NetEmbs
As a part of the Master's thesis at the University of Amsterdam and ITMO University.
Framework for Representation Learning on Financial Statement Networks.
Keywords – REPRESENTATION LEARNING, FINANCIAL STATEMENTS, NETWORKS, AUDIT, SAMPLING STRATEGY, SKIP-GRAM MODEL, TRANSACTION DATA
The solution relies on both modelling techniques and machine learning. We give a detail definition of sampling strategy, finWalk on a Financial statement network. The novelty of it is to follow directions of relationships on the network rather than directions of edges. As a result, after learning embeddings, one allows merging a large number of business processes into groups as well as revealing an actual meaning of these groups.
In the experiments, we demonstrate the results of applying our coarse-graining procedure to simulated. Moreover, we establish the fact that plausible relationship models considering the predicted labels have the same order of accuracy as the models operating with expert labels. Owing to the framework for data simulation (Simulation folder), we ensure the repeatability of our findings and encourage further investigation and improvements.
Used literature
- Marcel Boersma et al. “Financial statement networks: an application of network theory in the audit”. In: Journal of Network Theory in Finance 4 (2018), pp. 59–85. ISSN: 0033-3174. DOI: 10.21314/JNTF.2018.048
- Bryan Perozzi, Rami Al-Rfou, and Steven Skiena. “DeepWalk: Online Learning of Social Representations”. In: Proceedings of the 20th ACM SIGKDD interna- tional conference on Knowledge discovery and data mining - KDD ’14. New York, New York, USA: ACM Press, Mar. 2014, pp. 701–710. ISBN: 9781450329569. DOI: 10.1145/2623330.2623732
- Yuxiao Dong, Nitesh V. Chawla, and Ananthram Swami. “metapath2vec”. In: Proceedings of the 23rd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining - KDD ’17. New York, New York, USA: ACM Press, 2017, pp. 135–144. ISBN: 9781450348874. DOI: 10.1145/3097983.3098036