Trend detection in a messaging platform

Patent metadata

Field	Value
Publication number	US-10277693-B2
Application number	US-201615174853-A
Country	US
Kind code	B2
Filing date	Jun 6, 2016
Priority date	Jun 4, 2015
Publication date	Apr 30, 2019
Grant date	Apr 30, 2019

Abstract

A system and method for trend detection in a messaging platform. A trend detection model is selected and a time series having a plurality of instances of social data is received, wherein the instances of social data share a countable parameter. A count is made of occurrences of countable parameters in each instance of social data assigned to that bin and a trend detected based at least in part on the trend detection model and on the count for each bin.

First claim

The invention claimed is:

1. A method comprising:
receiving, by a computing device, a set of historical time series of social data;
labeling, by the computing device, each of the set of historical time series of social data as trending or non-trending;
selecting, by the computing device, and based on the set of historical time series of social data, a trend detection model;
receiving, by the computing device, a time series having a plurality of instances of social data, wherein the instances of social data share a countable parameter;
for each of a number of bins, counting, by the computing device, occurrences of one or more of the countable parameters in each instance of social data assigned to that bin;
determining, by the computing device, based at least in part on the trend detection model, on the count for each bin, and on the set of historical time series of social data and the one or more instances of social data in the time series of social data that correspond to a particular event, a measure of a trend associated with the countable parameter;
determining, by the computing device, a distance between at least one historical time series of the trend detection model and the time series of social data that corresponds to the particular event;
determining, by the computing device, based at least in part on the distance and a scaling parameter, a particular weight;
determining, by the computing device, a trending value based on a trending score generated using the particular weight, the trending value representing a ratio of a first aggregation of weights and a second aggregation of weights, wherein:
the first aggregation of weights is based on a first plurality of weights, the first plurality of weights based at least in part on one or more historical time series of social data labeled as trending,
the second aggregation of weights is based on a second plurality of weights, the second plurality of weights based at least in part on one or more historical time series of social data labeled as non-trending, and
the particular weight is included in at least one of the first aggregation of weights or the second aggregation of weights;

implementing cycle-correction to the measure of the trend, based on pattern information associated with respective time data associated with each of the one or more historical time series of social data labeled, as trending to obtain a cycle-corrected measure of the trend;
determining, by the computing device, that the cycle-corrected measure of the trend satisfies a trend threshold; and
responsive to determining that the cycle-corrected measure of the trend satisfies the trend threshold, outputting, by the computing device, at least one indication of the detected trend.

2. The method of claim 1 , wherein the method further comprises associating the trend with an event.

3. The method of claim 1 , wherein the computing device is a first computing device, wherein outputting the at least one indication of the detected trend comprises at least one of:
sending, by the first computing device and to a second computing device of a user, a notification that indicates the detected trend;
providing, by the first computing device, a graphical user interface that indicates the detected trend; or
sending, by the first computing device and to a third computing device that analyzes detected trends for particular events, the indication of the detected trend.

4. The method of claim 1 , further comprising:
determining, by the computing device and based on a plurality of previous instances of social data that correspond to the countable parameter, an expected count of instances of social data; and
determining, by the computing device, an actual count of the one or more instances of social data in the time series of social data,
wherein determining the measure of the trend includes determining, by the computing device, a trending score based at least in part on the expected count and the actual count for each bin.

5. The method of claim 4 ,
wherein the expected count of instances of social data is based on a time period in a first day, and
wherein the actual count of instances of social data is based on the same time period in a second day.

6. The method of claim 1 , further comprising:
storing, by the computing device and in the trend detection model, each of the labeled historical time series of social data as a set of time series of historical social data.

7. The method of claim 1 , wherein each respective instance of social data in the time series of social data indicates at least one of: text of a social media communication, a count of mentions in a social network for the respective event, one or more hashtags of a social media communication, one or more users following the respective event in a social network, one or more friends of a user in a social network associated with the instance of social data, one or more hyperlinks, or a time-dependent quantity that is countable over time.

8. The method of claim 1 , wherein the trend detection model includes a parameter, and wherein the parameter modifies at least of:
a time difference between a first time that a real-world event occurs and a second time that a detection of the real-world event occurs in the social data;
a fraction of identified trends that are not statistical false positives; or
a fraction of real-world trends, which are detected by the model that indicates the likelihood of a trend.

9. The method of claim 1 , wherein the detected trend represents at least one of:
a continuing increase in a quantity of instances of social data that correspond to a particular event, wherein the continuing increase follows a prior steady state in the quantity of instances of social data that correspond to the particular event;
an increase in the quantity of instances of social data that correspond to a particular event within a first defined time duration, which is greater than a threshold, wherein the increase follows the prior steady state in the quantity of instances of social data that correspond to the particular event; or
an increase in the quantity of instances of social data within a second defined time duration that is greater than a threshold, followed by a decline in the quantity of instances of social data within the second defined time duration, wherein the increase follows the prior steady state in the quantity of instances of social data that correspond to a particular event.

10. The method of claim 1 , wherein selecting a trend detection model includes receiving, from a content provider, an indication of the trend detection model to select, and wherein outputting at least one indication of the detected trend includes transmitting a trend notice and trend data to the content provider.

11. The method of claim 1 , wherein selecting the trend detection model comprises selecting the trend detection model from a plurality of available trend detection models, wherein each respective available trend detection model of the plurality of available trend detection models includes a respective plurality of trend values distributed over time, and wherein each trend value is a function of the count from two or more bins.

12. A computing device comprising:
at least one processor; and
at least one non-transitory computer-readable storage medium storing instructions that are executable by the at least one processor to:
receive a set of historical time series of social data;
label each of the set of historical time series of social data as trending or non-trending;
select, based on the set of historical time series of social data, a trend detection model;
receive a time series

Assignees

• Twitter Inc

Inventors

• Hendrickson Scott
• Kolb Jeffrey Aaron
• Montague Joshua
• Lehman Brian

Classifications

• H04L51/16

  Electricity · mapped topic

• H04L67/306

  User profiles · CPC title

• H04L51/32

  Electricity · mapped topic

• H04L67/02

  based on web technology, e.g. hypertext transfer protocol [HTTP] · CPC title

• G06F17/30864

  Physics · mapped topic