Systems and Methods for Expert Driven Document Identification

A method for identifying data strings in electronic documents using pattern recognition, the method comprising:

• receiving, by a server computing device, a first data string corresponding to a first sentence of a first plurality of sentences of an electronic reference document from a first database;
• receiving, by the server computing device, a second data string corresponding to a second sentence of a second plurality of sentences of an electronic legal document from a second database;
• processing, by the server computing device, the first data string corresponding to the first sentence into a first processed data string, wherein processing the first data string comprises at least one of removing stop words, removing punctuation, removing digits, converting all characters to lower-case, or lemmatization;
• processing, by the server computing device, the second data string corresponding to the second sentence into a second processed data string, wherein processing the second data string comprises at least one of removing stop words, removing punctuation, removing digits, converting all characters to lower-case, or lemmatization;
• computing, by the server computing device, a term frequency-inverse document frequency (TF-IDF) matrix based upon the first processed data string and the second processed data string;
• generating, by the server computing device, a real-valued vector for each word in the first processed data string and for each word in the second processed data string;
• calculating, by the server computing device, a centroid-based cosine similarity score using the TF-IDF matrix and the real-valued vectors for the words in each of the first processed data string and the second processed data string;
• receiving, by the server computing device, a feedback score from a user, wherein the feedback score corresponds to an accuracy of the centroid-based cosine similarity score between the first processed data string and the second processed data string;
• and
• calculating, by the server computing device, an adjusted centroid-based cosine similarity score between the first processed data string and the second processed data string based on the centroid-based cosine similarity score and the feedback score.