Learning topics by simulation of a stochastic cellular automaton

• US 10,394,872 B2
• Filed: 11/04/2015
• Issued: 08/27/2019
• Est. Priority Date: 05/29/2015
• Status: Active Grant

• Current Assignee: Oracle International Corporation (Oracle Corporation)
• Original Assignee: Oracle International Corporation (Oracle Corporation)
• Inventors: Tristan, Jean-Baptiste, Green, Stephen J., Steele, Jr., Guy L., Zaheer, Manzil
• Primary Examiner(s): Reyes, Mariela
• Assistant Examiner(s): Harmon, Courtney
• Application Number:
  US14/932,825
• Publication Number:
  US 20160350411A1
• Time in Patent Office: 1,392 Days
• CPC Class Codes: G06F 16/3344 using natural language analysis
• CPC Class Codes: G06F 16/353 into predefined classes
• CPC Class Codes: G06F 16/951 Indexing; Web crawling techniques

Abstract

Herein is described an unsupervised learning method to discover topics and reduce the dimensionality of documents by designing and simulating a stochastic cellular automaton. A key formula that appears in many inference methods for LDA is used as the local update rule of the cellular automaton. Approximate counters may be used to represent counter values being tracked by the inference algorithms. Also, sparsity may be used to reduce the amount of computation needed for sampling a topic for particular words in the corpus being analyzed.

Claims

1. A method for identifying sets of correlated words comprising:

receiving information for a set of documents;

wherein the set of documents comprises a plurality of words;

wherein a particular document of the set of documents comprises a particular word of the plurality of words;

running an inference algorithm over a Dirichlet distribution of the plurality of words in the set of documents to produce sampler result data, further comprising;

retrieving a first counter value from a first data structure,

based, at least in part, on the first counter value, assigning a particular topic, of a plurality of topics, to the particular word in the particular document to produce a topic assignment for the particular word,

after assigning the particular topic to the particular word, updating a second counter value in a second data structure to produce an updated second counter value,

wherein the updated second counter value reflects the topic assignment, and

wherein the first data structure is stored and accessed independently from the second data structure; and

determining, from the sampler result data, one or more sets of correlated words;

wherein the method is performed by one or more computing devices.

2. The method of claim 1, wherein the inference algorithm is one of:

Mean-for-Mode Gibbs sampling, and collapsed variational Bayesian inference.

3. The method of claim 1, further comprising:

performing one or more of a particular group of steps in parallel with one or more of the steps of retrieving, assigning, and updating;

wherein the particular group of steps comprises;

retrieving a third counter value from the first data structure;

based, at least in part, on the third counter value, assigning a second topic, of the plurality of topics, to a second word, of the plurality of words, to produce a topic assignment for the second word; and

after assigning the second topic to the second word, updating a fourth counter value, in the second data structure, to reflect the topic assignment for the second word;

wherein the fourth counter value reflects the topic assignment for the second word.

4. The method of claim 1, wherein one or more of said first counter value and said second counter value are represented as approximate counter values.

5. The method of claim 1, further comprising:

prior to performing an iteration of the inference algorithm that includes said steps of retrieving, assigning, and updating;

calculating values for one or more tables based on counts in the first data structure;

wherein assigning the particular topic to the particular word comprises performing Walker'"'"'s “

alias method”

based on the one or more tables.

6. The method of claim 1, wherein assigning the particular topic to the particular word is performed without respect to one or more of parameters theta and phi.

7. The method of claim 1, wherein assigning the particular topic to the particular word is performed without respect to stored values indicating previous assignments of words to topics.

8. One or more non-transitory computer-readable media storing instructions, which, when executed by one or more processors, cause:

receiving information for a set of documents;

wherein the set of documents comprises a plurality of words;

wherein a particular document of the set of documents comprises a particular word of the plurality of words;

running an inference algorithm over a Dirichlet distribution of the plurality of words in the set of documents to produce sampler result data, further comprising;

retrieving a first counter value from a first data structure,

based, at least in part, on the first counter value, assigning a particular topic, of a plurality of topics, to the particular word in the particular document to produce a topic assignment for the particular word,

after assigning the particular topic to the particular word, updating a second counter value in a second data structure to produce an updated second counter value,

wherein the updated second counter value reflects the topic assignment, and

wherein the first data structure is stored and accessed independently from the second data structure; and

determining, from the sampler result data, one or more sets of correlated words.

9. The one or more non-transitory computer-readable media of claim 8, wherein the inference algorithm is one of:

Mean-for-Mode Gibbs sampling, and collapsed variational Bayesian inference.

10. The one or more non-transitory computer-readable media of claim 8, wherein the instructions further comprise instructions, which, when executed by one or more processors, cause:

performing one or more of a particular group of steps in parallel with one or more of the steps of retrieving, assigning, and updating;

wherein the particular group of steps comprises;

retrieving a third counter value from the first data structure;

based, at least in part, on the third counter value, assigning a second topic, of the plurality of topics, to a second word, of the plurality of words, to produce a topic assignment for the second word; and

after assigning the second topic to the second word, updating a fourth counter value, in the second data structure, to reflect the topic assignment for the second word;

wherein the fourth counter value reflects the topic assignment for the second word.

11. The one or more non-transitory computer-readable media of claim 8, wherein one or more of said first counter value and said second counter value are represented as approximate counter values.

12. The one or more non-transitory computer-readable media of claim 8, wherein the instructions further comprise instructions, which, when executed by one or more processors, cause:

prior to performing an iteration of the inference algorithm that includes said steps of retrieving, assigning, and updating;

calculating values for one or more tables based on counts in the first data structure;

wherein assigning the particular topic to the particular word comprises performing Walker'"'"'s “

alias method”

based on the one or more tables.

13. The one or more non-transitory computer-readable media of claim 8, wherein assigning the particular topic to the particular word is performed without respect to one or more of parameters theta and phi.

14. The one or more non-transitory computer-readable media of claim 8, wherein assigning the particular topic to the particular word is performed without respect to stored values indicating previous assignments of words to topics.