feedback-ai
   

Running the Analysis

1. Preprocess Data

Clean and prepare the raw dataset for analysis:

python src/preprocessing/clean_data.py

2. Run Sentiment Analysis

Perform sentiment analysis on user reviews:

python src/sentiment_analysis/analyze_sentiment.py

3. Generate Topics

Use topic modeling to identify themes in user feedback:

python src/topic_modeling/model_topics.py

4. Classify Urgency

Categorize feedback into urgency levels (immediate, short-term, long-term):

python src/classification/urgency_classification.py

License

This project is licensed under the MIT License.