An Approach For Scalable First-Order Rule Learning On Twitter Data
No Thumbnail Available
Authors
Meeting name
Sponsors
Date
Journal Title
Format
Subject
Abstract
Scalable Rule Learning (SRLearn) is a scalable divide-and-conquer approach with graph-based modeling of social media data, to scale up first-order rule learning through Markov Logic Networks on a commodity cluster on large scale Twitter data. SRLearn takes advantage of distributed systems to partition large-scale data into smaller but meaningful partitions based on user interaction and incorporates a gradient boosting approach with a tool called BoostSRL for first-order rule mining. We show how this scalable solution on first order predicates is more accurate and efficient than existing systems, such as ProbKB (a scalable system to construct probabilistic knowledge base) and XGBoost (extreme gradient boosting) on relational data.
Table of Contents
Introduction -- Background -- Related work -- SRLearn -- Experimentation and evaluation -- Conclusion and future work