Modeling sequence and time series data in predictive analytics

• US 7,747,641 B2
• Filed: 04/28/2005
• Issued: 06/29/2010
• Est. Priority Date: 07/09/2004
• Status: Active Grant

• Current Assignee: Microsoft Technology Licensing LLC (Microsoft Corporation)
• Original Assignee: Microsoft Corporation
• Inventors: Tang, ZhaoHui, MacLennan, C. James, Kim, Pyungchul
• Primary Examiner(s): Rones; Charles
• Assistant Examiner(s): Khoshnoodi; Fariborz
• Application Number:
  US11/116,832
• Publication Number:
  US 20060010142A1
• Time in Patent Office: 1,888 Days
• Field of Search: 707/1, 707/100, 707/102, 707/10, 707/2, 707/3
• US Class Current: 707/769
• CPC Class Codes: G06F 16/2465 Query processing support for facilitating data mining operations in structured databases
• CPC Class Codes: G06F 16/2474 Sequence data queries, e.g. querying versioned data

Abstract

The subject invention relates to systems and methods to extend the capabilities of declarative data modeling languages. In one aspect, a declarative data modeling language system is provided. The system includes a data modeling language component that generates one or more data mining models to extract predictive information from local or remote databases. A language extension component facilitates modeling capability in the data modeling language by providing a data sequence model or a time series model within the data modeling language to support various data mining applications.

Claims

1. A declarative data modeling language system for predicting sequences and time series data automatically, and by identifying patterns without manual pattern identification or validation, comprising:

a processor and memory;

a data modeling language component that automatically generates at least one data mining model to extract predictive information from at least one database, and in a manner that does not require manual identification or validation of a predictive pattern;

a plurality of language extension components configured in the data modeling language, the plurality of language extension components providing at least;

a data sequence model in the data modeling language to generate sequence predictions;

a time series model in the data modeling language and facilitating generating time series predictions of at least one of a casual or discrete subsequent data value in a time series, wherein the sequence model and the time series model are separate models, and in which the data sequence model predicts events based at least in part on historical event data, and the time series model predicts numerical time values based on historical numerical time value data;

wherein one or both of the data sequence model or the time series model include schema rowsets stores that include contents of a mining model according to a transition matrix for clustering sequences and storing probabilities of transitions between different states;

wherein the schema rowsets include All, Cluster and Sequence, in which;

All is a node that is a root and represents a model;

Cluster is a child of All; and

Sequence is a child of All that stores a marginal transition matrix, and in which each Cluster has a Sequence child that contains a set of children, each of which is a column in the transition matrix; and

wherein the memory configured to the processor retains at least one piece of information that pertains to the data modeling language component or the language extension components when directed to the processor.

2. The system of claim 1, wherein the data modeling language is associated with a relational database language.

3. The system of claim 2, wherein the relational database language is a structured query language.

4. The system of claim 1, wherein the data mining models are employed with at least one automated application that queries a database for sequence model predictions or time series model predictions.

5. The system of claim 1, wherein the data sequence model is associated with a data sequence that includes at least one of a DNA sequence, a web sequence, a weather sequence, or a buying sequence.

6. The system of claim 5, wherein the data sequence model is associated with a sequence clustering algorithm.

7. The system of claim 6, wherein the sequence clustering algorithm is employed to cluster user navigation patterns with respect to a web site.

8. The system of claim 1, wherein the time series model is associated with continuous variables.

9. The system of claim 1, wherein the time series model is associated with an Online Analytical Processing (OLAP) cube or a relational data structure.

10. The system of claim 1, wherein the time series model is associated with a data component that is associated with a time parameter.

11. The system of claim 1, wherein the data sequence model is associated with a predict sequence function.

12. The system of claim 1, wherein the time series model is associated with a predict time series function.

13. The system of claim 1, wherein the data sequence model or the time series model are associated with a nested table or a flat t table.

14. The system of claim 13, wherein the nested table is associated with a probability.

15. The system of claim 14, wherein the nested table is associated with a state transition schema.

16. A computer implemented method for generating data mining models and predicting sequences and time series data automatically, and by identifying patterns without manual pattern identification or validation, comprising:

at a computing system, executing computer-executable instructions using one or more processors, wherein execution of the computer-executable instructions directs the computing system to;

provide a plurality of language extensions to a database modeling language, the plurality of language extensions including;

at least one data sequence model in the database modeling language to generate sequence predictions;

at least one time series model in the database modeling language to generate time series predictions, wherein the at least one data sequence model and the at least one time series model are separate models, and in which the data sequence model predicts events based at least in part on historical event data, and the time series model predicts numerical values based on historical numerical value data;

wherein the schema rowsets include All, Cluster and Sequence, in which;

All is a node that is a root and represents a model;

Cluster is a child of All; and

Sequence is a child of All that stores a marginal transition matrix, and in which each Cluster has a Sequence child that contains a set of children, each of which is a column in the transition matrix; and

automatically, with a computing system and without manual pattern identification or validation, generate data mining models from the plurality of language extensions;

generate a query for a database; and

automatically generate at least one sequence prediction and at least one time series prediction from the database based on the query and the data mining models, wherein the sequence prediction predicts a future event and the time series prediction predicts a future numerical time value and is based on at least one of casual data or discrete data.

17. The method of claim 16, further comprising generating a session identifier key, a sequence identifier or a sequence table in response to the query.

18. The method of claim 16, further comprising generating a probability prediction based at least in part on continuous data, a flat rowset, or a nested rowset.

19. A system to facilitate data mining operations and predict sequences at time series data automatically, and by identifying patterns without manual pattern identification or verification, comprising:

one or more computer-readable media having stored thereon computer executable instructions that, when executed by a processor, cause the system to;

A processor and memory;

query a relational database;

generate a data mining model to determine predictive information from the database;

modify the data mining model to each of a casual data time series, discrete data time series, and a data sequence;

generate probabilities from the database in view of the data time series or the data sequence, such that;

probabilities associated with the casual data time series and the discrete data time series predict future numerical time values based on historical numerical time values; and

wherein the schema rowsets include All, Cluster and Sequence, in which;

All is a node that is a root and represents a model;

Cluster is a child of All; and

Sequence is a child of All that stores a marginal transition matrix, and in which each Cluster has a Sequence child that contains a set of children, each of which is a column in the transition matrix; and

probabilities associated with the data sequence predict future events based on historical event data; and

wherein the memory configured to the processor to the one or more computer-readable media and which, upon request, executes at least one command in relation to the aforementioned querying or generating.

20. The system of claim 19, wherein:

the data mining model is adaptable to a relational database structure and includes schema rowsets stores that include contents of a mining model according to a transition matrix for clustering sequences and storing probabilities of transitions between different states, and in which generating the probabilities includes evaluating schema considerations and algorithm parameters;

algorithm parameters usable with the data mining model for data sequence prediction include;

a parameter for a maximum number of clusters to be formed;

a parameter for a minimum size of each cluster;

a parameter for a maximum number of states for a clustering algorithm attribute; and

an integer-type parameter allowing setting of a seed for a cluster; and

algorithm parameters with the data mining model for time sequence prediction include;

a parameter that provides hints about periodicities in mined data;

a parameter that defines a number of future steps cached for prediction purposes during training;

a parameter specifying a gap between historical models;

a parameter specifying how to handle a missing time slice; and

a floating parameter in which a higher value indicates a higher likelihood of finding patterns.