# US 2020/0004870 A1 · Identifying homogenous clusters

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2020/0004870 A1
- Document ID: US20200004870A1
- Status: Published application
- Kind: A1
- Application number: 16/025,936
- Filed: 2018-07-02
- Published: 2020-01-02
- Earliest claimed: 2018-07-02
- Assignees: salesforce.com, inc.
- Applicants: salesforce.com, inc.
- Inventors: Ganeswara Rao DULAM
- Web page: https://patentgrep.com/patent/US20200004870A1

## CPC

- G06F 16/35: Clustering; Classification
- G06F 16/285: {Clustering or classification}
- G06F 16/93: Document management systems
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/23213: with fixed number of clusters, e.g. K-means clustering
- G06F 40/30: Semantic analysis
- G06N 7/01: Probabilistic graphical models, e.g. probabilistic networks
- G06F 40/284: Lexical analysis, e.g. tokenisation or collocates
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Homogeneous clusters are generated from a first plurality of documents for generation of regular expressions. Documents that share similar characteristics are clustered, and for each cluster, features are generated for use by a homogeneity model to determine a homogeneity score for the cluster. Clusters determined to be homogenous are sent to a regular expression generator.

## Claims

### Claim 1

A computer implemented method for generating homogeneous clusters from a first plurality of documents for generation of regular expressions, the method comprising:
clustering the first plurality of documents into a first plurality of clusters, wherein each of the first plurality of documents is included in only one of the clusters, wherein each of the clusters includes one or more of the documents and wherein the documents in each cluster share certain characteristics more closely with each other than the documents of the other clusters in the first plurality of clusters;
for each cluster in the first plurality of clusters:
generating a word distribution for each document in the cluster;
assigning, using the word distribution, each word to a probability group;
determining the percentage of words in each probability group;
determining features for the cluster using the probability groups;
determining a homogeneity score by applying a homogeneity model to the features for the cluster; and
send those of the first plurality of clusters for which the homogeneity score exceeds a homogeneity threshold to an automatic regular expression generator.

### Claim 2 (depends on 1)

The method of claim 1, wherein the word distribution indicates in which documents in the cluster that each word in each document occurs.

### Claim 3 (depends on 1)

The method of claim 1, wherein the percentage indicates what percentage of documents in the cluster each word occurs at least once.

### Claim 4 (depends on 1)

The method of claim 1, wherein the probability groups bin each word within a predetermined number of probability groups.

### Claim 5 (depends on 4)

The method of claim 4, wherein the predetermined number of probability groups is 10.

### Claim 6 (depends on 1)

The method of claim 1, wherein the homogeneity model is a logistic regression model.

### Claim 7 (depends on 1)

The method of claim 1, further comprising:
assigning documents in a second plurality of documents to the first plurality of clusters using the regular expressions, wherein the first plurality of documents comprises a plurality of error messages and wherein each cluster corresponds to one or more related software bugs and wherein the second plurality of documents corresponds to a more recent plurality of error messages.

### Claim 8

A non-transitory machine-readable storage medium that provides instructions for generating homogeneous clusters from a first plurality of documents for generation of regular expressions that, if executed by a processor, will cause said processor to perform operations comprising:
clustering the first plurality of documents into a first plurality of clusters, wherein each of the first plurality of documents is included in only one of the clusters, wherein each of the clusters includes one or more of the documents and wherein the documents in each cluster share certain characteristics more closely with each other than the documents of the other clusters in the first plurality of clusters;
for each cluster in the first plurality of clusters:
generating a word distribution for each document in the cluster;
assigning, using the word distribution, each word to a probability group;
determining the percentage of words in each probability group;
determining features for the cluster using the probability groups;
determining a homogeneity score by applying a homogeneity model to the features for the cluster; and
send those of the first plurality of clusters for which the homogeneity score exceeds a homogeneity threshold to an automatic regular expression generator.

### Claim 9 (depends on 8)

The non-transitory machine-readable storage medium of claim 8, wherein the word distribution indicates in which documents in the cluster that each word in each document occurs.

### Claim 10 (depends on 8)

The non-transitory machine-readable storage medium of claim 8, wherein the percentage indicates what percentage of documents in the cluster each word occurs at least once.

### Claim 11 (depends on 8)

The non-transitory machine-readable storage medium of claim 8, wherein the probability groups bin each word within a predetermined number of probability groups.

### Claim 12 (depends on 11)

The non-transitory machine-readable storage medium of claim 11, wherein the predetermined number of probability groups is 10.

### Claim 13 (depends on 8)

The non-transitory machine-readable storage medium of claim 8, wherein the homogeneity model is a logistic regression model.

### Claim 14 (depends on 8)

The non-transitory machine-readable storage medium of claim 8, the operations further comprising:
assigning documents in a second plurality of documents to the first plurality of clusters using the regular expressions, wherein the first plurality of documents comprises a plurality of error messages and wherein each cluster corresponds to one or more related software bugs and wherein the second plurality of documents corresponds to a more recent plurality of error messages.

### Claim 15

An article of manufacture for generating homogeneous clusters from a first plurality of documents for generation of regular expressions, the article comprising:
a processor; and
a memory coupled to the processor, the memory storing instructions which, when executed by the processor, cause the article to:
cluster the first plurality of documents into a first plurality of clusters, wherein each of the first plurality of documents is included in only one of the clusters, wherein each of the clusters includes one or more of the documents and wherein the documents in each cluster share certain characteristics more closely with each other than the documents of the other clusters in the first plurality of clusters;
for each cluster in the first plurality of clusters:
generate a word distribution for each document in the cluster;
assign, using the word distribution, each word to a probability group;
determine the percentage of words in each probability group;
determine features for the cluster using the probability groups;
determine a homogeneity score by applying a homogeneity model to the features for the cluster; and
send those of the first plurality of clusters for which the homogeneity score exceeds a homogeneity threshold to an automatic regular expression generator.

### Claim 16 (depends on 15)

The article of claim 15, wherein the word distribution indicates in which documents in the cluster that each word in each document occurs.

### Claim 17 (depends on 15)

The article of claim 15, wherein the percentage indicates what percentage of documents in the cluster each word occurs at least once.

### Claim 18 (depends on 15)

The article of claim 15, wherein the probability groups bin each word within a predetermined number of probability groups.

### Claim 19 (depends on 18)

The article of claim 18, wherein the predetermined number of probability groups is 10.

### Claim 20 (depends on 15)

The article of claim 15, the instructions further causing the article to:
assigning documents in a second plurality of documents to the first plurality of clusters using the regular expressions, wherein the first plurality of documents comprises a plurality of error messages and wherein each cluster corresponds to one or more related software bugs and wherein the second plurality of documents corresponds to a more recent plurality of error messages.
