Is it feasible to detect FLOSS version release events from textual messages? A case study on Stack Overflow.

Area of Science:

• Text Mining
• Natural Language Processing
• Machine Learning

Background:

• Topic Detection and Tracking (TDT) typically focuses on single-message events.
• Micro-events, undetectable from single messages, require new detection methods.
• Existing TDT methods are insufficient for nuanced, multi-message event detection.

Purpose of the Study:

• Investigate the feasibility of detecting micro-events in textual data.
• Evaluate different feature spaces and detection pipelines for micro-event identification.
• Determine the detectability threshold for micro-event classifiers.

Main Methods:

• Utilized Stack Overflow and Libraries.io datasets for micro-event detection.
• Developed detection pipelines with three estimators, optimized via grid search.
• Employed LDA and hSBM topic modeling with sentiment analysis, optimized with RFECV.
• Performed detailed statistical analysis and generated synthetic datasets to test limits.

Main Results:

• Evaluated the capacity of different pipeline variants to detect micro-events.
• Investigated characteristic changes in topic distribution and sentiment features around micro-events.
• Established micro-event detectability thresholds for evaluated classifiers.

Conclusions:

• Demonstrated the feasibility of micro-event detection using topic and sentiment analysis.
• Highlighted the importance of feature space selection and pipeline optimization.
• Provided a framework for understanding the limits of micro-event detection in textual data.