A boutique investment bank automates its Entity & Product mapping for Taxonomy using Neural network.

Hierarical Neural network helped to classifying entities & product according to a hierarchy and helped to build automated taxonomy mapping.

Summary

A boutique investment financial firm aimed to optimize its data operation by automating taxonomy mapping process which was manual and time consuming. The current process was rule based and prone to error as every new file contains changes in hierarchy and names of both entities and its underlying structure. Identifying these changes and mapping against to its won structured Taxonomy was tedious & time consuming task.

Business Challenges 

• Standardized list and mapped entities were not uniform as they had additional words and casing issues.

• Many entity mappings consisted of variations greater than 2. Thus, causing ambiguity.

• There were many overlapping entities, abbreviations, spelling mistakes.

• Large amount of data to be processed including those entries as well which were not in scope, causing additional noises.

• Current Rule based and similarity models failed to provide more than 20% of matching for every fresh run.

CRISIL Approach

CRISIL proposed a comprehensive framework & workflow design which includes two approaches that helped incorporate business rules through data & model: 

• Deep Learning approach along with sufficient historical data to capture entity name change, map with high accuracy for overlapping classes.

• Hierarchical Neural Networks (HNN) helped to establish hierarchical relationships (parent-subsidiaries).

An approach that helped to incorporate both business rules and hierarchical relationships to arrive accurate mapping for each run.

Value Delivered

Integrating with NLP, data processing, rules & HNN data operation team could able to achieve higher accuracy, higher productivity and reduction in run time, resulting in speed of delivery and better client satisfaction.