Methods and systems for profile-based forecasting with dynamic profile selection

• US 7,664,671 B2
• Filed: 12/01/2004
• Issued: 02/16/2010
• Est. Priority Date: 12/01/2004
• Status: Expired due to Fees

• Current Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Original Assignee: Hewlett-Packard Development Company, L.P. (HP Inc.)
• Inventors: Shan, Jerry Z.
• Primary Examiner(s): Jarrett; Scott L
• Application Number:
  US11/001,455
• Publication Number:
  US 20060116921A1
• Time in Patent Office: 1,903 Days
• Field of Search: 705/10
• US Class Current: 705/7.11
• CPC Class Codes: G06Q 10/063 Operations research, analysis or management
• CPC Class Codes: G06Q 10/0631 Resource planning, allocation, distributing or scheduling for enterprises or organisations
• CPC Class Codes: G06Q 10/06315 Needs-based resource requirements planning or analysis
• CPC Class Codes: G06Q 10/0635 Risk analysis of enterprise or organisation activities
• CPC Class Codes: G06Q 10/06375 Prediction of business process outcome or impact based on a proposed change
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G06Q 30/0202 Market predictions or forecasting for commercial activities

Abstract

Disclosed herein are systems and methods for profile-based forecasting with dynamic profile selection. Some method embodiments may comprise determining a reference set of profiles from a source set of profiles, and using the reference set of profiles to generate a forecast. The reference set determination comprises at least comparing a current, partial profile to each profile in the source set to obtain a similarity measurement for each profile in the source set.

Claims

1. A method comprising:

(a) determining, by a processor, whether there is sufficient information that has been acquired so far about a current time period to perform similarity measurements;

(b) in response to determining that there is not sufficient information that has been acquired so far about the current time period to perform similarity measurements,

operating in a first mode in which the processor selects a reference set of profiles from a source set of profiles, using a plurality of initial selection rules;

(c) in response to determining that there is sufficient information that has been acquired so far about the current time period to perform similarity measurements,

operating in a second mode in which the processor selects the reference set of profiles by, for each previous profile of one or more previous profiles indicative of prior time periods,

obtaining a similarity measurement through the comparison, by the processor, a current partial profile indicative of the information acquired so far about the current time period to the previous profile;

where the similarity measurement denotes low dissimilarity between the current partial profile and the previous profile, adding, by the processor, the previous profile to the reference set of profiles;

where the similarity measurement denotes high dissimilarity between the current partial profile and the previous profile, removing, by the processor, the previous profile to the reference set of profiles if present; and

,

(d) thereafter, using, by the processor, the reference set of profiles and the current partial profile to generate a forecast,

repeating (a), (b), and (c) one or more times before performing (d), such that the first mode is initially used to select the reference set of profiles until sufficient information has been acquired about the current period to perform similarity measurements, and such that the second mode is thereafter used to refine the reference set of profiles after sufficient information has been acquired about the current period to perform similarity measurements; and

wherein the first mode is used to select the reference set of profiles in a rule-based manner where insufficient information has been acquired so far about the current time period to perform similarity measurements, and the second mode is used to select the reference set of profiles in a similarity measurement-based manner where sufficient information has been acquired so far about the current time period to perform similarity measurements.

2. The method of claim 1, and wherein each profile comprises:

a past history of a selected attribute that is to be forecast in the future;

a pattern of the selected attribute over time;

identification of other attributes relevant to forecasting the selected attribute; and

,

identification of other, non-attribute information relevant to forecasting the selected attribute.

3. The method of claim 1, wherein the initial selection rules are event-based, in that the initial selection rules govern selection of the reference set of profiles from the source set of profiles based on one or more events that are to occur within the current time period.

4. The method of claim 3, wherein the events are selected from:

holidays within the current time period;

sales promotions within the current time period; and

, weather information for the current time period.

5. The method of claim 1, wherein the similarity measurement is

$D_i={({\int }_{}^{}}^{}$

0 T ⁢



S ⁡

( t ) - S i ⁡

( t ) 

p ⁢

⁢

ⅆ

t ) 1 p ,

where D_i is a difference in distance between the current partial profile and the previous profile, T is a current time, S(t) is the current partial profile, S_i(t) is the previous profile, and p is a constant.

6. The method of claim 5, wherein p equals 2 so that the similarity measurement is a Euclidean distance measurement.

7. A computer comprising:

a display;

a processor coupled to the display; and

,

a memory couple do the processor,

wherein the memory stores software that upon being executed by the processor generates a forecast using a reference set of profiles and a current partial profile,

wherein the reference set of profiles is generated in accordance with one or more of a first mode and a second mode based on whether there is sufficient information that has been acquired so far about a current time period,

wherein in the first mode, the reference set of profiles are selected from a source set of profiles using a plurality of initial selection rules,

wherein in the second mode, the reference set of profiles are selected by, for each previous profile of one or more previous profiles indicative of prior time periods,

obtaining a similarity measurement through the comparison of a current partial profile indicative of the information acquired so far about the current time period to the previous profile;

where the similarity measurement denotes low dissimilarity between the current partial profile and the previous profile, adding the previous profile to the reference set of profiles;

where the similarity measurement denotes high dissimilarity between the current partial profile and the previous profile, removing the previous profile to the reference set of profiles if present,

and repeating the above steps one or more times, such that the first mode is initially used to select the reference set of profiles until sufficient information has been acquired about the current period to perform similarity measurements, and the second mode is thereafter used to refine the reference set of profiles after sufficient information has been acquired about the current period to perform similarity measurements,

wherein the first mode is used to select the reference set of profiles in a rule-based manner where insufficient information has been acquired so far about the current time period to perform similarity measurements, and the second mode is used to select the reference set of profiles in a similarity measurement-based manner where sufficient information has been acquired so far about the current time period to perform similarity measurements.

8. The computer of claim 7, and wherein each profile comprises:

a past history of a selected attribute that is to be forecast in the future;

a pattern of the selected attribute over time;

identification of other attributes relevant to forecasting the selected attribute; and

,

identification of other, non-attribute information relevant to forecasting the selected attribute.

9. The computer of claim 7, wherein the initial selection rules are event-based, in that the initial selection rules govern selection of the reference set of profiles from the source set of profiles based on one or more events that are to occur within the current time period.

10. The computer of claim 9, wherein the events are selected from:

holidays within the current time period;

sales promotions within the current time period; and

, weather information for the current time period.

11. The computer of claim 7, wherein the similarity measurement is

$\begin{array}{c}{D}_i={({\int }_{}^{}}^{}\end{array}$

0 T ⁢



S ⁡

( t ) - S i ⁡

( t ) 

p ⁢

⁢

ⅆ

t ) 1 p ,

where D_i is a difference in distance between the current partial profile and the previous profile, T is a current time, S(t) is the current partial profile, S_i(t) is the previous profile, and p is a constant.

12. The computer of claim 11, wherein p equals 2 so that the similarity measurement is a Euclidean distance measurement.

13. A computer-readable medium storing software that is executable by a processor to:

determine whether there is sufficient information that has been acquired so far about a current time period to perform similarity measurements;

in response to determining that there is not sufficient information that has been acquired so far about the current time period to perform similarity measurements,

operate in a first mode in which a reference set of profiles is selected from a source set of profiles, using a plurality of initial selection rules;

in response to determining that there is sufficient information that has been acquired so far about the current time period to perform similarity measurements,

operate in a second mode in which the reference set of profiles is selected by, for each previous profile of one or more previous profiles indicative of prior time periods,

obtaining a similarity measurement by comparing a current partial profile indicative of the information acquired so far about the current time period to the previous profile;

where the similarity measurement denotes low dissimilarity between the current partial profile and the previous profile, adding the previous profile to the reference set of profiles;

where the similarity measurement denotes high dissimilarity between the current partial profile and the previous profile, removing the previous profile to the reference set of profiles if present; and

,

(d) thereafter, use the reference set of profiles and the current partial profile to generate a forecast,

repeating (a), (b), and (c) one or more times before performing (d), such that the first mode is initially used to select the reference set of profiles until sufficient information has been acquired about the current period to perform similarity measurements, and wherein the second mode is thereafter used to refine the reference set of profiles after sufficient information has been acquired about the current period to perform similarity measurements, and

wherein the first mode is used to select the reference set of profiles in a rule-based manner where insufficient information has been acquired so far about the current time period to perform similarity measurements, and the second mode is used to select the reference set of profiles in a similarity measurement-based manner where sufficient information has been acquired so far about the current time period to perform similarity measurements.

14. The medium of claim 13, and wherein each profile comprises:

a past history of a selected attribute that is to be forecast in the future;

a pattern of the selected attribute over time;

identification of other attributes relevant to forecasting the selected attribute; and

,

identification of other, non-attribute information relevant to forecasting the selected attribute.

15. The method of claim 13, wherein the initial selection rules are event-based, in that the initial selection rules govern selection of the reference set of profiles from the source set of profiles based on one or more events that are to occur within the current time period.

16. The method of claim 15, wherein the events are selected from:

holidays within the current time period;

sales promotions within the current time period; and

, weather information for the current time period.

17. The method of claim 13, wherein the similarity measurement is

$\begin{array}{c}{D}_i={({\int }_{}^{}}^{}\end{array}$

0 T ⁢



S ⁡

( t ) - S i ⁡

( t ) 

p ⁢

⁢

ⅆ

t ) 1 p ,

where D_i is a difference in distance between the current partial profile and the previous profile, T is a current time, S(t) is the current partial profile, S_i(t) is the previous profile, and p is a constant.

18. The method of claim 17, wherein p equals 2 so that the similarity measurement is a Euclidean distance measurement.