Time series pattern extraction apparatus and method

• US 7,464,060 B2
• Filed: 09/30/2005
• Issued: 12/09/2008
• Est. Priority Date: 11/30/2004
• Status: Expired due to Fees

• Current Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Original Assignee: Kabushiki Kaisha Toshiba (Toshiba Corporation)
• Inventors: Suyama, Akihiro, Orihara, Ryohei, Sakurai, Shigeaki, Ueno, Ken
• Primary Examiner(s): Holmes; Michael B
• Application Number:
  US11/239,422
• Publication Number:
  US 20060129554A1
• Time in Patent Office: 1,166 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06F 16/2474 Sequence data queries, e.g. querying versioned data
• CPC Class Codes: G06F 16/2477 Temporal data queries
• CPC Class Codes: G06Q 30/02 Marketing; Price estimation or determination; Fundraising
• CPC Class Codes: G16H 15/00 ICT specially adapted for medical reports, e.g. generation or transmission thereof

Abstract

A time series pattern generation unit extracts a key concept corresponding to a text of each data element of a plurality of time series data by referring to an extraction rule memory, and generates a plurality of time series patterns. A colony generation unit unifies a plurality of data elements having the same key concept and substantially the same time duration, and generates colonies of unified elements. In response to a selection of a first colony, a colony extraction unit specifies a second colony having a key concept contrary to the first colony, and extracts colonies earlier than the first colony and the second colony. A time reason extraction unit extracts a time reason as a divergence condition of the first colony from the second colony from the extracted colonies.

Claims

1. An apparatus for extracting a time series pattern, comprising:

an input unit inputting a plurality of time series data, each time series data being an electronic daily report including a plurality of data elements, each data element including a time and a text;

a data memory storing the plurality of time series data;

an extraction rule memory storing a correspondence relation between a key concept and a text expression;

a time series pattern generation unit configured to extract key concepts corresponding to the text of each data element of the plurality of time series data from said extraction rule memory, and to generate a plurality of time series patterns from the plurality of time series data, each pattern including data elements each having a time and a key concept;

a colony generation unit configured to unify a plurality of data elements having the same key concept and substantially the same time duration, and to generate colonies of unified data elements;

a select unit selecting a first colony from the colonies in response to a user'"'"'s operation, the first colony having a first key concept;

a colony extraction unit configured to specify a second colony having a second key concept contrary to the first key concept, and to extract colonies each having a time earlier than the first colony and the second colony;

a time reason extraction unit configured to extract a time reason as a divergence condition of the first colony from the second colony from the extracted colonies; and

a display unit displaying the first colony, the second colony, the extracted colonies, and the time reason.

2. The apparatus according to claim 1, wherein said extraction rule memory hierarchically stores the correspondence relation among an upper level key concept, a lower level key concept, and the text expression.

3. The apparatus according to claim 1, wherein said time series pattern generation unit replaces a text of each data element of the time to a series data with the key concept corresponding text expression included in the text, and generates the plurality of time series patterns in place of the plurality of time series data, each time series pattern including a plurality of data elements, each data element including the time and the key concept.

4. The apparatus according to claim 3, wherein said colony generation unit calculates a minimum and a maximum time duration between data elements having the same key concept and previous data elements having the same key concept, and unifies the data elements of the same key concept of which minimums and maximums are respectively equal as the same colony.

5. The apparatus according to claim 4, wherein said select unit selects the first colony of which key concept is an analysis object in response to the user'"'"'s operation.

6. The apparatus according to claim 5, wherein said colony extraction unit traces other colonies in retroactive time series from the first colony or the second colony.

7. The apparatus according to claim 6, wherein said time reason extraction unit extracts a third colony from which the first colony and the second colony originally diverge from the other colonies, calculates a first time duration from the third colony to the first colony, calculates a second time duration from the third colony to the second colony, and generates the time reason based on a difference between the first time duration and the second time duration.

8. The apparatus according to claim 7, wherein said time reason extraction unit calculates a probability of transition from each colony to a next colony, and calculates a support ratio of transition from an origin colony to each colony in all transitions of the colonies, and wherein said display unit additionally displays the probability and the support ratio with an arrow between two colonies.

9. A method for extracting a time series pattern, comprising:

inputting a plurality of time series data, each time series data being an electronic daily report including a plurality of data elements, each data element including a time and a text;

storing the plurality of time series data in a data memory;

storing a correspondence relation between a key concept and a text expression in an extraction rule memory;

extracting key concepts corresponding to the text of each data element of the plurality of time series data from the extraction rule memory;

generating a plurality of time series patterns from the plurality of time series data, each pattern including data elements each having a time and a key concept;

unifying a plurality of data elements having the same key concept and substantially the same time durations;

generating colonies of unified data elements;

selecting a first colony from the colonies in response to a user'"'"'s operation, the first colony having a first key concept;

specifying a second colony having a second key concept contrary to the first key concept;

extracting colonies each having a time earlier than the first colony and the second colony;

extracting a time reason as a divergence condition of the first colony from the second colony from the extracted colonies; and

displaying the first colony, the second colony, the extracted colonies, and the time reason.

10. The method according to claim 9, further comprising:

hierarchically storing the correspondence relation among an upper level key concept, a lower level key concept, and the text expression in the extraction rule memory.

11. The method according to claim 9, at the first generating step, further comprising:

replacing a text of each data element of the time series data with the key concept corresponding to a text expression included in the text; and

generating the plurality of time series patterns in place of the plurality of time series data, each time series pattern including a plurality of data elements, each data element including the time and the key concept.

12. The method according to claim 11, at the unifying step, further comprising:

calculating a minimum and a maximum time durations between data elements having the same key concept and previous data elements having the same key concept; and

unifying the data elements of the same key concept of which minimums and maximums are respectively equal as the same colony.

13. The method according to claim 12, at the indicating step, further comprising:

selecting the first colony of which key concept is an analysis object in response to the user'"'"'s operation.

14. The method according to claim 13, at the extracting step, further comprising:

tracing other colonies in retroactive time series from the first colony or the second colony.

15. The method according to claim 14, at the second extracting step, further comprising:

extracting a third colony from which the first colony and the second colony originally diverge from the other colonies;

calculating a first time duration from the third colony to the first colony;

calculating a second time duration from the third colony to the second colony; and

generating the time reason based on a difference between the first time duration and the second time duration.