Automated decision modelling from text
The present inventive concept provides for a method for automated decision modelling from text including obtaining a text corpus including a policy. Terms and syntax are identified within the text corpus related to the policy. Sentence similarities and co-references based on the terms and syntax are identified. Discourse and sentence level semantic parsing is performed based on the terms and the sentence similarities and the co-references using machine learning. A decision model template is generated based on the discourse and semantic parsing, and the decision model template is transformed into an automated decision model.
1 . A computer-implemented method for automated decision modelling from text, the method comprising:
obtaining, via a network, a text corpus including a policy;
applying natural language processing (NLP) to the text corpus to extract terms and syntax within the text corpus related to the policy;
determining sentence similarities and co-references, within the text corpus, based on the terms and the syntax,
wherein determining the sentence similarities and co-references comprises:
generating, using a first transformer model, text embedding vectors for sentence fragments of the sentence similarities and co-references, and
applying a second transformer model to the text embedding vectors and the sentence fragments to determine the sentence similarities and co-references within the text corpus,
wherein the second transformer model is applied to the text embedding vectors and the sentence fragments to determine redundant sentences of a same rule for a same decision within the text corpus, and
wherein similarity between a pair of the sentence fragments is calculated, using the second transformer model, as an aggregated similarity of the text embedding vectors and a bag of noun-phrases and verb-phrases extracted from the sentence fragments;
performing discourse and sentence level semantic parsing based on the terms and the sentence similarities and the co-references using machine learning;
generating a decision model template based on the discourse and sentence level semantic parsing; and
transforming the decision model template into an automated decision model.
2 . The computer-implemented method of claim 1 , wherein the text corpus includes a single document, and wherein the terms include terms that are related to a decision and criteria therefor.
3 . The computer-implemented method of claim 1 , wherein the NLP includes named-entity recognition (NER) and a parsing tree,
wherein applying the NLP to the text corpus to extract the terms and syntax comprises:
applying the NLP to the text corpus to extract named-entities,
wherein the same decision includes an entitlement, a user agreement, a benefit, a penalty, or a compliance certification.
4 . The computer-implemented method of claim 3 , wherein identifying the terms includes using a knowledge graph, and
wherein the knowledge graph represents a network of terms and illustrates relationships between the terms.
5 . The computer-implemented method of claim 1 , wherein the text corpus includes a decision, and wherein the similar sentences are rules in the decision.
6 . The computer-implemented method of claim 5 , wherein determining the sentence similarities and co-references comprises determining the sentence similarities and co-references based on predetermined thresholds of matching terms, synonyms, semantics, syntax, and ontology.
7 . The computer-implemented method of claim 6 , wherein anaphora resolution (AR) is used to identify synonyms to identified terms and pronouns, wherein the sentence fragments of the sentence similarities and co-references are given text embedding vectors using the first transformer model, wherein the bag of noun-phrases and verb-phrases is extracted from the sentence fragments using an abstract meaning representation (AMR) parser, and wherein similarity between a pair of sentence fragments is calculated, using the second transformer model, as the aggregated similarity of the text embedding vectors and the bags of noun-phrases and verb phrases using similarity between sentence embeddings.
8 . The computer-implemented method of claim 1 , further comprising:
identifying categories for text spans of the policy related to a decision in the text corpus; and
generating a decision model template outlining decision logic and identified categories from the text corpus.
9 . The computer-implemented method of claim 8 , further comprising:
parsing text spans based upon rhetoric structure theory (RST); and
using a cluster algorithm to group text spans based on embeddings into the identified categories.
10 . The computer-implemented method of claim 9 , wherein the decision model includes a decision requirement diagram (DRD) depicting the identified categories and decision logic in the text corpus.
11 . The computer-implemented method of claim 10 , wherein the DRD includes a hierarchy of decision boxes based upon the categories and decision logic, wherein each decision box includes a decision table with rules expressed in Boolean, wherein the Boolean is extracted from text spans using curated labelled data and AMR parsers with provided syntactical dependency rules for extracting logical expressions, and wherein the rules expressed inside the decision tables are in a standardized standard friendly enough expression language (S-FEEL) language.
12 . A computer program product for automated decision modelling from text, the computer program comprising:
one or more computer-readable storage media and program instructions stored on the one or more computer-readable storage media, the program instructions comprising:
program instructions to obtain a text corpus including a policy;
program instructions to apply natural language processing (NLP) to the text corpus to extract terms and syntax within the text corpus related to the policy;
program instructions to determine, using transformer models, sentence similarities and co-references based on the terms and the syntax,
wherein the program instructions to determine the sentence similarities and co-references include:
program instructions to generate, using a first transformer model, text embedding vectors for sentence fragments of the sentence similarities and co-references, and
program instructions to apply a second transformer model to the text embedding vectors and the sentence fragments to determine the sentence similarities and co-references within the text corpus,
wherein similarity between a pair of the sentence fragments is calculated, using the second transformer model, as an aggregated similarity of the text embedding vectors and a bag of noun-phrases and verb-phrases extracted from the sentence fragments;
program instructions to perform discourse and sentence level semantic parsing based on the terms and the sentence similarities and the co-references using machine learning;
program instructions to generate a decision model template based on the discourse and sentence level semantic parsing; and
program instructions to transform the decision model template into an automated decision model.
13 . The computer program product of claim 12 , wherein the terms include terms that are related to a decision and criteria therefor.
14 . The computer program product of claim 13 , wherein the NLP includes named-entity recognition (NER) and a parsing tree.
15 . The computer program product of claim 14 , wherein the program instructions to identify the terms comprise program instructions to use a knowledge graph.
16 . The computer program product of claim 12 , wherein the text corpus includes a decision, and wherein the similar sentences are rules in the decision.
17 . The computer program product of claim 16 , wherein the program instructions to identify the sentence similarities and co-references include program instructions to identify the sentence similarities and co-references based on predetermined thresholds of matching terms, synonyms, semantics, syntax, and ontology.
18 . The computer program product of claim 17 , wherein anaphora resolution (AR) is used to identify synonyms to identified terms and pronouns, wherein the sentence fragments of the sentence similarities and co-references are given text embedding vectors using pretrained language models, wherein the pretrained language models include bidirectional encoder representations from transformers (BERT) model, wherein the bag of noun-phrases and verb-phrases is extracted from the sentence fragments using an abstract meaning representation (AMR) parser, and wherein similarity between a pair of sentence fragments is calculated as an aggregated similarity of the text embedding vectors and the bags of noun-phrases and verb-phrases using similarity between sentence embeddings.
19 . The computer program product of claim 12 , the program instructions further comprising:
program instructions to identify categories for text spans of the policy related to a decision in the text corpus; and
program instructions to generate a decision model template outlining decision logic and identified categories from the text corpus.
20 . A computer system for automated decision modelling from text, the computer system comprising:
one or more computer processors, one or more computer-readable storage media, and program instructions stored on the one or more of the computer-readable storage media for execution by at least one of the one or more processors, the program instructions comprising:
program instructions to obtain a text corpus including a policy;
program instructions to apply natural language processing (NLP) to the text corpus to extract terms and syntax within the text corpus related to the policy;
program instructions to determine, using transformer models, sentence similarities and co-references based on the terms and the syntax,
wherein the program instructions to determine the sentence similarities and co-references include:
program instructions to generate, using a first transformer model, text embedding vectors for sentence fragments of the sentence similarities and co-references, and
program instructions to apply a second transformer model to the text embedding vectors and the sentence fragments to determine the sentence similarities and co-references within the text corpus,
wherein similarity between a pair of the sentence fragments is calculated, using the second transformer model, as an aggregated similarity of the text embedding vectors and a bag of noun-phrases of verb-phrases extracted from the sentence fragments;
program instructions to perform discourse and sentence level semantic parsing based on the terms and the sentence similarities and the co-references using machine learning;
program instructions to generate a decision model template based on the discourse and sentence level semantic parsing; and
program instructions to transform the decision model template into an automated decision model.
21 . The computer system of claim 20 , wherein the terms include terms that are related to a decision and criteria therefor.
22 . The computer system of claim 20 , wherein the NLP includes named-entity recognition (NER) and a parsing tree.
23 . The computer system of claim 20 , wherein the program instructions to apply NLP to the text corpus to extract the terms include program instructions to use a knowledge graph to identify the terms.
24 . The computer system of claim 20 , wherein the text corpus includes a decision, wherein the decision includes an entitlement, a user agreement, a benefit, a penalty, or a compliance certification, and wherein the similar sentences are rules in the decision.
25 . The computer system of claim 24 , wherein the sentence similarities and co-references are determined based on predetermined thresholds of matching terms, synonyms, semantics, syntax, and ontology.