US 12,591,749 B2 · Machine learning system for multi-domain long document clustering

Abstract

A model pipeline has been created that allows an organization to gain a clear understanding of contents of a collection of documents despite varying lengths and multiple domain content. A collection of documents with content of different domains are normalized by summarizing the documents. For summarizing, the pipeline uses a language model that has been trained for text summarization across domains to a constrained summary length or length limit. The pipeline extracts embeddings of the summaries and clusters the summary embeddings. The pipeline uses the summaries to label the clusters using another language model that has been trained to generate a label or name for a cluster based on a set of summaries corresponding to a sample of cluster members. The labeled clusters can then be used to generate an organized presentation of the content of the documents in the document collection.

Claims21

Dependent claims

A method comprising:

summarizing each of a plurality of documents with a first language model, wherein the first language model has been tuned to generate summaries of approximately a same length and has been tuned with documents of different domains at least including natural language documents and program code documents;
obtaining embeddings of the summaries;
clustering the embeddings which generates a plurality of clusters;
and
labeling each cluster of the plurality of clusters based on the summaries of those of the plurality of documents corresponding to at least a subset of members of the cluster, wherein labeling each cluster of the plurality of clusters comprises generating labels with a second language model based on the summaries of those of the plurality of documents corresponding to at least a subset of members of the cluster.

2 more elements

An apparatus comprising:

a processor;
a machine-readable medium having instructions stored thereon that are executable by the processor to cause the apparatus to,
generate an organized presentation of a plurality of documents across multiple domains with a model pipeline, wherein the instructions to generate the organized presentation of the plurality of documents comprise instructions to,
generate a summary of each of the plurality of documents with a first language model that has been trained to generate summaries constrained to a defined limit;
for each summary, obtain an embedding of the summary;
hierarchically cluster the embeddings;
for each leaf cluster, select a subset of members of the leaf cluster and generate a label for the leaf cluster with a second language model based on those of the summaries corresponding to the selected subset;
and
for each parent cluster, select a subset of the members of the parent cluster and generate a label for the parent cluster with the second language model based on labels of the selected subset of members of the parent cluster.

6 more elements

Dependencies are inferred from the claim text.

Machine learning system for multi-domain long document clustering

Life of this patent

Machine learning system for multi-domain long document clustering

Life of this patent

Abstract

Claims21