International Business Machines Corporation

Pathirage D S U Perera

Sheng Hua Bao

Ramani Ranjan Routray

Cartic Ramakrishnan

Eitan D Farchi

Orna Raz

Marcel Zalmanovici

Vivek Krishnamurthy

A computer system trains a machine learning model. A vector representation is generated for each document in a collection of documents. The documents are clustered based on the vector representations of the documents to produce a plurality of clusters. A training set is produced by selecting one or more documents from each cluster, wherein the selected documents represent a sample of the collection of documents to train the machine learning model. The machine learning model is trained by applying the training set to the machine learning model. Embodiments of the present invention further include a method and program product for training a machine learning model in substantially the same manner described above.

Generating training sets to train machine learning models

A method, a system, and a computer program product are provided. A machine learning model is generated to process adverse event information and produce multiple corresponding medical codes associated with the adverse event information, wherein the multiple medical codes are semantically and hierarchically related in a medical taxonomy. The machine learning model includes multiple parallel output layers, each of which is associated with a corresponding medical code. The machine learning model is trained with training data elements, each of which includes adverse event information mapped to respective multiple medical codes, wherein results from each of the output layers adjusts the machine learning model. After completing the training, information pertaining to an adverse event is applied to the machine learning model to determine the corresponding multiple medical codes within the medical taxonomy.

Architecture for machine learning model to leverage hierarchical semantics between medical concepts in dictionaries

A method, a system, and a computer program product are provided. A training set of adverse event text fragments assigned to medical codes is analyzed to determine first text fragments having frequently occurring medical code assignments and second text fragments having infrequently occurring medical code assignments. The training set is modified to undersample the first text fragments and to oversample the second text fragments such that the text fragments of the modified training set correspond to a substantially uniform assignment of the medical codes. At least one machine learning model is generated and trained with the modified training set. Some parameters of the at least one machine learning model are updated based on errors detected during the training. After completing the training, an adverse event text fragment is applied to the at least one machine learning model to assign at least one medical code.

Automatic classification of adverse event text fragments

Growing community of inventors

San Jose, CA, United States of America

Pathirage D S U Perera