Remediation site portfolio risk scoring

A system comprising:

• one or more processors;
• and
• one or more tangible, non-transitory memories configured to communicate with the one or more processors,
• the one or more tangible, non-transitory memories having instructions stored thereon that, in response to execution by the one or more processors, cause the one or more processors to perform operations comprising:
• developing, by the one or more processors, an extreme event risk score based on operational environment data, legislative environment data and a data environment;
• determining, by the one or more processors, a remediation score based on a plurality of remediation terms including harm reduction data, cost data, sustainability data, resiliency data, the extreme event risk score and the data environment;
• determining, by the one or more processors, how each remediation term of the plurality of remediation terms influences the remediation score by combining each of the plurality of remediation terms via linear combination with a weighting term,
• wherein the weighting term is determined by statistical analysis of at least one of previous sites, portfolio priorities or an emphasis of a regulatory environment on each of the plurality of remediation terms;
• determining, by the one or more processors, a remediated risk score for the site based on an initial risk score and the remediation score;
• determining, by the one or more processors, a risk rating for the site pre-remediation based on the initial risk score and an information depth score;
• determining, by the one or more processors, the risk rating for the site post-remediation based on the remediated riskscore and the information depth score;
• ingesting, by the one or more processors, legislative documents from the legislative environment and into the data environment;
• saving, by the one or more processors, the legislative documents as text documents in a data environment;
• parsing and converting, by the one or more processors, the text documents to at least one of a directed graph or state-specific decision trees using artificial intelligence (AI) models that include deep learning models and natural language processing (NLP) routines,
• wherein the directed graph is an object that describes decision nodes and edges,
• wherein the edges are a directional relationship between the decision nodes, and
• wherein the directed graph enforces directionality constraints on the nodes to encode at least one of causality or sequence;
• detecting, by the one or more processors, knowledge to obtain detected knowledge in the form of attributes comprising key terms, rules, topic summaries, relationships between various legal terms, semantically similar terminologies, deontic expressions and cross-referenced legal facts and rules;
• tagging, by the one or more processors, the detected knowledge from the text documents as a decision node or a outcome node;
• building, by the one or more processors, a knowledge graph using deep learning technologies, word embeddings, text summarization, embedded provenance and topic modeling;
• extracting, by the one or more processors, cross-referenced rules from the knowledge graph;
• identifying and classifying, by the one or more processors, the cross-referenced rules into the deontic expressions;
• re-training, by the one or more processors, a predictive model with definitions of categories of each section of the legislative documents;
• implementing, by the one or more processors, content segmentation of the legislative documents to categorize each section of the legislative documents into the categories by type;
• re-training, by the one or more processors, the predictive model using the risk rating and the legislative documents;
• adjusting, by the one or more processors, probabilities of the extreme event risk score based on a severity of an outcome to obtain adjusted probabilities, wherein the adjusted probabilities are visually indicated by a weight of the edges in the decision tree;
• adjusting the weight of the edges in the decision tree by using the machine learning models;
• computing, by the one or more processors, a training objective by re-calculating probabilities underlying risk calculations of the initial risk score, by using the adjusted probabilities of the extreme event risk score and by defining portfolio priorities with the operational environment data and the legislative environment data;
• updating, by the one or more processors, the predictive model to create an updated predictive model, based on the training objective;
• and
• improving, by the one or more processors using the updated predictive model, an accuracy of the initial risk score and the risk rating to improve prediction accuracies for the other sites.