Artificial intelligence engine for entity resolution and standardization

Abstract

Techniques for training an entity resolution model are presented. The techniques include inputting raw training data into the entity resolution model. The training data includes historical transaction data including a plurality of transactions. A label dictionary is generated by performing natural language processing (NLP) on the training data. The NLP includes scanning text of each transaction, extracting one or more entities from the text, and storing the label dictionary in a database. The label dictionary includes the extracted entities. Tagged data is generated from the training data using the label dictionary. Vocabulary training is performed on the training data, including tokenizing the text of each transaction and converting the tokenized text into a transformer model specific format. The entity resolution model is then trained using a transformer model, which uses the tokenized text and the tagged data. The trained entity resolution model is then stored in a database.

What is claimed is: 
     
         1 . A computer-implemented method performed by a server for training an entity resolution model for entity determination and standardization of financial transactions, the method comprising:
 inputting raw training data into the entity resolution model, the raw training data including historical transaction data comprising a plurality of transactions;   generating a label dictionary, the generating comprising performing natural language processing on the raw training data using an automatic engine, the natural language processing including:
 scanning text of each transaction; 
 extracting one or more entities from the text; and 
 storing the label dictionary in a database, the label dictionary including the one or more entities; 
   generating tagged data from the raw training data using the label dictionary;   performing vocabulary training on the raw training data, including tokenizing the text of each transaction and converting the tokenized text into a transformer model specific format;   training the entity resolution model using a transformer model, the tokenized text in the transformer model specific format, and the tagged data; and   storing the trained entity resolution model in a database.   
     
     
         2 . The computer-implemented method in accordance with  claim 1 , further comprising receiving the raw training data from a data source. 
     
     
         3 . The computer-implemented method in accordance with  claim 2 ,
 receiving the raw training data comprises one or more of the following: retrieving the raw training data from the database and receiving the raw training data from one or more data source computing devices.   
     
     
         4 . The computer-implemented method in accordance with  claim 1 ,
 performing the natural language processing on the raw training data comprises using Python-based Natural Language Toolkit (NLTK).   
     
     
         5 . The computer-implemented method in accordance with  claim 1 , extracting the one or more entities from the text comprises performing the following processes on the text of each transaction: word tokenization, word lemmatization, stop word removal, punctuation removal, and accented character and word conversion. 
     
     
         6 . The computer-implemented method in accordance with  claim 5 ,
 performing the word tokenization process comprises splitting the text into individual words and defining a token for each word.   
     
     
         7 . The computer-implemented method in accordance with  claim 1 ,
 generating tagged data from the raw training data using the label dictionary comprises performing name entity recognition (NER) tagging on each transaction, including labelling each respective transaction with the one or more entities identified in the respective transaction.   
     
     
         8 . The computer-implemented method in accordance with  claim 1 ,
 tokenizing the text of each transaction of the raw training data includes performing subword-based tokenization on the raw training data.   
     
     
         9 . The computer-implemented method in accordance with  claim 8 ,
 performing the subword-based tokenization on the processed training data comprises:
 initializing a word unit inventory with all the characters in the text of each transaction of the processed training data; 
 building a language model on the processed training data using the word unit inventory; and 
 generating a new word unit by combining two units out of the word inventory. 
   
     
     
         10 . The computer-implemented method in accordance with  claim 1 , further comprising determining whether training criteria for the entity resolution model are met, the training criteria comprising one or more of the following: a target loss accuracy, a depth for recall at a specified percentage, and an F1 measure. 
     
     
         11 . A server for training an entity resolution model for entity determination and standardization of financial transactions, the server comprising:
 a database;   one or more processors; and   a memory storing computer-executable instructions, that when executed by the one or more processors, cause the one or more processors to:
 input the raw training data into the entity resolution model, the raw training data including historical transaction data comprising a plurality of transactions; 
 generate a label dictionary, the generating comprising performing natural language processing on the raw training data using an automatic engine, the natural language processing including:
 scanning text of each transaction; 
 extracting one or more entities from the text; and 
 storing the label dictionary in a database, the label dictionary including the one or more entities; 
 
 generate tagged data from the raw training data using the label dictionary; 
 perform vocabulary training on the raw training data, including tokenizing text of each transaction of the processed training data and converting the tokenized text into a transformer model specific format; 
 train the entity resolution model using a transformer model, the tokenized text in the transformer model specific format, and the tagged data; and 
 store the trained entity resolution model in the database. 
   
     
     
         12 . The server in accordance with  claim 11 ,
 said computer-executable instructions causing the one or more processors to receive the raw training data from a data source.   
     
     
         13 . The server in accordance with  claim 12 ,
 receiving the raw training data comprises one or more of the following: the one or more processors retrieving the raw training data from the database and the one or more processors receiving the raw training data from one or more data source computing devices.   
     
     
         14 . The server in accordance with  claim 11 ,
 performing the natural language processing on the raw training data comprises the one or more processors using Python-based Natural Language Toolkit (NLTK).   
     
     
         15 . The server in accordance with  claim 11 ,
 extracting the one or more entities from the text comprises the one or more processors performing the following processes on the text of each transaction: word tokenization, word lemmatization, stop word removal, punctuation removal, and accented character and word conversion.   
     
     
         16 . The server in accordance with  claim 15 ,
 performing the word tokenization process comprises the one or more processors splitting the text into individual words and defining a token for each word.   
     
     
         17 . The server in accordance with  claim 11 ,
 generating tagged data from the raw training data using the label dictionary comprises the one or more processors performing name entity recognition (NER) tagging on each transaction, including labelling each respective transaction with the one or more entities identified in the respective transaction.   
     
     
         18 . The server in accordance with  claim 11 ,
 tokenizing the text of each transaction of the processed training data includes the one or more processors performing subword-based tokenization on the processed training data.   
     
     
         19 . The server in accordance with  claim 18 ,
 performing the subword-based tokenization on the processed training data comprises the one or more processors:
 initializing a word unit inventory with all the characters in the text of each transaction of the processed training data; 
 building a language model on the processed training data using the word unit inventory; and 
 generating a new word unit by combining two units out of the word inventory. 
   
     
     
         20 . The server in accordance with  claim 11 ,
 said computer-executable instructions causing the one or more processors to determine whether training criteria for the entity resolution model are met, the training criteria comprising one or more of the following: a target loss accuracy, a depth for recall at a specified percentage, and an F1 measure.