International Business Machines Corporation

Bharath Dandala

Lakshminarayanan Krishnamurthy

Branimir Konstantinov Boguraev

Ravi S Sinha

Swaminathan Chandrasekaran

Benjamin P Segal

Ching-Huei Tsou

Ananya Aniruddha Poddar

Diwakar Mahajan

Alvin C Richardson

Murthy V Devarakonda

Aisha Walcott

Piyush Madan

Divya Pathak

Michal Rosen-Zvi

Venkata Naga Sreeram Joopudi

Parthasarathy Suryanarayanan

Systems and methods for adverse reaction detection. An electronic input is received, the electronic input comprising a drug and outcome pair. The drug and outcome pair is classified as having one of a positive control and a negative control and a probability measure of an adverse drug event relation is determined based on results of the classification. A neural network is trained to jointly leverage reported drug, outcome information and a heterogenous semantic hierarchy, and the classification of a candidate drug is performed using the trained neural network.

Hierarchy-aware adverse reaction embeddings for signal detection

Systems, devices, computer-implemented methods, and/or computer program products that facilitate artificial intelligence (AI)-assisted curation of non-pharmaceutical intervention (NPI) data from heterogeneous data sources. In one example, a system can comprise a processor that executes computer executable components stored in memory. The computer executable components can comprise an extraction component and a change detection component. The extraction component can extract candidate non-pharmaceutical intervention (NPI) events from data associated with a defined disease. The change detection component can evaluate the candidate NPI events for inclusion in a dataset storing NPI events in a defined format.

Artificial intelligence-assisted non-pharmaceutical intervention data curation

A clinical information extraction and training mechanism is provided for automatically extracting and identifying information in plain text narratives in a set of electronic medical records. The mechanism segments each clinical note in a plurality of clinical notes into one or more identified sections, labels each identified section with an associated tag, and generate a tag data structure utilizing explicitly tagged sequences of sentences and associated tags. The mechanism performs statistical analysis of the identified sections that contain tags identified in the tag data structure to identify one or more valid stop/start conditions; extracts a first set of positive examples of sentences for a selected type of information, and then trains a cognitive system to identify sentences in the plurality of clinical notes that fail to have a tag associated with the selected type using the positive examples of sentences for different types of information.

Identifying information in plain text narratives EMRs

Text data including at least named entities can be received. From the named entities, continuous entities, overlapping entities and disjoint entities can be identified. The overlapping entities can be transformed into continuous entities. The continuous entities, the transformed entities and the disjoint entities can be encoded. The encoded entities can be input to a machine learning language model to train the machine learning model to predict candidate entities. The predicted entities can be decoded to reconstruct the predicted entities.

Lightweight tagging for disjoint entities

An approach is provided to receive, at a question answering (QA) system, a question and identify a politeness corresponding to a number of terms corresponding to the question that are included in a corpus of the QA system. The approach identifies the politeness of one or more terms included in each of a set of candidate answers responsive to the question. Finally, the approach scores each of the candidate answers, with the scoring being based, in part, on the politeness identified for each of the terms.

Augment politeness in question answer systems

An approach is provided for automatically predicting an event occurrence based on a question from an end user presented using a near-real-time natural language processing (NLP) analysis to generate, score and rank a plurality of event occurrences based on a plurality of question context parameters extracted from the question, one or more user profile parameters for the end user, and the one or more historical questions, answers, and events having a specified spatial and/or temporal proximity to the question which are identified by an information handling system. In the approach, performed by an information handling system, a top ranked event occurrence from the ranked plurality of event occurrences is selected for inclusion in a notification message that is communicated or broadcast to the end user, as well as other users engaged with the information handling system and/or first responders in the affected area.

Near-real-time prediction, classification, and notification of events in natural language systems

Natural language processing (NLP) with awareness of textual polarity. An NLP system, such as a search engine or a Question-Answering (QA) system receives input text for processing. The input text may be a text fragment, a search phrase, a question having a general type, or a polar question having a yes or no answer. The NLP system identifies textual polarity and provides responses to the input text (for example, in answer form) based on identifying evidence whose selection, scoring, and processing, is informed by the textual polarity of the input text, and the textual polarity of candidate evidence passages.

Natural language processing based on textual polarity

Automatic near-real-time prediction, classification, and notification of events in natural language systems

An approach is provided that returns a simplified set of text to a user of a natural language processing (NLP) system with the simplified set of text having a complexity appropriate to the reading level of the user. The approach receives a word that belongs to a first natural language and retrieves a first set of complexity data pertaining to the word in the first natural language. The approach translates the word to one or more translated words, with each of the translated words corresponding to one or more second natural languages. The approach then retrieves sets of complexity data, with the sets of complexity data corresponding to a different translated word. The approach determines a complexity of the word in the first natural language based on an analysis of the first and second sets of complexity data.

Enhancing QA system cognition with improved lexical simplification using multilingual resources

An approach is provided that receives a word that belongs to a first natural language and retrieves a first set of complexity data pertaining to the word in the first natural language. The approach translates the word to one or more translated words, with each of the translated words corresponding to one or more second natural languages. The approach then retrieves sets of complexity data, with the sets of complexity data corresponding to a different translated word. The approach determines a complexity of the word in the first natural language based on an analysis of the first and second sets of complexity data.

Using multilingual lexical resources to improve lexical simplification

A mechanism is provided in a data processing system for question answering using multi-instance learning. The mechanism trains an answer ranking multi-instance learned model using a ground truth question and answer-key pairs set. When used for answering questions, the mechanism receives an input question from a user and generates one or more candidate answers to the input question. Each of the one or more candidate answers has an associated set of supporting passages. The mechanism determines a confidence value for each of the one or more candidate answers using an answer ranking multi-instance learned model based on the sets of supporting passages. The mechanism ranks the one or more candidate answers by confidence value to form a ranked set of answers, classifies supporting passages to identify the ones which truly support the answer, and presents a final answer from the ranked set of answers, the confidence value for the final answer, and supporting evidence for the final answer to the user.

Growing community of inventors

White Plains, NY, United States of America

Bharath Dandala