Smart contract generation system and methods
A system for generating smart contracts may include a first subsystem to receive a written or verbal contract, and a second subsystem to identify terms of the contract using natural language processing (NLP). The system may additionally include a third subsystem to correlate processed NLP terms of the contract with chaincode in a library, and a fourth subsystem to combine correlated NLP terms to generate a smart contract. Methods of generating a smart contract may include inputting natural language contract terms into a smart contract generation system and identifying the natural language contract terms with a natural language processing system. The method may further include correlating at least some of the contract terms to chaincodes stored in a library, generating chaincodes for any contract terms that do not correlate to any chaincodes stored in the library, and assembling the chaincodes into a smart contract.
1 . A system, comprising:
a first subsystem to receive a written or verbal contract;
a second subsystem to form a set of processed NLP terms by:
identifying terms of the contract using a machine-learning based natural language processing (NLP) model, and
abstracting compound NLP contract terms into a plurality of component contract terms to form a set of processed NLP;
a third subsystem to:
correlate processed NLP terms of the contract with chaincode in a library;
generate chaincode for the terms of the contract for which there is no correlating chaincode stored in the library;
receive per-term user feedback for reinforced learning to generate:
(i) a subset of reinforced new chaincodes approved by the user feedback, and
(ii) a subset of corrected new chaincodes that are corrected based on the user feedback, wherein the per-term user feedback either approves a generated chaincode as correctly correlating to an identified natural language contract term or supplies a corrected chaincode that replaces the generated chaincode for that identified term;
fine-tune the machine-learning based NLP model using the subset of reinforced new chaincodes and the subset of corrected new chaincodes as training data associated with their corresponding natural language contract terms, thereby decreasing inaccuracies of future chaincode generation;
store the subset of reinforced new chaincodes and the subset of corrected new chaincodes in the library; and
a fourth subsystem to combine correlated NLP terms to generate a smart contract using:
(i) the existing chaincodes that correlate to the natural language contract terms,
(ii) the reinforced new chaincodes approved by the user feedback, and
(iii) the corrected new chaincodes that are corrected based on the user feedback.
2 . The system of claim 1 , further comprising a computer-readable storage medium that stores at least one of the first subsystem, the second subsystem, the third subsystem, and the fourth subsystem.
3 . The system of claim 2 , wherein the computer-readable storage medium is a non-transitory computer-readable storage medium.
4 . The system of claim 2 , further comprising a bus connecting a processor, a memory, and a network interface to the computer-readable storage medium.
5 . The system of claim 1 further configured to utilize a FILBERT natural language model to generate chaincode.
6 . The system of claim 1 further configured to utilize a GPT based model to generate chaincode.
7 . A method of generating a smart contract, the method comprising:
causing natural language contract terms to be input into a smart contract generation system;
causing the natural language contract terms to be identified with a machine-learning based natural language processing model;
causing compound natural language contract terms to be abstracted into a plurlaity of component contract terms;
causing at least some of the component contract terms to be correlated to chaincodes stored in a library;
causing new chaincodes to be generated that correlate to at least some of the component contract contract terms for which existing chaincodes are not stored in the library;
inputting per-term user feedback on the correlation between the component contract contract terms and the associated new chaincodes for reinforced learning to generate:
(i) a subset of reinforced new chaincodes approved by the user feedback, and
(ii) a subset of corrected new chaincodes that are corrected based on the user feedback,
wherein the per-term user feedback either approves a generated chaincode as correlating to an identified natural-language contract term or supplies a corrected chaincode that replaces the generated chaincode for that identified term;
fine-tuning parameters of the machine-learning based natural language processing model using the reinforced new chaincodes and the corrected new chaincodes as labeled training data associated with their corresponding natural-language contract terms, thereby decreasing inaccuracies and contextual misalignments of future chaincode generation;
assembling a smart contract using:
(i) the existing chaincodes that correlate to the natural language contract terms,
(ii) the reinforced new chaincodes approved by the user feedback, and
(iii) the corrected new chaincodes that are corrected based on the user feedback; and
causing the assembled smart contract to be tested against the natural-language contract terms and:
responsive to successful testing, storing the reinforced new chaincodes and the corrected new chaincodes in the library for future use as existing chaincodes, and
disregarding generated chaincodes that are neither approved nor corrected.