Privacy-in-dialogue-systems-analysis

An analysis of dialogue system logs with regard to private/sensitive information

Contents

This analysis focuses on the detection of private user data in commercial dialogue systems' stored user inputs.

Data

The files data_analysis Google.ipynb and data_analysis Amazon.ipynb include all the analyses performed on the logs of commercial dialogue systems. The datasets themselves are not made publicly available since they contain private user information. However, bits of the data can be found in the Jupyter Notebook for better understanding.

Analysis

The analysis contains....

• ... sentence embeddings for all user inputs
• ... dimension reduction via Principal Component Analysis for visual representation of user input
• ... k-means clustering of inputs to detect patterns

Python packages needed for analysis

• sentence_transformers
• pandas
• sklearn
• matplotlib