System and method for providing large language model for sanctions artificial intelligence assisted automation

A system for reducing false-positives using neural network, the system comprising:

• a memory;
• and
• a processor,
• wherein the system is configured to perform:
• receiving a data envelope in a coding format;
• scanning the data envelope and detecting a presence of data corresponding to a value in a list of sanctioned entities, wherein the data corresponding to the value in the list of sanctioned entities is determined to be detected when a match between a text included in the data envelope and the value in the list of sanctioned entities is above a reference threshold, and wherein the value in the list of sanctioned entities includes a logo;
• when there is an exact match between the text included in the data envelope and the value in the list of sanctioned entities, flagging the data envelope and reporting the data envelope to a government entity;
• when there is no exact match between the text included in the data envelope and the value in the list of sanctioned entities, but a level of match between the text included in the data envelope and the value in the list of sanctioned entities is above the reference threshold for a potential match:
• transforming the coding format of the data envelope into a text format;
• identifying and extracting target information including the data corresponding to the value in the list of sanctioned entities and surrounding information, wherein the target information includes text information corresponding to a target key field;
• reformatting the extracted target information and surrounding information into a natural language question;
• processing, by a neural network model executed by the processor, the natural language question;
• determining, by the neural network model executed by the processor, whether a sanctioned entity is included or not based on context of the natural language question, wherein the context of the natural language question is provided based on a data field value for a corresponding text value and adjacent text values, and wherein the adjacent text values include a street name;
• and
• when the sanctioned entity is determined to be included in the natural language question, reporting the data envelope with a corresponding confidence level to the government entity,
• wherein the context of the natural language question is further determined by:
• tokenizing text of the natural language question to generate a plurality of tokens, and tracking a sequence of the plurality of tokens;
• processing the plurality of tokens through a plurality of transformer layers;
• removing a portion of the plurality of tokens from further processing while retaining a portion of the plurality of tokens;
• and
• performing a classification of the target information included in the natural language question based on the retained portion of the plurality of tokens.

A non-transitory computer readable storage medium that stores a computer program for reducing false-positives using neural network, the computer program, when executed by a processor, causing a system to perform a plurality of processes comprising:

• receiving a data envelope in a coding format;
• scanning the data envelope and detecting a presence of data corresponding to a value in a list of sanctioned entities, wherein the data corresponding to the value in the list of sanctioned entities is determined to be detected when a match between a text included in the data envelope and the value in the list of sanctioned entities is above a reference threshold, and wherein the value in the list of sanctioned entities includes a logo;
• when there is an exact match between the text included in the data envelope and the value in the list of sanctioned entities, flagging the data envelope and reporting the data envelope to a government entity;
• when there is no exact match between the text included in the data envelope and the value in the list of sanctioned entities, but a level of match between the text included in the data envelope and the value in the list of sanctioned entities is above the reference threshold for a potential match:
• transforming the coding format of the data envelope into a text format;
• identifying and extracting target information including the data corresponding to the value in the list of sanctioned entities and surrounding information, wherein the target information includes text information corresponding to a target key field;
• reformatting the extracted target information and surrounding information into a natural language question;
• processing, by a neural network model executed by the processor, the natural language question;
• determining, by the neural network model executed by the processor, whether a sanctioned entity is included or not based on context of the natural language question, wherein the context of the natural language question is provided based on a data field value for a corresponding text value and adjacent text values, and wherein the adjacent text values include a street name;
• and
• when the sanctioned entity is determined to be included in the natural language question, reporting the data envelope with a corresponding confidence level to the government entity,
• wherein the context of the natural language question is further determined by:
• tokenizing text of the natural language question to generate a plurality of tokens, and tracking a sequence of the plurality of tokens;
• processing the plurality of tokens through a plurality of transformer layers;
• removing a portion of the plurality of tokens from further processing while retaining a portion of the plurality of tokens;
• and
• performing a classification of the target information included in the natural language question based on the retained portion of the plurality of tokens.