Predicting a user behavior number of a word

• US 8,849,738 B2
• Filed: 07/29/2011
• Issued: 09/30/2014
• Est. Priority Date: 08/02/2010
• Status: Active Grant

• Current Assignee: Alibaba Group Holding Ltd.
• Original Assignee: Alibaba Group Holding Ltd.
• Inventors: Zhang, Tao, Guo, Jiaqing, Guo, Ning
• Primary Examiner(s): Chaki, Kakali
• Assistant Examiner(s): Zidanic, Michael
• Application Number:
  US13/318,439
• Publication Number:
  US 20130144822A1
• Time in Patent Office: 1,159 Days
• Field of Search: 706/46
• US Class Current: 706/46
• CPC Class Codes: G06F 16/90 Details of database functions independent of the retrieved data types
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06Q 10/00 Administration; Management
• CPC Class Codes: G06Q 30/00 Commerce

Abstract

The present disclosure introduces a method, an apparatus and memory of predicting a user behavior number of a word for reducing the amount and the complexity of operation, saving the consumption of the equipment, and improving the accuracy and reliability of predictions. In an embodiment, a historical data sequence of the user behavior number of a word is converted from a time domain to a frequency domain. Based on the converted frequency domain, each estimated cycle and its effect rate value of the historical data sequence are ascertained. If the historical data sequence is stable, an average value of user behavior numbers of some historical data points before a prediction point is calculated as a user behavior number of the prediction point. Otherwise, the user behavior number is calculated based on a selected main cycle and a selected singularity.

Claims

1. A method of predicting a user behavior number of a word, the method comprising:

converting a historical data sequence of user behavior numbers associated with the word from a time domain to a frequency domain;

ascertaining one or more estimated cycles of the historical data sequence, and an effect rate value of each of the one or more estimated cycles based on the frequency domain of the historical data sequence;

determining whether the historical data sequence is stable based on each of the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, the determining including determining whether the effect rate value of each of the one or more estimated cycles exceeds a configured effect rate threshold;

if the historical data sequence is stable, calculating a user behavior number of a prediction point based on an average value of the user behavior numbers of the historical data sequence before the prediction point; and

if the historical data sequence is not stable;

selecting a main cycle and a singularity of the historical data sequence based on the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, the selecting including;

selecting an estimated cycle of the one or more estimated cycles as the main cycle, the estimated cycle being within a configured main cycle range and having a largest effect rate value; and

selecting another estimated cycle of the one or more estimated cycles as the singularity, an effect rate value of the another estimated cycle being larger than effect rate values of other estimated cycles of the one or more estimated cycles, the other estimated cycles excluding the estimated cycle selected as the main cycle, the one or more estimated cycles including multiple estimated cycles; and

calculating the user behavior number of a prediction point based on the selected main cycle and the selected singularity.

2. The method as recited in claim 1, wherein the calculating of the user behavior number of the prediction point based on the selected main cycle and the selected singularity comprises:

forming a training data sequence comprising a set of data associated with individual historical data points of the historical sequence after a point corresponding to the selected singularity; and

calculating the user behavior number of the prediction point based on a model of the training data sequence, the model being created by using a time sequence model.

3. The method as recited in claim 1, wherein the calculating of the user behavior number of the prediction point based on the selected main cycle and the selected singularity comprises:

forming a first training data sequence comprising a set of data associated with individual historical data points of the historical data sequence after a historical data point corresponding to the selected singularity;

obtaining a cycle average value by averaging a set of first training data of the first training sequence, each of the set of first training data corresponding to one or more main cycle locations;

calculating a set of second training data by subtracting the cycle average value from each of the set of first training data;

forming a second training sequence without a cycle that comprises the set of second training data;

calculating one user behavior number of the prediction point without the cycle based on a model of the second training data sequence, the model being created by using a time sequence model; and

calculating the user behavior number of the prediction point by adding the one user behavior number and the cycle average value corresponding to one of the one or more main cycle locations.

4. The method as recited in claim 3, further comprising:

selecting an average value of training data of the second training sequence without the cycle as the one user behavior number of the prediction point without the cycle after determining a deviation of the user behavior number of the prediction point exceeds a configured deviation threshold, the training data being associated with one main cycle before the prediction point.

5. The method as recited in claim 1, wherein the converting from the time domain to the frequency domain of the historical data sequence comprises converting by using a Fast Fourier Transformation (FFT) or a Wavelet Transformation.

6. The method as recited in claim 1, wherein the user behavior number of the word comprises traffic or a click rate of the word.

7. The method as recited in claim 1, wherein the one or more estimated cycles have two or more different cycles.

8. An apparatus of predicting a user behavior number of a word, comprising:

one or more processors;

a computer storage medium having stored thereon computer executable instructions that are executable by the one or more processors to form multiple instruction devices comprising;

a conversion unit to convert a historical data sequence of user behavior numbers of the word from a time domain to a frequency domain;

a deciding unit to decide one or more estimated cycles of the historical data sequence, and an effect rate value of each of the one or more estimated cycles based on the frequency domain of the historical data sequence;

a determination unit to determine whether the historical data sequence is stable based on each of the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, the determining including determining whether the effect rate value of each of the one or more estimated cycles exceeds a configured effect rate threshold;

a first prediction unit to calculate a user behavior number of a prediction point based on an average value of user behavior numbers of the historical data sequence before the prediction point, if the historical data sequence is stable;

a selection unit to select a main cycle and a singularity of the historical data sequence based on the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, if the historical data sequence is not stable, the selection unit including;

a storing sub-unit to store a configured main cycle range;

a first selection sub-unit to select a first estimated cycle of the one or more estimated cycles as the main cycle, the first estimated cycle being within the configured main cycle range and having a largest effect rate value; and

a second selection sub-unit to select a second estimated cycle of the one or more estimated cycles as the singularity, an effect rate value of the second estimated cycle being larger than effect rate values of other estimated cycles of the one or more estimated cycles, the other estimated cycles excluding the first estimated cycle, the one or more estimated cycles comprising multiple estimated cycles; and

a second prediction unit to calculate the user behavior number of a prediction point based on the selected main cycle and the selected singularity.

9. The apparatus as recited in claim 8, wherein the second prediction unit comprises:

a choosing sub-unit to form a training data sequence comprising a set of data associated with individual historical data points of the historical sequence after a historical data point corresponding to the selected singularity; and

a prediction sub-unit to calculate the user behavior number of the prediction point based on a model of the training data sequence, the model being created by using a time sequence model.

10. The apparatus as recited in claim 8, wherein the second prediction unit comprises:

a choosing sub-unit to form a first training data sequence comprising a set of data associated with individual historical data points of the historical data sequence after a point corresponding to the selected singularity;

an operation sub-unit to obtain a cycle average value by averaging a set of first training data of the first training sequence, each of the set of first training data being corresponding to one or more main cycle locations;

a de-cycle process sub-unit to;

calculate a second training data sequence without a cycle by subtracting, from each training data of the training data sequence, the cycle average value corresponding to individual main cycle locations; and

form a second training sequence without the cycle that comprises the set of second training data;

a prediction sub-unit to calculate one user behavior number of the prediction point without the cycle based on a model of the training data sequence, the model being created by using a time sequence model; and

a cycle recovery process sub-unit to calculate the user behavior number of the prediction point by adding the one user behavior number and the cycle average value corresponding to one of the one or more main cycle locations.

11. The apparatus as recited in claim 10, wherein the second prediction unit comprises:

a re-prediction sub-unit to;

select an average value of training data of the second training sequence without a cycle as the user behavior number of the prediction point without the cycle after determining that a deviation of the behavior number of the prediction point exceeds a configured deviation threshold, the training data being associated with one main cycle before the prediction point.

12. The apparatus as recited in claim 8, wherein the one or more estimated cycles have two or more different cycles.

13. The apparatus as recited in claim 8, wherein the user behavior number of the word comprises traffic or a click rate of the word.

14. The apparatus as recited in claim 8, wherein the converting from the time domain to the frequency domain of the historical data sequence comprises converting by using a Fast Fourier Transformation (FFT) or a Wavelet Transformation.

15. A computer storage medium having stored thereon computer executable instructions configured to predict a user behavior number of a word, that are executable by one or more processors to perform acts comprising:

converting a historical data sequence of user behavior numbers associated with the word from a time domain to a frequency domain;

ascertaining one or more estimated cycles of the historical data sequence, and an effect rate value of each of the one or more estimated cycles based on the frequency domain of the historical data sequence;

determining whether the historical data sequence is stable based on each of the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, the determining including determining whether the effect rate value of each of the one or more estimated cycles exceeds a configured effect rate threshold;

if the historical data sequence is stable, calculating a user behavior number of a prediction point based on an average value of the user behavior numbers of the historical data sequence before the prediction point; and

if the historical data sequence is not stable;

selecting a main cycle and a singularity of the historical data sequence based on the one or more estimated cycles and the effect rate value of each of the one or more estimated cycles, the selecting including;

selecting an estimated cycle of the one or more estimated cycles as the main cycle, the estimated cycle being within a configured main cycle range and having a largest effect rate value; and

selecting another estimated cycle of the one or more estimated cycles as the singularity, an effect rate value of the another estimated cycle being larger than effect rate values of other estimated cycles of the one or more estimated cycles, the other estimated cycles excluding the estimated cycle selected as the main cycle, the one or more estimated cycles including multiple estimated cycles; and

calculating the user behavior number of a prediction point based on the selected main cycle and the selected singularity.

16. The computer storage medium as recited in claim 15, wherein the calculating of the user behavior number of the prediction point based on the selected main cycle and the selected singularity comprises:

forming a training data sequence comprising a set of data associated with individual historical data points of the historical sequence after a point corresponding to the selected singularity; and

calculating the user behavior number of the prediction point based on a model of the training data sequence, the model being created by using a time sequence model.

17. The computer storage medium as recited in claim 15, wherein the calculating of the user behavior number of the prediction point based on the selected main cycle and the selected singularity comprises:

forming a first training data sequence comprising a set of data associated with individual historical data points of the historical data sequence after a historical data point corresponding to the selected singularity;

obtaining a cycle average value by averaging a set of first training data of the first training sequence, each of the set of first training data being corresponding to one or more main cycle locations;

calculating a set of second training data by subtracting the cycle average value from each of the set of first training data;

forming a second training sequence without a cycle that comprises the set of second training data;

calculating one user behavior number of the prediction point without the cycle based on a model of the second training data sequence, the model being created by using a time sequence model; and

calculating the user behavior number of the prediction point by adding the one user behavior number and the cycle average value corresponding to one of the one or more main cycle locations.

18. The computer storage medium as recited in claim 15, wherein the converting from the time domain to the frequency domain of the historical data sequence comprises converting by using a Fast Fourier Transformation (FFT) or a Wavelet Transformation.

19. The computer storage medium as recited in claim 15, wherein the user behavior number of the word comprises traffic or a click rate of the word.

20. The computer storage medium as recited in claim 15, wherein the one or more estimated cycles have two or more different cycles.