Hierarchical and multi-label transaction cleansing and categorization

A non-transitory computer-readable storage medium comprising executable instructions that when executed by at least one processor perform operations, comprising:

• receiving a transaction string associated with payment processing for a transaction by a financial institution (FI);
• cleansing and normalizing the transaction string as normalized data;
• labeling entities identified in the normalized data;
• wherein labeling further includes:
• providing the normalized data as input to a first machine learning model trained on named entity recognition, the first machine learning model processing a bidirectional encoder representations from transformers (BERT) algorithm that converts words in the transaction string into a numerical format while maintaining context of the words within the normalized data;
• adjusting embeddings of each word during training to learn context of where merchants appear in transaction strings;
• and
• receiving an entity-labeled normalized string as output from the first machine learning model;
• and
• assigning categories to each entity based on a unique hierarchy associated with each entity using the normalized data to produce a multi-entity labeled and multi-categorized string for the transaction;
• wherein assigning further includes:
• providing the entity-labeled normalized string as input to a second machine learning model comprising a hierarchical convolutional neural network (HCNN) trained on hierarchies associated with the entities, wherein the second machine learning model remaps merchant category codes present in the entity-labeled normalized string to predefined categories and cascades each labeled entity and the entity-labeled normalized string through a top of a corresponding hierarchy associated with a corresponding entity as a cascading HCNN that works on each subspace of each corresponding hierarchy until there are no subspaces through which to cascade;
• receiving the multi-entity labeled and multi-categorized string as output from the second machine learning model, wherein probabilities assigned to categories are independent such that sum of category probabilities do not equal 1, enabling multiple categories to be simultaneously assigned to a single transaction;
• and
• providing the multi-entity labeled and multi-categorized string to a system associated with the FI for subsequent analytics.

A system, comprising:

• at least one processor comprising a non-transitory computer-readable storage medium with instructions;
• and
• the instructions when executed by the at least one processor cause the at least one processor to perform operations, comprising:
• receiving a transaction string associated with payment processing for a transaction by a financial institution (FI);
• cleansing and normalizing the transaction string as normalized data;
• labeling entities identified in the normalized data;
• wherein labeling further includes:
• providing the normalized data as input to a first machine learning model trained on named entity recognition, the first machine learning model processing a bidirectional encoder representations from transformers (BERT) algorithm that converts words in the transaction string into a numerical format while maintaining context of the words within the normalized data;
• adjusting embeddings of each word during training to learn context of where merchants appear in transaction strings;
• and
• receiving an entity-labeled normalized string as output from the first machine learning model;
• and
• assigning categories to each entity based on a unique hierarchy associated with each entity using the normalized data to produce a multi-entity labeled and multi-categorized string for the transaction;
• wherein assigning further includes:
• providing the entity-labeled normalized string as input to a second machine learning model comprising a hierarchical convolutional neural network (HCNN) trained on hierarchies associated with the entities, wherein the second machine learning model remaps merchant category codes present in the entity-labeled normalized string to predefined categories and cascades each labeled entity and the entity-labeled normalized string through a top of a corresponding hierarchy associated with a corresponding entity as a cascading HCNN that works on each subspace of each corresponding hierarchy until there are no subspaces through which to cascade;
• receiving the multi-entity labeled and multi-categorized string as output from the second machine learning model, wherein probabilities assigned to categories are independent such that sum of category probabilities do not equal 1, enabling multiple categories to be simultaneously assigned to a single transaction;
• and
• providing the multi-entity labeled and multi-categorized string to a system associated with the FI for subsequent analytics.