Thomson Reuters Enterprise Centre GmbH

A method performed by a machine learning system that involves obtaining a first ontology that includes one or more labels. Each label is associated with a sample that includes text. The ML system is configured to use a particular label to retrieve one or more samples associated with the particular label. The method further involves receiving an identification of a label of a first ontology associated with a first machine learning model to share with a second ontology associated with a second machine learning model and sharing the label and the information with the second ontology. The method further involves training the second machine learning model using the shared information associated with the label.

Aspects of the present disclosure provide systems, methods, apparatus, and computer-readable storage media that support dynamic cross-contextual taxonomy search and interpretation. A multi-layer hierarchical taxonomy may be indicated by metadata that is retrieved based on input search term(s). The multi-layer hierarchical taxonomy may includes a first plurality of nodes that are arranged as parent-child node pairs, and that are grouped into node groups based on common parent nodes. The multi-layer hierarchical taxonomy may be dynamically modified to include a second plurality of nodes that includes duplicates of the parent-child node pairs in which the parent-child relationships are reversed. The nodes may then be regrouped based on common parent nodes, and one or more node groups may be deleted if all the parent-child node pairs are duplicated in another node group. A graphical user interface (GUI) may display a ranked listing of informational elements that correspond to the node groups.

A system and method for using a large language model to automatically identify legal claims or legal causes of action that may apply to a fact pattern provided by a user are described. Automatically determining a legal claim from the fact pattern may include using a structured data collection and a large language model to address a fact pattern and suggest potentially valid causes of action. For each potential legal claim, the systems and methods may, as nonlimiting examples, identify whether the user a) has a potentially valid claim fully supported by the fact pattern, b) might have a potentially valid claim if additional facts were present, or c) does not appear to have a valid claim. The system may then generate a description of the potential claims, framed in terms of the user's facts, and display the description at a display of a graphical user interface.

Computer systems and computer implemented methods for training a machine learning model are provided that includes: selecting seed data from an unlabeled dataset; labeling the seed data and storing the labeled seed data in a data store; training the machine learning model in an initial iteration using the labeled seed data, where the machine learning model is trained to select a next subset of the unlabeled dataset; selecting a next subset of the unlabeled dataset; computing difficulty scores for at least the next subset of the unlabeled dataset; labeling the next subset of the unlabeled data; and training the machine learning model in a second iteration using the labeled next subset of the unlabeled dataset. The machine learning model is generally trained to select the next subset of the unlabeled dataset for a subsequent training iteration by presenting the labeled next subset of the unlabeled dataset in an order sorted based on the difficulty scores.

This disclosure provides systems, methods, and devices for automatic product classification using deep learning and generative artificial intelligence (AI) models for a harmonized system (HS) product classification. A method includes generating embeddings based on an input dataset. The method includes applying a deep learning model to the embeddings to produce a prediction set including classifications corresponding to the embeddings. The method includes converting the classifications to a first set of similarity metrics. The method includes determining a second set of similarity metrics based on the embeddings using a semantic similarity model. The method includes generating a third set of similarity metrics based on an output of the semantic similarity model. The method includes outputting a ranked set of predictions corresponding to the input dataset based on the first set of similarity metrics, the second set of similarity metrics, and the third set of similarity metrics.

The present disclosure is directed towards systems and methods for extracting structure and headers from a body of text. This computational extraction is based on the visual and logical similarities between portions of text. Boilerplate is removed from chunks of text making up potential headers and the cleaned result is compared against other potential headers and the remainder of the body of text.

A system and method of tax regulatory change management is provided, comprising: receiving tax filing instructions and e-file schemas of tax forms; converting the filing instructions into output filing instructions and the e-file schemas into output e-file schemas; executing a mapping operation to generate an instruction-to-schema mapping; executing a tax form descriptions operation that generates form descriptions and per-line descriptions of each of the tax forms from the instruction-to-schema mapping using a first generative model; receiving a tax article identifying a tax regulation change; executing an impacted forms operation that uses the form descriptions and the identified tax regulation change as inputs to a second generative model that identifies an impacted tax form; executing an impacted lines operation that uses a third generative model to identify an impacted line from the impacted form; and generating a notification including the impacted line in the impacted form.

Aspects of the present disclosure provide systems, methods, apparatus, and computer-readable storage media that support relevance-based analysis and filtering of documents and media for one or more enterprises. Aspects disclosed herein leverage custom-built taxonomies, natural language processing (NLP), and machine learning (ML) for identifying and extracting features from highly-relevant documents. The extracted features are vectorized and then filtered based on entities (e.g., enterprises, organizations, individuals, etc.) and compliance-based risks (e.g., illegal or non-compliant activities) that are highly relevant to a particular client. The filtered feature vectors are used to identify and highlight relevant information in the corresponding documents, enabling decision making to resolve compliance-related risks. The aspects described herein generate fewer false positive or otherwise less relevant results than conventional document screening applications or manual techniques.

Aspects of the present disclosure involve an automated, machine-learning technique for generating a representation of an ontology of a corpus of documents. This unsupervised generation of the ontology of the content of the documents may describe, based on the semantics of the language in the corpus and on the structure and format of the documents in that corpus, potentially key differentiable topics and sub-topics within the documents and the potential relationship between the topics and sub-topics. The unsupervised, or automated, generation of the ontology may provide a foundation of potential topics and sub-topics of a corpus of documents from which a complete ontology for the corpus of documents may be created. This ontology may be both pertinent in defining a structure through which an end user may interpret the data identified from a document or set of documents and/or to inform a machine-learning model to extract document information and classification.

Systems and methods are disclosed for generating graphical views of searches. For example, a set of search results may be identified based on a search of legal content maintained at a data source and the search results may be output to a graphical user interface (GUI). Information associated with particular search results of the set of search results may be displayed in response to inputs corresponding to activation of interactive elements of the GUI are received, and the disclosed systems and methods may track one or more metrics associated with the interactions corresponding to each displayed search result. A graphical view of the displayed search results (i.e., ones viewed by the researcher) may be generated based on the one or more metrics and output to a display device. The tracked metrics may be used to quantify a relevancy of the search results, which may be indicated in the graphical view.