Computer-implemented contract risk assessment platform leveraging transformers
Methods, systems, and computer-readable storage media for receiving a document provided as a computer-readable file, receiving a set of questions, for each question in the set of questions, generating an inference input including a question, at least a portion of text of the document, and multiple tokens, processing, by a PLM, the inference input to generate a set of text embeddings, processing, by a neural network, the set of text embeddings to provide sets of tokens, each set of tokens being specific to a segment of the document and including a start token and an end token respectively identifying a start position and an end position of the segment, determining, from the sets of tokens, a segment for display, and displaying at least a portion of the document in a UI and an annotation indicating the segment within the at least a portion of the document.
1 . A computer-implemented method for automatic identification and display of segments of text in computer-readable documents, the method being executed by one or more processors and comprising:
receiving a document provided as a computer-readable file;
receiving a set of questions;
for each question in the set of questions, generating an inference input comprising a question, at least a portion of text of the document, and multiple tokens;
processing, by a pre-trained language model (PLM), the inference input to generate a set of text embeddings;
processing, by a neural network, the set of text embeddings to provide sets of tokens, each set of tokens being specific to a segment of the document and comprising a start token and an end token respectively identifying a start position and an end position of the segment;
determining, from the sets of tokens, a segment for display by determining a score difference based on a null score and a non-null score and selecting the segment; and
displaying at least a portion of the document in a user interface (UI) and an annotation indicating the segment within the at least a portion of the document.
2 . The method of claim 1 , wherein tokens in each set of tokens is associated with a logit comprising a non-normalized value that is normalized by a softmax function.
3 . The method of claim 1 , wherein the segment is selected based on a null prediction represented by the null score in response to the score difference exceeding a threshold.
4 . The method of claim 1 , wherein the document comprises a contract and the segment comprises a portion of a clause of the contract.
5 . The method of claim 1 , wherein the annotation comprises highlighting of the segment in the UI, the highlighting extending from the start token and the end token determined for the segment.
6 . The method of claim 1 , wherein the PLM comprises a Bidirectional Encoder Representations from Transformers (BERT) model.
7 . 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 automatic identification and display of segments of text in computer-readable documents, the operations comprising:
receiving a document provided as a computer-readable file;
receiving a set of questions;
for each question in the set of questions, generating an inference input comprising a question, at least a portion of text of the document, and multiple tokens;
processing, by a pre-trained language model (PLM), the inference input to generate a set of text embeddings;
processing, by a neural network, the set of text embeddings to provide sets of tokens, each set of tokens being specific to a segment of the document and comprising a start token and an end token respectively identifying a start position and an end position of the segment;
determining, from the sets of tokens, a segment for display by determining a score difference based on a null score and a non-null score and selecting the segment; and
displaying at least a portion of the document in a user interface (UI) and an annotation indicating the segment within the at least a portion of the document.
8 . The non-transitory computer-readable storage medium of claim 7 , wherein tokens in each set of tokens is associated with a logit comprising a non-normalized value that is normalized by a softmax function.
9 . The non-transitory computer-readable storage medium of claim 7 , wherein the segment is selected based on a null prediction represented by the null score in response to the score difference exceeding a threshold.
10 . The non-transitory computer-readable storage medium of claim 7 , wherein the document comprises a contract and the segment comprises a portion of a clause of the contract.
11 . The non-transitory computer-readable storage medium of claim 7 , wherein the annotation comprises highlighting of the segment in the UI, the highlighting extending from the start token and the end token determined for the segment.
12 . The non-transitory computer-readable storage medium of claim 7 , wherein the PLM comprises a Bidirectional Encoder Representations from Transformers (BERT) model.
13 . 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 automatic identification and display of segments of text in computer-readable documents, the operations comprising:
receiving a document provided as a computer-readable file;
receiving a set of questions;
for each question in the set of questions, generating an inference input comprising a question, at least a portion of text of the document, and multiple tokens;
processing, by a pre-trained language model (PLM), the inference input to generate a set of text embeddings;
processing, by a neural network, the set of text embeddings to provide sets of tokens, each set of tokens being specific to a segment of the document and comprising a start token and an end token respectively identifying a start position and an end position of the segment;
determining, from the sets of tokens, a segment for display by determining a score difference based on a null score and a non-null score and selecting the segment; and
displaying at least a portion of the document in a user interface (UI) and an annotation indicating the segment within the at least a portion of the document.
14 . The system of claim 13 , wherein tokens in each set of tokens is associated with a logit comprising a non-normalized value that is normalized by a softmax function.
15 . The system of claim 13 , wherein the segment is selected based on a null prediction represented by the null score in response to the score difference exceeding a threshold.
16 . The system of claim 13 , wherein the document comprises a contract and the segment comprises a portion of a clause of the contract.
17 . The system of claim 13 , wherein the annotation comprises highlighting of the segment in the UI, the highlighting extending from the start token and the end token determined for the segment.