Predicting future trending topics

• US 10,380,249 B2
• Filed: 10/02/2017
• Issued: 08/13/2019
• Est. Priority Date: 10/02/2017
• Status: Active Grant

• Current Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Original Assignee: Meta Platforms, Inc. (f/k/a Facebook, Inc.)
• Inventors: Tiwari, Parth
• Primary Examiner(s): Lerner, Martin
• Application Number:
  US15/723,095
• Publication Number:
  US 20190102374A1
• Time in Patent Office: 680 Days
• Field of Search: 704 1, 704 9, 707724, 707725, 707727, 707750
• CPC Class Codes: G06F 18/24 Classification techniques
• CPC Class Codes: G06F 40/284 Lexical analysis, e.g. tokenisation or collocates
• CPC Class Codes: G06F 40/30 Semantic analysis
• CPC Class Codes: G06Q 50/01 Social networking

Abstract

A prediction system can predict future trending topics. The prediction system can classify social media posts by region and vertical, extract text from the posts, tokenize the extracted text, and organizing the tokens into n-grams. The prediction system can store the n-grams from the posts in a cumulative set of n-grams, with each n-gram tagged with the originating post'"'"'s identified region, vertical, and a time value. The prediction system can compute, for each n-gram, a frequency within each category defined by a region/vertical pair. The prediction system can fit occurrence data for n-grams to a polynomial and identify the slope of the point on for the current time. The slope can be used as a prediction of growth or decline for the n-gram. The prediction system can identify n-grams with a comparatively large slope within that region/vertical as likely to be trending in the future.

Claims

1. A method for identifying future trending n-grams, comprising:

for at least one particular content item of multiple content items;

extracting text from the particular content item;

identifying a plurality of classifications for the particular content item, the plurality of classifications including a geographical region classification and a subject-based classification;

organizing the extracted text into one or more n-grams;

adding the one or more n-grams to a cumulative set of n-grams, wherein each n-gram in the cumulative set is associated with a time-based value for the particular content item;

sorting the n-grams in the cumulative set into groups by the plurality of classifications of the content item that the n-gram originated from;

computing a frequency value, within each group, for each unique n-gram in that group;

selecting unique n-grams, for at least one of the groups, that have a frequency value above a frequency threshold;

computing a predicted change in frequency value for the selected unique n-grams, the computing for a given unique n-gram comprising fitting a polynomial to the time-based values for the n-grams that have the same sequence of words as the given unique n-gram and that are in the same group as the given unique n-gram, wherein the computed predicted change in frequency is a slope of the polynomial at a point corresponding to a current time; and

selecting, as the future trending n-grams, for a geographical region specified in the geographical region classification, one or more n-grams with a predicted change in frequency value above a predicted change threshold.

2. The method of claim 1, wherein the geographical region classification for the content item is identified based on region data for a user who provided the content item or for a device the content item originated from.

3. The method of claim 1, wherein the subject-based classification for the content item is identified based on the extracted text from the particular content item.

4. The method of claim 1, wherein extracting text from the particular content item comprises one or more of:

converting audio associated with the particular content item to text;

performing text recognition on an image associated with the particular content item;

performing text recognition on video associated with the particular content item.

5. The method of claim 1, wherein organizing the extracted text into one or more n-grams comprises:

normalizing the extracted text;

tokenizing the normalized text; and

grouping the tokenized text into groups of sequential tokens, the groups having a fixed number of tokens.

6. The method of claim 5, wherein the fixed number of tokens is two tokens.

7. The method of claim 5, wherein at least two of the groups of sequential tokens are overlapping in the normalized text.

8. The method of claim 1 further comprising:

identifying at least one invalid n-gram, wherein each particular invalid n-gram is identified as invalid based on an amount of words, of the particular invalid n-gram that match words on a pre-defined stop word list, being above a stop-word threshold; and

removing from the cumulative set the identified invalid n-grams.

9. A non-transitory computer-readable storage medium storing instructions that, when executed by a computing system, cause the computing system to perform operations for identifying one or more future trending n-grams, the operations comprising:

for at least one particular content item of multiple content items;

identifying a plurality of classifications for the particular content item, the plurality of classifications includes a geographical region classification and a subject-based classification;

organizing text associated with the particular content item into one or more n-grams;

adding, from the one or more n-grams into to a cumulative set of n-grams, at least one n-gram;

computing a frequency value for each unique n-gram in the cumulative set of n-grams, the frequency value computed for a frequency within the group of n-grams in the cumulative set of n-grams that have the same one or more classifications;

computing a predicted change in frequency value for at least some of the unique n-grams, the computing for a given n-gram comprising fitting a polynomial to time-based values associated with the n-grams in the cumulative set that have the same sequence of words as the given unique n-gram and that have the same one or more classifications as the given unique n-gram, wherein the computed change in frequency is a slope of the polynomial at a point corresponding to a current time; and

selecting, as the future trending n-grams, for a geographical region specified in the geographical region classification, one or more n-grams with a predicted change in frequency value above a predicted change threshold.

10. The non-transitory computer-readable storage medium of claim 9, wherein at least one of the one or more classifications for each particular content item is identified by performing natural language topic recognition on the text associated with the particular content item.

11. The non-transitory computer-readable storage medium of claim 9, wherein the operations further comprise:

receiving an indication of user input choosing a selected geographical region and a selected subject; and

in response to the indication of user input, providing a subset of the selected future trending n-grams whose one or more classifications include both the region classification matching the selected geographical region and the subject-based classification matching the selected subject.

12. The non-transitory computer-readable storage medium of claim 11, wherein at least one chosen n-gram of the provided subset of future trending n-grams is used to generate marketing materials prior to the chosen n-gram reaching a peak in trending among users of a social media system.

13. The non-transitory computer-readable storage medium of claim 9, wherein the operations further comprise selecting the at least some of the unique n-grams to be used in predicting a change frequency by:

selecting unique n-grams, for at least one of the groups, that have a frequency value above a frequency threshold.

14. The non-transitory computer-readable storage medium of claim 9, wherein the geographical region classification for the content item is identified based on region data for a user who provided the content item or for a device the content item originated from.

15. The non-transitory computer-readable storage medium of claim 9, wherein the subject based classification of the content item is identified based on text associated with the particular content item.

16. The non-transitory computer-readable storage medium of claim 9, wherein the operations further comprise extracting text from each particular content item, extracting the text from each particular content item including one or more of:

converting audio associated with the particular content item to text;

performing text recognition on an image associated with the particular content item;

or

performing text recognition on video associated with the particular content item.

17. The non-transitory computer-readable storage medium of claim 16, wherein organizing the text into one or more n-grams comprises:

normalizing the text;

tokenizing the normalized text; and

grouping the tokenized text into groups of sequential tokens, the groups having a fixed number of tokens.

18. The non-transitory computer-readable storage medium of claim 9, wherein the operations further comprise:

identifying at least one invalid n-gram, wherein each particular invalid n-gram is identified as invalid based on an amount of words, of the particular invalid n-gram that match words on a pre-defined stop word list, being above a stop-word threshold; and

removing from the cumulative set the identified invalid n-grams.