JP5548243B2 - Event occurrence point extraction apparatus, method, and program - Google Patents

Description

  The present invention relates to an event occurrence point extraction device, method, and program, and more particularly, to an event occurrence point extraction device, method, and program for extracting a point where a sudden event occurs.

  There is a technology to extract the things that are currently attracting attention in the world by statistically processing the data on people's behavior gathered through web services such as web search and blog. For example, keywords currently attracting attention are extracted and presented from the frequency of keywords appearing in microblog articles.

  There is a technique of giving attention to events instead of attention to keywords (Patent Document 1). This is realized by creating a keyword list representing an event in advance and obtaining the attention level of the keyword related to the event among the keywords included in the keyword list. In such a technique, since it is necessary to create a keyword list regarding events that occur in advance, it is not possible to extract the attention level of events that occur suddenly, such as incidents, accidents, and guerrilla live events.

  As a method for realizing a system for suddenly extracting events in a person skilled in the art, a simple method using the above-described technique can be considered. Explain the procedure of the simple method. First, the points that are attracting attention are extracted. This can be realized by replacing the event list used in the above method with a point list. Next, the point where the event is occurring is excluded from the points attracting attention. This is possible by using a list in which an event and its occurrence point are paired. In this way, it can be seen that a sudden event has occurred at points that have attracted attention, excluding those where it is known that an event has occurred in advance.

Japanese Patent No. 4496900

  However, when implemented by the simple method described above, two problems arise. The first point is that it is difficult to create an event list that covers all event information. As a result, there is a problem that the event is accidentally extracted as a sudden event because it is not in the event list even though the event is scheduled in advance. The second point is that a sudden event occurring at a point where a previously predicted event is occurring cannot be extracted.

  The present invention has been made in view of the above facts, and it is possible to accurately extract a point where a sudden event occurs without creating an event list that covers all event information. An object is to provide a point extraction apparatus, method, and program.

In order to achieve the above object, an event occurrence point extraction apparatus according to the present invention is an event occurrence point extraction apparatus that extracts a point where a sudden event has occurred, and time information indicating each transmission time is given. From each text in the text set, a place name database that stores a place name and location information indicated by the place name is searched for each place name, and the word information of the text and the text for each of the place names searched for the text Based on the pre-processing means for generating the time information and the location information indicated by the place name and registering it in the real-time point information database, and the geographical range and time information input as the extraction condition of the sudden event, From the real-time location information database, location information is included in the input geographic range, and time information Includes record acquisition means for acquiring a record corresponding to the input time information, and for each position information included in the input geographic range, the position information in the record acquired by the record acquisition means is included. The number of records is stored in the real-time point information database, the position information is included in the input geographic range, and the time information is in a record corresponding to the input time information on another date. The point-of-interest extraction means for determining whether or not the point of interest is included by comparing with the number of records including the position information included, and extracting the position information determined to be the point of interest; and the record acquisition unit And a record including position information that is a point of interest extracted by the point of interest extraction means. For each word in the word information contained in de, and the number of records that contain the words in the word information in the record, the real-time location information stored in the database, the position information and the predetermined is the point of interest Calculated on the basis of all records including time information corresponding to the input time information on the other date of the number of days, the number of days of the other date on which the record including the word exists in the word information, and Based on a predetermined number of days, the attention factor degree of the word is calculated, the attention factor word extracting means for extracting the word having the largest attention factor degree as the attention factor word, and the stored in the real-time location information database, the position information is the point of interest, and the time information in the time zone from the time indicated by the input time information to the time traced back a predetermined time including The number of records in which the attention factor word is included in the word information in all records, the position information that is the attention point stored in the real-time point information database, and the time within the time zone at another date The number of records in which the attention factor word is included in the word information in all records including information , the position information of the attention point stored in the real-time point information database, and the time zone at another date Based on the number of records in which the attention factor word is included in the word information in all the records including the time information within, it is determined whether the attention point is a point where a sudden event has occurred And event occurrence point extraction means for extracting a point where a sudden event has occurred.

The event occurrence point extraction method according to the present invention is an event occurrence point extraction method for extracting a point where a sudden event has occurred, and is a text set to which time information indicating each transmission time is given by a preprocessing means. Each place name obtained from the location information database storing the place name and the position information indicated by the place name is searched from each text, and the word information of the text and the time information of the text are retrieved for each of the texts searched for the place name. And a record including location information indicated by the place name, registered in the real-time point information database, and based on the geographical range and time information input as a sudden event extraction condition by the record acquisition means, the real-time From the point information database, location information is included in the input geographic range, and The record corresponding to the input time information is acquired by the time information, and the position in the record acquired by the record acquisition unit is acquired for each position information included in the input geographic range by the point of interest extraction unit. The number of records including position information is stored in the real-time point information database, the position information is included in the input geographic range, and the time information corresponds to the input time information on another date. By comparing with the number of records including the position information included in the record to be determined, whether or not it is a point of interest, extracting the position information determined to be the point of interest, and attention factor word extraction means The point of interest is a record acquired by the record acquisition unit and extracted by the point of interest extraction unit For each word in the word information included in the record including a location information, and the number of records that contain the words in the word information in the records, stored in the real-time location information database, with the point of interest The other date on which there is a record including the word in the word information, which is obtained based on all records including certain position information and time information corresponding to the input time information on another date of a predetermined number of days Based on the number of days and the predetermined number of days, the attention factor degree of the word is calculated, the word having the largest attention factor degree is extracted as the attention factor word, and the real-time point information database is extracted by the event occurrence point extraction unit stored in, when the a point of interest at a location, going back a predetermined time from the input time information indicating a time The number of records in which the attention factor word is included in the word information in all records including the time information in the time zone until the time, and the position information that is the attention point stored in the real-time point information database; , The number of records in which the attention factor word is included in the word information in all records including the time information in the time zone on other dates , and the attention point accumulated in the real-time point information database Based on the number of records in which the attention factor word is included in the word information in all the records including the position information and the time information in the time zone on another date , a sudden event occurs at the attention point. By determining whether or not the point is a spot, a point where a sudden event occurs is extracted.

  The program which concerns on this invention is a program for functioning a computer as each means of said event occurrence point extraction apparatus.

  As described above, according to the event occurrence point extraction apparatus, method, and program of the present invention, the point of interest is based on the number of records including position information for each position information included in the input geographic range. The position information is extracted, the attention factor is calculated for each word of the word information included in the record including the position information that is the attention point, the attention factor word is extracted, and the position information that is the attention point Based on the time change of the number of records including the attention factor word obtained from the word information of all records including the time information in the past predetermined period before the input time information, the point of interest is the sudden event. A place where a sudden event has occurred without creating an event list that covers all event information by determining whether or not it is an occurrence point It is possible to accurately extract an advantage of being.

It is the schematic which shows the structure of the event occurrence point extraction apparatus which concerns on embodiment of this invention. It is a figure which shows the example of text information. It is a figure which shows the example of the place name expression and the latitude / longitude pair memorize | stored in the point information database. It is a figure which shows the example of the record memorize | stored in the real-time point information database. It is a figure for demonstrating the method of extracting an attention point. It is a flowchart which shows the content of the event occurrence point process routine in the event occurrence point extraction apparatus which concerns on embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. An example in which the present invention is applied to an event occurrence point extraction apparatus that extracts a point where a sudden event has occurred from a large amount of text information transmitted like a microblog will be described.

<Configuration of event occurrence point extraction device>
As shown in FIG. 1, an event occurrence point extraction device 100 according to an embodiment of the present invention executes a CPU (Central Processing Unit), a RAM (Random Access Memory), and an event occurrence point extraction processing routine described later. The computer is provided with a ROM (Read Only Memory) storing a program for the purpose, and is functionally configured as follows. The event occurrence point extraction device 100 includes an input unit 10, a real-time data database 12, a point information database 14, a preprocessing unit 16, a real-time point information database 18, a sudden event information extraction unit 20, and an output unit 22. The sudden event information extraction unit 20 is an example of a record acquisition unit, a point of interest extraction unit, a point of interest word extraction unit, and an event occurrence point extraction unit.

  The input unit 10 accepts a set of text information input from a text information database that accumulates transmitted text information such as a microblog. The input text information holds data of a service such as a microblog where the content of the transmitted information is close to the transmission time. An example of text information is shown in FIG. The text information holds the transmitted information and the time when the information was transmitted.

  In addition, the input unit 10 receives a geographic range and time as an extraction condition for extracting a sudden event, which is input by an input device such as a known keyboard or mouse.

  The real-time data database 12 stores a set of text information received by the input unit 10 together with a transmission time of each text information. In the example of FIG. 2, the transmitted text information is stored in the BODY column, and the transmission time at which the information is transmitted is stored in the LAST_MOD column.

  The point information database 14 stores a plurality of pairs of latitude / longitude associated with place name expressions and place names. An example is shown in FIG. A combination of place name expression and address information included in a Web document collected in advance is extracted, and if it is necessary to convert from address information to latitude and longitude, the converted version is stored in the point information database 14. .

  The pre-processing unit 16 performs morphological analysis on the text information held in the real-time data database 12, and when the morphological analysis result includes the place name expression included in the point information database 14, the transmission time of the text information, the morphological analysis result Is stored in the real-time location information database 18 with the word information representing the noun included in the transmitted text information and the latitude and longitude associated with the place name expression.

  The real-time location information database 18 holds a plurality of records including transmission time, latitude / longitude, and word information. An example is shown in FIG. Each record is created from the real-time data database 12 and the point information database 14 via the preprocessing unit 16.

  The sudden event information extraction unit 20 receives the geographic range and time received by the input unit 10 and passes the extracted occurrence point of the sudden event to the output unit 22. The sudden event information extraction unit 20 is mainly composed of three processes: attention point extraction, attention factor word extraction, and attention pattern determination. Details of each process will be described below.

<Point of interest extraction>
First, a process for extracting a point of interest in the geographic range and time received from the input unit 10 is performed. The point of interest is extracted by using a statement (text information) relating to a certain point held in the real-time point information database 18. Several extraction methods can be considered, for example, a method of extracting a point where the frequency of utterances including point expressions rapidly increases as a point of interest, or a point where the interval between utterances including point expressions is rapidly shortened. A method of extracting as a point can be considered.

  Here, an embodiment in which a point of interest is extracted based on the utterance frequency will be described. First, a specific example is given to explain the basic concept of the extraction method. For example, it is assumed that the time of May 31, 2012, 19:40 and the geographical range representing Tokyo 23 wards are received from the input unit 10. In order to extract the points of interest in Tokyo's 23 wards at 19:40 on May 31, the real-time location information database 18 uses the range from 19:20 to 19:40 on May 31 as the transmission time. And, the utterance (record) including the longitude and latitude of each point in Tokyo 23 wards is obtained, the utterance frequency (number of records) is obtained for each longitude and latitude, and the utterance frequency (records) of the past few weeks for each longitude and latitude Number). Specifically, on May 10, 2012, May 24, 2012, and May 31, 2012, Tokyo that had an outgoing time included between 19:20 and 19:40 and was received as input A record having the latitude and longitude included in the geographical range representing the 23 wards is acquired from the real-time point information database 18. The total number of utterances (number of records) is calculated for each longitude / latitude and each day. Compare the average number. If the number of utterances on May 31 regarding a certain point (longitude and latitude) exceeds twice the average number of utterances on other days, that point (longitude and latitude) is extracted as a point of interest.

This is generalized and described. The time received as the extraction condition is set to t _0,0, and the time going back from t _0,0 by a certain time s (for example, 20 minutes) is set to t _1,0 . Also, a time that goes back a certain period c (for example, one week) from t _0,0 is t _0,1 , and similarly a time that goes back a period c from t _1,0 is t _1,1 . Furthermore, a period from t _{0,0 to} t _1,0 is s _0,0, and similarly a period from t _0,1 to t _1,1 is s _0,1 . Records from the real-time location information database 18 using the geographical range specified as the extraction condition and the periods s _{0, i} (i = 0,1,2, ..., n) based on the time specified as the extraction condition as keys. get. When the number of records related to the point p in the period s _{0, i} is p (s _{0, i} ), a point (longitude / latitude) p that satisfies the discriminant expressed by the following equation (1) is extracted as a point of interest.

Here, θ is a preset threshold value, and a large value is set to increase the extraction accuracy, and a small value is set to increase the coverage. Further, it is preferable to select the time s and the period c so that p (s _{0, i} ) is as constant as possible.

  FIG. 5 shows an arrangement of the time variables used in the above embodiment.

<Retrieving attention factor word>
Next, an attention factor word that is a factor attracting attention is extracted from the remarks about the attention point p extracted by the attention point extraction process. The degree of attention factor is calculated for each word in the word information of the record having the longitude and latitude of the point of interest p and the transmission time in the period s _0,0 , and the word with the highest score is used as the attention factor word Extract. Word w _h of the attention factor degree a (w _h) is calculated according to the following equation (2).

Here, wf _h (s _0,0) has a transmission time between the period s _0,0, and frequency of the word w _h included in the word information record having the longitude and latitude of the point p (record number), n is the (1) period s _{0, i} other day of days obtained records in formula, df _h was in the formula (1) obtained for the period s _{0, i} of other dates period _{s 0, i (i = 1,2} , ..., n) that contains the word w _h to record the words information indicating the frequency of (the number of days the other day).

  Using the attention factor degree calculated by the above equation (2), a word having a particularly high appearance frequency (number of records) can be extracted as the attention factor word as compared with the same time zone on another day.

<Attention pattern judgment>
Finally, attention pattern determination is performed to determine whether or not the attention point attracted attention has been predicted in advance. The determination is made by looking at the appearance distribution of the attention factor word. In data such as microblogs, when an event that has been predicted in advance occurs, relevant remarks can be seen before the spot gets noticeable. On the other hand, when an event that is difficult to predict occurs, there is almost no relevant remark before the spot gets noticeable. By using such characteristics, it is determined that an event that could not be predicted has occurred if the attention factor word has not substantially appeared before the time zone in which the attention point is extracted. In other words, whether or not the occurrence frequency of the attention word at the attention point in the time zone before the start of attracting attention was more or less compared with the same time zone on other days, was predicted in advance Is determined.

Let t _{a, 0} (a = x / s) be a time _dating from a specified time t _0,0 by a certain time x (for example, 12 hours), a period t _{0, i to} t _{a, i} be x _{0, i} , and in the period x _{0, i,} and the point of interest p and attention factor word w _h wf the number of records in real-time location information database 18 containing _h (x _{0, i).} A point of interest p that satisfies the discriminant represented by the following equation (3) is extracted as a point of occurrence of a sudden event.

Here, θ is a preset threshold value, and is set to a large value when it is desired to increase the extraction accuracy of sudden events, and a small value when it is desired to increase the coverage of sudden events. The time zone x _{0, i} (i = 0,..., N) is an example of a past predetermined period before the input time information.

In accordance with the discriminant of the above equation (3), the time zone x _0,0 before starting to attract attention based on the temporal change of the appearance frequency (number of records) of the attention factor word in the time zone x _{0, i} If the appearance frequency is higher than the appearance frequency (number of records) of the attention factor word in that time zone x _{0, k} on the day of the day, it is determined that the event has been predicted in advance, and the occurrence frequency may be as usual. It is determined that a sudden event has occurred.

  The output unit 22 outputs the latitude and longitude of the point where the sudden event occurs, received from the sudden event information extraction unit, and presents it to the user.

<Operation of event occurrence point extraction device>
Next, the operation of the event occurrence point extraction apparatus 100 according to the present embodiment will be described. First, when a set of text information to which a transmission time is given is input to the event occurrence point extraction device 100, it is stored in the real-time data database 12. In the event occurrence point extraction apparatus 100, the preprocessing unit 16 stores a record in the real-time point information database 18 based on the real-time data database 12 and the point information database 14. Then, when the geographic range and time as the extraction condition are input to the event occurrence point extraction device 100, the event occurrence point extraction device 100 executes an event occurrence point extraction processing routine shown in FIG.

First, in step S101, the geographic range and time input by the input unit 10 are received. Then, in step S102, the real-time point is set by using each period s _{0, i} (i = 0,1,2,..., N) based on the geographical range received in step S101 and the time received in step S101 as a key. A corresponding record is acquired from the information database 18.

In the next step S103, whether or not each point of the longitude and latitude of the record corresponding to the geographical range and the period s _0,0 acquired in step S102 is a point of interest based on the discriminant expressed by the equation (1). Is determined.

And the process after step S104 mentioned later is performed about each point of the longitude latitude of the record corresponding to the said geographical range and period _s0,0 acquired by said step S102.

  In step S104, it is determined whether or not the point to be processed is a point of interest. If the point to be processed does not satisfy the discriminant expressed by the above equation (1), the point to be processed is determined. Does nothing. On the other hand, when the point to be processed satisfies the discriminant represented by the above equation (1), the process proceeds to step S105.

In step S105, calculated for each word w _h word information record corresponding to the latitude and longitude and time s _0,0 of the point of interest p to be processed, according to the above (2), attention factor degree a a (w _h) To do. Specifically, for each word w _h, having a transmission time between the period s _0,0, and the frequency (number of records) of a word w _h included in the word information record having the longitude and latitude of the point p Is determined as wf _h (s _0,0 ). For each word w _h , the word w _h is added to the word information of the record acquired from the real-time point information database 18 using the target point of interest p and each period s _{0, i} (i = 1, 2,..., N) as keys. The frequency df _h (the number of days on another day) of the period s _{0, i} (i = 1, 2,..., N) in which is included is obtained. Then, for each word w _h, according to the above (2), thereby calculating the target factor of a (w _h).

In the next step S106, based on the attention factor degree a (w _h) of each word w _h calculated in step S105, attention factor degree a (w _h) is the highest word is extracted as a target factor words. Then, based on the discriminant represented by the above formula (3), it is determined whether or not the time-series change in the appearance frequency of the attention factor word (swell of the attention factor word) is sudden. Specifically, the period x transmission time in the _0,0, obtains latitude and longitude of the point of interest p, and a record including the target factor word w _h from the real-time location information database 18, the acquired record number wf _h (x _{0 , 0} ). A period _{x 0, k (k = 1,2} , ..., n) for the period x _0, transmission time in _k, latitude and longitude of the point of interest p to be processed, and the record including the target factor word w _h The number of records wf _h (x _{0, k} ) acquired from the real-time location information database 18 is obtained. And it is discriminate | determined whether the discriminant shown by said Formula (3) is satisfy | filled.

  In step S107, it is determined whether or not the target point p to be processed satisfies the discriminant relating to the target pattern determination. If the discriminant represented by the above equation (3) is not satisfied, the target point to be processed is determined. Do nothing for p. On the other hand, when the point of interest p to be processed satisfies the discriminant represented by the above equation (3), the process proceeds to step S108. In step S108, the point of interest p to be processed is output by the output unit 22 as a point where a sudden event has occurred, and the event occurrence point extraction processing routine ends.

  As described above, according to the event occurrence point extraction device according to the present embodiment, the number of records including the latitude and longitude in the real-time point information database for each latitude and longitude included in the geographic range input as the extraction condition. The attention point is extracted, the attention factor degree is calculated for each word of the word information included in the record including the latitude and longitude of the attention point, the attention factor word is extracted, and the latitude that is the attention point The point of interest based on the time change of the number of records including the attention factor word, which is obtained from the word information of all records including the longitude and time information in the past predetermined period before the time information input as the extraction condition Create an event list that covers all event information by determining whether or not a point where a sudden event has occurred It without a point where unexpected event has occurred can be accurately extracted.

  In the conventional method, sudden events were extracted by removing points included in the event list from points that attracted attention, but it was difficult to create a comprehensive event list and an event occurred. There was a problem that sudden events could not be extracted at certain locations. In this embodiment, it is further scheduled in advance without creating an exhaustive event list by determining whether or not the cause of the attention is due to a sudden event at a spot that attracts attention. A sudden event can be extracted even at a point where the event has occurred.

  In addition, it is possible to present the point where an unexpected event that is not known in advance, such as an incident, accident, or guerrilla live, and that attracts people's attention, is currently occurring. .

  The present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, the present invention can be realized by installing a program on a known computer via a medium or a communication line.

  In addition, the event occurrence point extraction apparatus described above has a computer system inside, but the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. Shall be.

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium.

DESCRIPTION OF SYMBOLS 10 Input part 12 Real time data database 14 Point information database 16 Pre-processing part 18 Real time point information database 20 Sudden event information extraction part 22 Output part 100 Event occurrence point extraction apparatus

Claims (3)

An event occurrence point extraction device that extracts points where sudden events occur,
Each location name obtained from the location information database storing the location name and the location information indicated by the location name is searched from each text of the text set to which the time information indicating the transmission time is given, and each text from which the location name is searched A pre-processing means for generating a record including word information of the text, time information of the text, and position information indicated by the place name, and registering it in a real-time location information database;
Based on the geographic range and time information input as an extraction condition for sudden events, the time information from the real-time point information database includes location information in the input geographic range and the time information is input. Record acquisition means for acquiring records corresponding to
For each position information included in the input geographic range, the number of records including the position information in the records acquired by the record acquisition means is stored in the real-time point information database, and the position information is the By comparing with the number of records including the position information included in the record corresponding to the input time information on another date and included in the input geographical range , A point-of-interest extracting means for determining whether or not there is, and extracting position information determined to be the point of interest;
For each word of the word information included in the record that is the record acquired by the record acquisition unit and includes the position information that is the point of interest extracted by the point of interest extraction unit, the word in the record The number of records in which the word is included in the word information, the position information that is the point of interest, and the time information corresponding to the time information input on another date of a predetermined number of days, stored in the real-time point information database. The attention factor of the word is calculated based on the number of days of the other date on which the record including the word exists in the word information and the predetermined number of days, which is obtained based on all the records including An attention factor word extracting means for extracting a word having the highest degree as an attention factor word;
The real-time location information stored in the database, the and the point of interest at a location, in all records containing the time information in the time zone from the input time information indicating time to time going back a predetermined time All including the number of records in which the attention factor word is included in the word information, and the position information that is the point of interest, and the time information in the time zone at other dates, stored in the real-time point information database The number of records in which the attention factor word is included in the word information in the record of the record , the position information that is the point of interest accumulated in the real-time point information database, and time information in the time zone at another date, based on the number of records that contain the attention factor word to the word information in all records, including, the attention area But by determining whether a point a sudden event has occurred, an event occurrence point extracting means for extracting a point where sudden event has occurred,
An event occurrence point extraction device including An event occurrence point extraction method for extracting a point where a sudden event has occurred,
The pre-processing means searches each place name obtained from the location information database storing the place name and the position information indicated by the place name from each text of the text set to which the time information indicating the sending time is given, and the place name is searched. For each of the texts generated, generate a record including the word information of the text, the time information of the text, and the position information indicated by the place name, and registered in the real-time point information database,
Based on the geographical range and time information input as a sudden event extraction condition by the record acquisition means, location information is included in the input geographical range from the real-time location information database, and time information is Get the record corresponding to the input time information,
The number of records including the position information in the records acquired by the record acquisition means is stored in the real-time point information database for each position information included in the input geographic range by the point of interest extraction means . The position information is included in the input geographic range, and the time information is included in records corresponding to the input time information on another date, and compared with the number of records including the position information. Thus, it is determined whether or not it is a point of interest, the position information determined to be a point of interest is extracted,
For each word of the word information included in the record that is the record acquired by the record acquisition unit by the attention factor word extraction unit and that includes the position information that is the attention point extracted by the attention point extraction unit The number of records in which the word is included in the word information in the record, the position information that is the point of interest, and the time information that is input in another date of a predetermined number of days, stored in the real-time point information database The word attention factor based on the number of days of the other date when the record containing the word exists in the word information and the predetermined number of days, which are obtained based on all records including time information corresponding to The degree of attention factor is extracted as the attention factor word,
By the event occurrence point extracting means, the real-time location information stored in the database, and the point of interest at which the position information, the time information in the time zone from the time indicated by the input time information to the time traced back a predetermined time The number of records in which the attention factor word is included in the word information in all the records including the position information of the attention point accumulated in the real-time point information database, and the time zone at another date The number of records in which the attention factor word is included in the word information in all records including the time information , and the position information that is the attention point stored in the real-time point information database, and the date at another date The attention factor word is included in the word information in all records including time information within a time zone. Based on the number of records, the point of interest is by determining whether a point a sudden event has occurred, an event occurrence point extracting method for extracting a point where sudden event has occurred. A program for causing a computer to function as each means of the event occurrence point extraction device according to claim 1 .

Images