Enhanced model explanations using dynamic tokenization for entity matching models

A non-transitory computer-readable storage medium coupled to one or more processors and having instructions stored thereon which, when executed by the one or more processors, cause the one or more processors to perform operations for providing explanation text for predicted matches of query entities to target entities using machine learning (ML) models, the operations comprising:

• receiving query data and target data, the query data representative of a plurality of query entities of a query entity table and the target data representative of a plurality of target entities of a target entity table, the query entity table comprising rows, each row recording query data of a respective query entity, the target entity table comprising rows, each row recording target data of a respective target entity;
• determining, by an attention machine learning (ML) model, a set of character-level embeddings based on the query data and the target data;
• providing, by a sub-word-level tokenizer, a set of sub-word-level tokens from the query data and the target data, each sub-word-level token comprising a string of multiple characters;
• generating, by the attention ML model, a set of sub-word-level embeddings based on the set of sub-word-level tokens;
• providing, by the attention ML model, at least one attention matrix comprising, for at least a sub-set of sub-word-level tokens, attention scores, each attention score representative of a relative importance of a respective sub-word-level token in a predicted match of a set of predicted matches provided by a matching ML model, each predicted match in the set of predicted matches comprising a match between a query entity of the plurality of query entities and a target entity of the plurality of query entities, the set of predicted matches comprising a first predicted match as a single match that matches a first query entity to a first target entity and a second predicted match as a multi match that matches a second query entity to a second target entity and a third target entity;
• and
• outputting a set of explanations, each explanation corresponding to a query entity and target entity pair and comprising a query text string representative of a portion of a query entity of the query entity and target entity pair and a target text string representative of a portion of a target entity of the query entity and target entity pair that is found in the portion of the query entity, the query text string and the target text string of each explanation in the set of explanations being provided based on the at least one attention matrix.

A system, comprising:

• a computing device;
• and
• a computer-readable storage device coupled to the computing device and having instructions stored thereon which, when executed by the computing device, cause the computing device to perform operations for providing explanation text for predicted matches of query entities to target entities using machine learning (ML) models, the operations comprising:
• receiving query data and target data, the query data representative of a plurality of query entities of a query entity table and the target data representative of a plurality of target entities of a target entity table, the query entity table comprising rows, each row recording query data of a respective query entity, the target entity table comprising rows, each row recording target data of a respective target entity;
• determining, by an attention machine learning (ML) model, a set of character-level embeddings based on the query data and the target data;
• providing, by a sub-word-level tokenizer, a set of sub-word-level tokens from the query data and the target data, each sub-word-level token comprising a string of multiple characters;
• generating, by the attention ML model, a set of sub-word-level embeddings based on the set of sub-word-level tokens;
• providing, by the attention ML model, at least one attention matrix comprising, for at least a sub-set of sub-word-level tokens, attention scores, each attention score representative of a relative importance of a respective sub-word-level token in a predicted match of a set of predicted matches provided by a matching ML model, each predicted match in the set of predicted matches comprising a match between a query entity of the plurality of query entities and a target entity of the plurality of query entities, the set of predicted matches comprising a first predicted match as a single match that matches a first query entity to a first target entity and a second predicted match as a multi match that matches a second query entity to a second target entity and a third target entity;
• and
• outputting a set of explanations, each explanation corresponding to a query entity and target entity pair and comprising a query text string representative of a portion of a query entity of the query entity and target entity pair and a target text string representative of a portion of a target entity of the query entity and target entity pair that is found in the portion of the query entity, the query text string and the target text string of each explanation in the set of explanations being provided based on the at least one attention matrix.