Exploring Time-Sensitive Variational Bayesian Inference LDA for Social Media Data
- Submitting institution
-
University of Glasgow
- Unit of assessment
- 11 - Computer Science and Informatics
- Output identifier
- 11-05327
- Type
- E - Conference contribution
- DOI
-
10.1007/978-3-319-56608-5_20
- Title of conference / published proceedings
- 39th European Conference in Information Retrieval
- First page
- 252
- Volume
- -
- Issue
- -
- ISSN
- -
- Open access status
- -
- Month of publication
- April
- Year of publication
- 2017
- URL
-
http://eprints.gla.ac.uk/135028/
- Supplementary information
-
-
- Request cross-referral to
- -
- Output has been delayed by COVID-19
- No
- COVID-19 affected output statement
- -
- Forensic science
- No
- Criminology
- No
- Interdisciplinary
- No
- Number of additional authors
-
4
- Research group(s)
-
-
- Citation count
- 3
- Proposed double-weighted
- No
- Reserve for an output with double weighting
- No
- Additional information
- ORIGINALITY: First to integrate time in Variational Bayesian (VB) LDA topic modelling and to show this generates more coherent topics. RIGOUR: Mathematically shows how time-sensitive VB topic modelling can be soundly implemented using an EM algorithm. Evaluation is conducted quantitatively using 8 different events from Twitter in comparison to 4 strong baselines and through a user study. SIGNIFICANCE: Published at a top IR conference; Best Student-Paper Award (2/248. Allows to generate significantly more coherent topics (confirmed both quantitatively and through a user study). These topics are more easily interpreted e.g. when analysing discussions on social media.
- Author contribution statement
- -
- Non-English
- No
- English abstract
- -