JP2012164018A - Tag recommendation device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide technique for recommending a tag to a document.SOLUTION: A tag recommendation device includes: gathering means of gathering a document including a tag comprising a specific symbol and a character string; extraction means of extracting a word and the tag included in each of the gathered documents and a combination of the tag and the word included in the same document from each of the gathered documents; calculation means of calculating a tag-word co-occurrence scale indicative of a degree of co-occurrence of each tag and each word in the same document for each of combinations of words and tags based upon the word and the tag extracted by the extraction means, the combination of the tag and the word included in the same document, and the number of documents; and recommendation means of receiving a document, extracting words included in the received document, and calculating a recommendation score of each of tags with respect to the received documents based upon the word co-occurrence scales associated with all the extracted words.

Description

  The present invention relates to a tag recommendation device that presents tags.

  In recent years, services such as blogs and microblogs have become widespread. Microblogging is a service that allows an unspecified person to write and send a sentence of, for example, about 100 characters, or read a sentence sent by an unspecified person. These services are characterized by the ability to easily send documents written by individuals. Therefore, the amount of information distributed on the network is increasing rapidly.

  These services have a function that allows a poster (sender of a document) to group documents by adding a tag with a specific symbol (for example, # symbol) and a character string to a document to be posted. is there. For example, when searching for “#abcde”, a document group to which “#abcde” is added is extracted. This function has considerations such as making it easier for readers to find a desired document, and forming a loose communication between authors who use the same tag.

  FIG. 1 is a diagram illustrating an example of a tagged document. In the example of FIG. 1, a tag “#weather” is added to a document “The weather is fine today”. The contributor posts a document including this tag. A user of a service such as a microblog can extract a document related to this document by using a tag “#weather”.

JP 2010-282613 A

  In a service such as a blog or a microblog, a tag added to a document can be freely described by a contributor. Therefore, if the contributor does not know an existing appropriate tag, a new tag may be created, and a plurality of different tags having similar contents may occur. Also, there are so many kinds of tags that it is difficult for the poster to grasp all of them. Therefore, there is a problem that information that should be related originally is dispersed depending on tags, and the burden of searching for information of the reader increases.

  An object of the present invention is to provide a technique for recommending a tag to a document.

  In order to solve the above-mentioned problems, the following configuration is adopted in the aspect of the present invention.

One embodiment of the present invention provides:
A collection means for collecting documents including tags with specific symbols and character strings;
Extraction means for extracting the words included in each collected document, the tags, and combinations of tags and words included in the same document from the collected documents;
Tag word co-occurrence scale indicating the degree of co-occurrence between each tag and each word in the same document based on the word extracted by the extraction means, the tag, the combination of the tag and word contained in the same document, and the number of documents Calculating means for each word and tag combination;
A recommendation means for receiving a document, extracting words included in the received document, and calculating a recommendation score for each tag for the received document based on a tag word co-occurrence scale for all the extracted words;
Is a tag recommendation device.

  Note that, as another aspect of the present invention, a method, a program, and a computer-readable recording medium recording the program may be used to realize any one of the above configurations.

  According to the aspect of the present invention, it is possible to provide a technique for recommending a tag to a document.

FIG. 1 is a diagram illustrating an example of a tagged document. FIG. 2 is a diagram illustrating an example of an information processing system. FIG. 3 is a diagram illustrating an example of a word appearance frequency table. FIG. 4 is a diagram illustrating an example of a tag appearance frequency table. FIG. 5 is a diagram illustrating an example of a tag word co-occurrence frequency table. FIG. 6 is a diagram illustrating an example of a tag usage history table. FIG. 7 is a diagram illustrating an example of a tag usage scale table. FIG. 8 is a diagram illustrating an example of a tag word co-occurrence scale table. FIG. 9 is a diagram illustrating a hardware configuration example of the information processing apparatus. FIG. 10 is a diagram illustrating an example (1) of the operation sequence of the information processing system. FIG. 11 is a diagram illustrating an example (2) of the operation sequence of the information processing system. FIG. 12 is a diagram illustrating an example of an operation flow of the collection unit. FIG. 13 is a diagram illustrating an example of an operation flow for calculating a tag usage scale by the calculation unit. FIG. 14 is a diagram illustrating an example of an operation flow for calculating a tag word co-occurrence scale by the calculation unit. FIG. 15 is a diagram illustrating an example (1) of the operation flow of the recommendation unit. FIG. 16 is a diagram illustrating an example (2) of the operation flow of the recommendation unit.

  Hereinafter, embodiments will be described with reference to the drawings. The configuration of the embodiment is an exemplification, and the present invention is not limited to the configuration of the disclosed embodiment.

Embodiment
(Configuration example)
FIG. 2 is a diagram illustrating an example of an information processing system according to the present embodiment. The information processing system 10 in FIG. 2 includes a server device 100, a storage device 200, and a user terminal 300. The server device 100 is connected to the storage device 200 and the user terminal 300 via a network or the like. A plurality of user terminals 300 can be connected to the server device 100. The server device 100 may include a storage device 200. The network or the like may be a public network such as the Internet, or an internal network such as a LAN (Local Area Network) or a WAN (Wide Area Network).

The server device 100 includes a service unit 110, a collection unit 120, a calculation unit 130, and a recommendation unit 140. Any one of the service unit 110, the collection unit 120, the calculation unit 130, and the recommendation unit 140 may be included in another server device. For example, a server device that includes a service unit 110, a server device that includes a collection unit 120, a server device that includes a calculation unit 130, and a server device that includes a recommendation unit 140 are connected via a network or the like. 100 may operate. The service unit 110, the collection unit 120, the calculation unit 130, and the recommendation unit 140 are realized by a plurality of server devices, so that the load of each processing unit is distributed.

  The service unit 110 provides services such as microblogging to the user terminal 300 and the like. The service unit 110 stores a document posted from the user terminal 300 or the like, a date and time when the document is posted, in a service such as a microblog. The service unit 110 provides the document or the like in response to a request from the collection unit 120. The provided document includes information on the date and time when the document was posted.

  The collection unit 120 requests a document posted to the service unit 110 and receives a document (document group) from the service unit 110. The collection unit 120 extracts a tagged document from the document group provided from the service unit 110. A tagged document is a document including a tag with a specific symbol (for example, # symbol) and a character string. The collection unit 120 requests a document posted to the service unit 110 every predetermined time.

  The collection unit 120 performs morphological analysis on all the extracted documents. The collection unit 120 acquires word information included in each document as a morphological analysis execution result. The collection unit 120 counts the number of appearances of each word (word appearance frequency) from the word information. In addition, the collection unit 120 acquires tag information included in each document. The collection unit 120 counts the number of appearances of each tag (tag appearance frequency) from the tag information. Furthermore, the collection unit 120 counts the number of documents in which a combination of an arbitrary tag and an arbitrary word appears in the same document (tag word co-occurrence frequency) for each combination of the tag and the word. The collection unit 120 uses the word appearance frequency data, the tag appearance frequency data, and the tag word co-occurrence frequency data as a word appearance frequency table 211, a tag appearance frequency table 212, and a tag word co-occurrence frequency table 213, respectively. Store in DB210.

  The collection unit 120 stores tag usage history data, which is a combination of a tag and information on the date and time when a document including the tag is posted, in the frequency DB 210 as a tag usage history table 214.

  The calculation unit 130 acquires tag usage history data from the tag usage history table 214 stored in the frequency DB 210 and calculates a tag usage scale for each tag. The tag usage scale is a scale that represents a change in tag usage frequency. The calculation of the tag usage scale will be described later. The calculation unit 130 stores the tag usage scale data in which the tag is associated with the calculated tag usage scale as the tag usage scale table 221 in the tag usage scale DB 220. The calculation unit 130 may acquire tag usage history data from the collection unit 120.

  Further, the calculation unit 130 acquires word appearance frequency data from the word appearance frequency table 211 stored in the frequency DB 210. The calculation unit 130 acquires tag appearance frequency data from the tag appearance frequency table 212 stored in the frequency DB 210. The calculation unit 130 acquires tag word co-occurrence frequency data from the tag word co-occurrence frequency table 213 stored in the frequency DB 210. The calculation unit 130 calculates a tag word co-occurrence scale for each tag-word combination based on the acquired data. The tag word co-occurrence scale is a scale representing the degree of tag-word co-occurrence. The calculation of the tag word co-occurrence scale will be described later. The calculation unit 130 stores the tag word co-occurrence scale data in which the tag-word combination and the calculated tag word co-occurrence scale are associated with each other in the tag word co-occurrence scale DB 230 as the tag word co-occurrence scale table 231. .

The recommendation unit 140 receives a document scheduled to be posted to a service such as a microblog from the user terminal 300. The recommendation unit 140 performs morphological analysis on the received document, and acquires word information included in the document. The recommendation unit 140 extracts tag word co-occurrence scale data including each word from the tag word co-occurrence scale table 231 stored in the tag word co-occurrence scale DB 230 based on the acquired word information. For the tags included in the tag word co-occurrence scale data, the recommendation unit 140 extracts tag use scale data including the tags from the tag use scale table 231 of the tag use scale DB 220. For each tag, the recommendation unit 140 calculates a tag recommendation score from the tag word co-occurrence scale and the tag usage scale. The tag recommendation score indicates the degree to which the tag is recommended to be added to a document to be posted. The recommendation unit 140 transmits the calculated tag recommendation score top N (N is a predetermined value) tag and its recommendation score to the user terminal 300.

The storage device 200 includes a frequency DB 210 (Data Base: database) and a tag usage scale DB.
220, tag term co-occurrence scale DB230 is included. The frequency DB 210, the tag usage scale DB 220, and the tag word co-occurrence scale DB 230 may each be included in separate storage devices.

  The frequency DB 210 includes a word appearance frequency table 211, a tag appearance frequency table 212, a tag word co-occurrence frequency table 213, and a tag usage history table 214.

  FIG. 3 is a diagram illustrating an example of a word appearance frequency table. The word appearance frequency table 211 stores word appearance frequency data in which a word that appears in a document is associated with the number of appearances of the word. A combination of one piece of information and one piece of information (for example, a word and the number of appearances of this word) in the table is also referred to as one record.

  FIG. 4 is a diagram illustrating an example of a tag appearance frequency table. The tag appearance frequency table 212 stores tag appearance frequency data in which tags appearing in a document are associated with the number of appearances of the tags.

  FIG. 5 is a diagram illustrating an example of a tag word co-occurrence frequency table. The tag word co-occurrence frequency table 213 stores tag word co-occurrence frequency data in which combinations of tags and words that appear in the same document are associated with the number of appearances of this combination.

  FIG. 6 is a diagram illustrating an example of a tag usage history table. The tag usage history table 214 stores tag usage history data that is a combination of a tag and information on the date and time when a document including the tag is posted.

  The tag usage scale DB 220 includes a tag usage scale table 221.

  FIG. 7 is a diagram illustrating an example of a tag usage scale table. The tag usage scale table 221 stores tag usage scale data in which tags are associated with tag usage scales calculated by the calculation unit 130.

  The tag word co-occurrence scale DB 230 includes a tag word co-occurrence scale table 231.

  FIG. 8 is a diagram illustrating an example of a tag word co-occurrence scale table. The tag word co-occurrence scale table 231 stores tag word co-occurrence scale data in which a combination of a tag and a word is associated with a tag word co-occurrence scale calculated by the calculation unit 130.

The user terminal 300 transmits a document to be posted to a microblog or the like input by the user to the recommendation unit 140. The user terminal 300 receives a tag recommended for the document transmitted to the recommendation unit 140 and its recommendation score from the recommendation unit 140. The user terminal 300 presents the tag received from the recommendation unit 140 and its recommendation score to the user, and causes the user to select a tag to be added to the document to be posted. When a tag to be added to the document is selected by the user, the user terminal 300 transmits (posts) the document with the tag added to the service unit 110.

The server apparatus 100 can be realized by using a general-purpose computer such as a personal computer (PC) or a dedicated computer such as a server machine.

  The user terminal 300 can be realized by using a dedicated or general-purpose computer such as a PC or PDA (Personal Digital Assistant), or an electronic device equipped with the computer. The user terminal 300 can be realized by using a dedicated or general-purpose computer such as a smartphone, a mobile phone, or a car navigation device, or an electronic device equipped with a computer.

  FIG. 9 is a diagram illustrating a hardware configuration example of the information processing apparatus. The server apparatus 100 and the user terminal 300 are realized by, for example, an information processing apparatus 1000 as illustrated in FIG.

The computer, that is, the information processing apparatus 1000 is a CPU (Central Processing Unit).
1002, a memory 1004, a storage unit 1006, an input unit 1008, an output unit 1010, and a communication unit 1012.

  In the information processing apparatus 1000, the CPU 1002 loads a program stored in the recording unit 1006 into the work area of the memory 1004 and executes the program, and the peripheral device is controlled through the execution of the program. Can be realized.

  The CPU 1002 performs processing according to a program stored in the storage unit 1006.

  The memory 1004 is used by the CPU 1002 to cache programs and data and to develop work areas. The memory 1004 includes, for example, a RAM (Random Access Memory) and a ROM (Read Only Memory).

The storage unit 1006 stores various programs and various data in a recording medium in a readable and writable manner. The storage unit 1006 is, for example, an EPROM (Erasable Programmable ROM), a solid state drive device, or a hard disk drive (HDD, Hard Disk Drive) device. As the storage unit 1006, for example, a CD (Compact Disc) drive device, a DVD (Digital Versatile Disk) drive device, a + R / + RW drive device, an HD DVD (High-Definition Digital Versatile Disk) drive device, or a BD (Blu- ray Disk) drive device. Examples of the recording medium include a silicon disk including a nonvolatile semiconductor memory (flash memory), a hard disk, a CD, a DVD, + R / + RW, an HD DVD, or a BD. CDs include CD-R (Recordable), CD-RW (Rewritable), and CD-ROM. Examples of DVD include DVD-R and DVD-RAM (Random Access Memory). BD includes BD-R, BD-RE (Rewritable), and BD-ROM. The storage unit 1006 can include a removable medium, that is, a portable recording medium. The removable medium is, for example, a USB (Universal Serial Bus) memory or a disk recording medium such as a CD or a DVD.

  The memory 1004 and the storage unit 1006 are computer-readable recording media.

The input unit 1008 receives an operation instruction or the like from a user or the like. The input unit 1008 is an input device such as a keyboard, a pointing device, a wireless remote controller, a microphone, or a camera. Information input from the input unit 1008 is notified to the CPU 1002.

  The output unit 1010 outputs data processed by the CPU 1002 and data stored in the memory 1004. The output unit 1010 is an output device such as a CRT (Cathode Ray Tube) display, an LCD (Liquid Crystal Display), a PDP (Plasma Display Panel), an EL (Electroluminescence) panel, a printer, or a speaker.

  The communication unit 1012 transmits / receives data to / from an external device. The communication unit 1012 is connected to an external device via, for example, a signal line. The external device is, for example, another information processing device or storage device. The communication unit 1012 is, for example, a LAN (Local Area Network) interface board or a wireless communication circuit for wireless communication.

  The information processing apparatus 1000 stores an operating system, various programs, various tables, and the like in the storage unit 1006.

  The operating system is software that mediates software and hardware, manages memory space, manages files, manages processes and tasks, and the like. The operating system includes a communication interface. The communication interface is a program for exchanging data with other external devices connected via the communication unit 1012.

  In the information processing apparatus 1000 that can implement the server apparatus 100, the CPU 1002 loads the program stored in the storage unit 1006 into the memory 1004 and executes the program, whereby the service unit 110, the collection unit 120, the calculation unit 130, and the recommendation unit 140. As a function.

Examples of the storage device 200 include a solid state drive device, a hard disk drive device, a CD drive device, a DVD drive device, a + R / + RW drive device, and an HD.
There is a DVD drive device or a BD drive device. The storage device 200 can include a removable medium, that is, a portable recording medium.

(Operation example)
<The entire>
An operation example of the information processing system 10 according to the present embodiment will be described.

  10 and 11 are diagrams illustrating an example of an operation sequence of the information processing system. “A”, “B”, “C”, “D”, “E”, “F”, “G”, and “H” in FIG. 10 respectively represent “A”, “B”, “ Connect with “C”, “D”, “E”, “F”, “G”, “H”.

  The collection unit 120 requests a document posted to the service unit 110 at a predetermined cycle, and collects documents (document groups) accumulated in the service unit 110 from the service unit 110 (SQ1002). The collection unit 120 may collect accumulated documents from other server devices.

  The collection unit 120 extracts a tagged document from the document group provided from the service unit 110. The collection unit 120 extracts words and tags included in each document. The collection unit 120 generates word appearance frequency data, tag appearance frequency data, tag word co-occurrence frequency data, and tag usage history data from the extracted words, tags, and the like (SQ1004).

The collection unit 120 generates the generated word appearance frequency data, tag appearance frequency data, tag word co-occurrence frequency data, and tag usage history data in the frequency DB 210 in a word appearance frequency table 211, a tag appearance frequency table 212, and a tag, respectively. The word co-occurrence frequency table 213 and the tag usage history table 214 are stored (SQ1006).

  The calculation unit 130 acquires tag usage history data from the tag usage history table 214 stored in the frequency DB 210 (SQ1008). The calculation unit 130 may acquire tag usage history data from the collection unit 120. The calculation unit 130 calculates a tag usage scale for each tag from the tag usage history data (SQ1010). The calculation unit 130 stores the tag usage scale data in which the tag is associated with the calculated tag usage scale in the tag usage scale DB 220 as the tag usage scale table 221 (SQ1012).

  The calculation unit 130 acquires word appearance frequency data, tag appearance frequency data, and tag word co-occurrence frequency data from the frequency DB 210 (SQ1014). The calculation unit 130 calculates a tag word co-occurrence scale for each combination of a tag and a word from these acquired data (SQ1016). The calculation unit 130 stores the tag word co-occurrence scale data in which the tag-word combination and the calculated tag word co-occurrence scale are associated with each other in the tag word co-occurrence scale DB 230 as the tag word co-occurrence scale table 231. (SQ1018).

  Through the operations so far, data for recommending a tag to the user terminal 300 is generated.

  The user terminal 300 transmits a document to be posted to a microblog or the like input by the user to the recommendation unit 140 (SQ1020). When the recommendation unit 140 receives a document to be posted to a service such as a microblog from the user terminal 300, the recommendation unit 140 performs morphological analysis on the received document and acquires word information included in the document. The recommendation unit 140 extracts tag word co-occurrence scale data including each word based on the acquired word information from the tag word co-occurrence scale table 231 stored in the tag word co-occurrence scale DB 230 (SQ1022). The recommendation unit 140 extracts the tag usage scale data including the tag from the tag usage scale table 231 of the tag usage scale DB 220 for the tags included in the extracted tag word co-occurrence scale data (SQ1024). For each tag, the recommendation unit 140 calculates a tag recommendation score from the tag word co-occurrence scale and the tag usage scale (SQ1026). The recommendation unit 140 transmits the calculated tag recommendation score top N (N is a predetermined number) tags and their recommendation scores to the user terminal 300 (SQ1028).

  The user terminal 300 receives a tag recommended for the document transmitted to the recommendation unit 140 and its recommendation score from the recommendation unit 140. The user terminal 300 presents the tag received from the recommendation unit 140 and its recommendation score to the user, and selects a tag to be added to the document to be posted (SQ1030). When a tag to be added to the document is selected by the user, the user terminal 300 transmits (posts) the document with the tag added to the service unit 110 (SQ 1032). Thereby, the user can add an appropriate tag to the document to be posted. Posted documents are accumulated in the service unit 110 and collected by the collection unit 120.

<Collection Department>
FIG. 12 is a diagram illustrating an example of an operation flow of the collection unit. The operation flow in FIG. 12 operates, for example, every predetermined time.

  The collection unit 120 requests a document posted to the service unit 110 and receives a document (document group) from the service unit 110 (S101). The collection unit 120 may request a posted document from another server device and collect the document (document group). The collected document is, for example, a document posted by a blog service or a microblog service. The collected document includes information on the date and time when the document was posted.

  The collection unit 120 extracts a tagged document from the collected document group (S102). A tagged document is a document including a tag with a specific symbol (for example, # symbol) and a character string. The collection unit 120 may collect only tagged documents from the service unit 110.

  The collection unit 120 performs morphological analysis on all the extracted documents. The collection unit 120 acquires word information included in each document and tag information included in each document as an execution result of the morphological analysis. The collection unit 120 counts the number of appearances of each word (word appearance frequency) from the word information. The number of appearances may be the number of appearances in document units. The number of appearances in units of documents means that the number of appearances of a word is counted as one even if the same word is included in one document. The collection unit 120 counts the number of appearances of each tag (tag appearance frequency) from the tag information. Furthermore, the collection unit 120 counts the number of documents in which a combination of an arbitrary tag and an arbitrary word appears in the same document (tag word co-occurrence frequency) for each combination of the tag and the word. The collection unit 120 uses the word appearance frequency data, the tag appearance frequency data, and the tag word co-occurrence frequency data as a word appearance frequency table 211, a tag appearance frequency table 212, and a tag word co-occurrence frequency table 213, respectively. Store in DB210. The collection unit 120 stores, in the frequency DB 210, tag usage history data that is a combination of a tag and information on the date and time when a document including the tag is posted (S103).

<Calculation unit>
[Tag Usage Scale]
FIG. 13 is a diagram illustrating an example of an operation flow for calculating a tag usage scale by the calculation unit. The operation flow of FIG. 13 operates, for example, every predetermined time.

  The calculation unit 130 acquires tag usage history data from the tag usage history table 214 stored in the frequency DB 210 (S201) for the latest (for example, from A day before to the present (A is a predetermined value)). The calculation unit 130 calculates the number of appearances of the tag for each predetermined time interval for each tag (S202). The calculation unit 130 calculates a tag usage scale for each tag from the number of appearances of the tag for each predetermined time interval for each tag (S203).

  Next, a specific example of calculating the tag usage scale will be shown.

<< Example of tag usage scale calculation (1) >>
The representative time of the predetermined time interval is time Xi, and the appearance frequency (number of times) of the tag in the predetermined time interval including time Xi is Yi. At this time, assuming that Yi can be approximated by a linear expression of Xi, the slope a of the linear expression is obtained as follows by the least square method.

  Here, n is the number of predetermined time intervals. That is, assuming that data from A day before to the present is acquired, n is a value obtained by dividing A day by a predetermined time interval.

  Using this inclination a, the tag utilization scale k can be obtained as follows.

That is, when the slope a is positive, the tag usage scale k is 1, and when the slope a is negative, the tag usage scale k is a value corresponding to the slope a. Therefore, when the tag usage increases with time, the tag usage scale k is 1 which is the maximum value. When the tag usage decreases with time, the tag usage scale k is cos (tan ⁻¹ (a)).

<< Example of tag usage scale calculation (2) >>
The representative time of the predetermined time interval is time Xi, and the appearance frequency of the tag in the predetermined time interval including time Xi is Yi. The current time is set as time p. At this time, the tag utilization scale k can be obtained as follows.

  The sum of the appearance frequency Yi at time Xi divided by the difference between the current date and time and time Xi is taken. The smaller the difference between the current date and time and the time Xi, the larger the value. In addition, since the value increases as the usage frequency increases, the tag usage scale k increases for tags that are frequently used.

<< Example of tag usage scale calculation (3) >>
Here, the tag usage scale k is obtained without using the number of appearances of the tag for each predetermined time interval. The tag usage scale is determined as follows.

  Here, f (tag) is the number of appearances of the tag “tag” in the latest (for example, from day A to the present). The value N is the number of documents with the latest tag collected by the collection unit 120. This tag usage scale k corresponds to the most recent appearance ratio of the tag “tag”.

[Tag word co-occurrence scale]
FIG. 14 is a diagram illustrating an example of an operation flow for calculating a tag word co-occurrence scale by the calculation unit. The operation flow in FIG. 14 operates, for example, every predetermined time. The tag word co-occurrence scale is a scale representing the degree of co-occurrence between a tag and a word.

  The calculation unit 130 acquires tag word co-occurrence frequency data one by one from the tag word co-occurrence frequency table 213 stored in the frequency DB 210 (S301). The calculation unit 130 acquires tag appearance frequency data for the tag acquired in step S301 from the tag appearance frequency table 212 stored in the frequency DB 210. Further, the calculation unit 130 acquires word appearance frequency data for the word acquired in step S301 from the word appearance frequency table 211 stored in the frequency DB 210. The calculation unit 130 calculates a tag word co-occurrence scale for a combination of a tag and a word based on the acquired data. The calculation unit 130 stores the tag word co-occurrence scale data in which the combination of the tag and the word is associated with the calculated tag word co-occurrence scale in the tag word co-occurrence scale DB 230 as the tag word co-occurrence scale table 231. (S303). The calculation unit 130 checks whether or not all the tag word co-occurrence frequency data stored in the frequency DB 210 has been acquired (S304). If there is tag word co-occurrence frequency data that has not yet been acquired (S304; NO), the calculation unit 130 returns the process to step S301. Moreover, the calculation part 130 complete | finishes a process, when all the tag word co-occurrence frequency data are acquired (S304; YES).

  Here, a specific example of calculating the tag word co-occurrence scale will be described. The tag word co-occurrence scale may be normalized to be 0 or more and 1 or less.

<< Example of tag word co-occurrence scale calculation (1) >>
The co-occurrence frequency f (term, tag) can be used as the tag word co-occurrence scale m. Here, the co-occurrence frequency f (term, tag) indicates the number of documents in which the word “term” and the tag “tag” appear in the same document. f (term, tag) is the co-occurrence observation.

<< Example of tag word co-occurrence scale calculation (2) >>
The ratio between the observed value and the expected value can be the tag word co-occurrence scale m. That is, it can be expressed as follows. The ratio between the observed value and the expected value means that the larger the value, the easier it is to co-occur.

Here, f (term) indicates the number of times that the word “term” appears (the number of documents). f (tag) indicates the number of times the tag “tag” appears (number of documents). The value N is the number of documents with the latest tag collected by the collection unit 120.

<< Example of tag word co-occurrence scale calculation (3) >>
By applying the t-test independence test, the tag word co-occurrence scale m can be obtained as follows.

<< Example of tag word co-occurrence scale calculation (4) >>
The tag word co-occurrence scale m can be obtained as follows, assuming that the co-occurrence of words and tags occurs randomly.

<< Example of calculating tag word co-occurrence scale (5) >>
Using log-likelihood ratio (LLR), tag word co-occurrence scale m
Can be requested.

  here,

It is. In principle, the base of the logarithm is e.

<< Example of tag word co-occurrence scale calculation (6) >>
Tag word co-occurrence scale m using PMI (Point-wise Mutual Information)
Can be requested.

  The tag word co-occurrence scale m has a very large value when the probability that the tag T is attached to the document in which the word A appears is high and the probability that the word A appears in the document to which the tag T is attached is high.

<Recommendation Department>
15 and 16 are diagrams illustrating an example of an operation flow of the recommendation unit. “A” and “B” in FIG. 15 are connected to “A” and “B” in FIG. 16, respectively. The operation flows of FIGS. 15 and 16 are started by receiving a document from the user terminal 300, for example.

  The recommendation unit 140 receives a document scheduled to be posted to a service such as a microblog from the user terminal 300 (S401). The recommendation unit 140 performs morphological analysis on the received document, divides the document into words, and acquires word information included in the document (S402). The recommendation unit 140 may acquire word information included in the document by a method other than morphological analysis. The recommendation unit 140 determines whether or not the number of words included in the received document is greater than or equal to a threshold value Wth (S403).

  When the number of words included in the received document is equal to or greater than the threshold value Wth (S403; YES), the recommendation unit 140 extracts tag word co-occurrence scale data regarding each word included in the document from the tag word co-occurrence scale DB 230. (S404). The tag word co-occurrence scale data related to a word is tag word co-occurrence scale data (record) including the word. A plurality of tag word co-occurrence scale data may be extracted for one word. The recommendation unit 140 collects the extracted tag word co-occurrence scale data for each tag. When a plurality of tag word co-occurrence scale data is extracted for one tag, the recommendation unit 140 integrates the tag word co-occurrence scale data of the tag word co-occurrence scale data of the same tag, and the basic recommendation scale of this tag And Here, integration refers to, for example, multiplying each tag word co-occurrence scale. Instead of multiplication, the sum of each tag word co-occurrence scale may be taken. Integration is not limited to multiplication or summation. When one tag word co-occurrence scale data is extracted for each tag, the recommendation unit 140 sets the tag word co-occurrence scale of the tag word co-occurrence scale data as the basic recommended scale of this tag. In this way, the recommendation unit 140 calculates a basic recommendation scale for each tag (S405).

For the tags included in the extracted tag word co-occurrence scale data, the recommendation unit 140 extracts tag use scale data including the tags from the tag use scale table 231 of the tag use scale DB 220. The recommendation unit 140 integrates the tag usage scale of the tag into the basic recommendation scale of each tag to obtain a recommendation score (S406). A tag with a higher recommendation score means that the tag is more suitable to be added to the received document. Here, integration is, for example, multiplying the basic recommendation scale and the tag usage scale. Moreover, you may add a basic recommendation scale and a tag utilization scale instead of multiplication. Further, instead of multiplication, the basic recommendation scale may be multiplied by a predetermined coefficient and added to the tag usage scale. The recommendation unit 140 sorts the tags in descending order of the recommendation score obtained in step S406. The recommendation unit 140 extracts the top N sorted tags, transmits the tag and the recommendation score of the tag to the user terminal 300, and ends the process (S407). Instead of using the tag utilization scale, the basic recommendation scale may be used as the recommendation score as it is.

  When the number of words included in the received document is less than the threshold Wth (S403; NO), the recommendation unit 140 cannot obtain sufficient information to recommend an appropriate tag to the user terminal 300. An error is transmitted (S408), and the process is terminated.

(Effect of embodiment)
The collection unit 120 of the server device 100 collects documents posted to a service such as a microblog from the service unit 110, the date and time when the document was posted, and the like. The collection unit 120 extracts word information and tag information from the collected documents and the like. The collection unit 120 obtains the number of times each word appears, the number of times each tag appears, and the tag word co-occurrence frequency based on the word information. Further, the collection unit 120 generates a tag usage history from the tag information and the date and time when the document is posted. The calculation unit 130 obtains a tag word co-occurrence scale from the number of times each word appears, the number of times each tag appears, and the tag word co-occurrence frequency. Further, the calculation unit 130 obtains a tag usage scale from the tag usage history. The recommendation unit 140 receives a document to be posted from the user terminal 300 and extracts words included in the document. The recommendation unit 140 extracts a tag to be recommended as a tag to be added to a document to be posted based on a word, a tag word co-occurrence scale, and a tag usage scale included in the document to be posted to a service such as a microblog. . The recommendation unit 140 extracts a recommended tag by calculating a tag recommendation score based on a word, a tag word co-occurrence scale, and a tag usage scale included in a document to be posted. The recommendation unit 140 transmits a recommended tag to the user terminal 300 as a tag to be added to the document scheduled to be posted. The server apparatus 100 can extract a tag determined to be appropriate as a tag to be added to a document scheduled to be posted based on a document posted in the past. Since the user of the user terminal 300 is presented with a tag to be added, an appropriate tag can be selected without knowing the tag exhaustively.

  Further, the server device 100 can easily extract tags that are actively used as recommended tags by using the tag usage scale. Also, tags that have been used a lot in the past but have not been used recently are less likely to be extracted as recommended tags. A tag recommended and used by the tag server device 100 is added to a document posted in a microblog service or the like, so that the server device 100 can use the document to which the tag is added as a usage scale or a co-occurrence scale. By providing feedback, it is possible to recommend tags with higher quality. Even when a plurality of similar tags are used, this feedback structure makes it easy to unify tags.

  According to the server apparatus 100, by transmitting a tag recommended as a tag to be added to a document to be posted to the user terminal 300, it is possible to easily select an appropriate tag to be added to the document posted by the user. it can.

(Modification)
In the above-described example, the tag word co-occurrence scale is obtained from the co-occurrence frequency between the tag and the word, and the tag recommendation score is calculated. In addition to this, additional information (context,
A co-occurrence scale (contents) and a tag (tag additional information co-occurrence scale) may be obtained and used to calculate a tag recommendation score. As additional information (context, context), for example, weather (temperature, atmospheric pressure, humidity, wind speed, precipitation, weather, etc.), time zone (morning, noon, night, every hour, etc.), location (latitude, longitude, facility) , Roads, routes, etc.), types of user terminals, and the like.

  When the user terminal 300 posts a document including a tag to a service such as a microblog, the user terminal 300 transmits the document together with additional information. The user terminal 300 acquires additional information using a function unique to the user terminal. The user terminal 300 may acquire the additional information by causing the user to input the additional information. When the service unit 110 receives additional information together with the document from the user terminal 300, the service unit 110 accumulates the additional information together with the posted document, the date and time when the document was posted, and the like. The collection unit 120 collects additional information from the service unit 110 together with the posted document, the date and time when the document was posted, and the like. The collection unit 120 generates additional information frequency data in the same manner as the word frequency data and the tag frequency data. Further, the collection unit 120 obtains the co-occurrence frequency of additional information (context, contents) and a tag regarding the same document. The calculation unit 130 obtains the tag additional information co-occurrence scale in the same manner as the tag word co-occurrence scale.

  The user terminal 300 transmits additional information to the recommendation unit 140 together with the document scheduled to be posted. The recommendation unit 140 extracts the tag additional information scale data regarding the additional information in the same manner as extracting the tag word co-occurrence scale data regarding each word included in the document. The recommendation unit 140 collects the extracted tag word co-occurrence scale data and tag additional information scale data for each tag. The recommendation unit 140 integrates the tag word co-occurrence scale of the tag word co-occurrence scale data and the tag additional information scale of the tag additional information scale data, and sets it as a basic recommendation scale for the tag. Here, integration refers to, for example, multiplying each tag word co-occurrence scale and each tag additional information scale. Instead of multiplication, the sum of each tag word co-occurrence measure and each tag additional information measure may be taken. When taking the sum, each tag word co-occurrence scale and each tag additional information scale may be given a predetermined weight. Integration is not limited to these.

  According to the server apparatus 100, it is possible to recommend a tag to be added to a document in consideration of additional information. When calculating the recommendation score, the server device 100 can recommend a more appropriate tag by using the additional information in addition to the word information.

[Computer-readable recording medium]
A program for causing a computer or other machine or device (hereinafter, a computer or the like) to realize any of the above functions can be recorded on a recording medium that can be read by the computer or the like. The function can be provided by causing a computer or the like to read and execute the program of the recording medium.

  Here, a computer-readable recording medium is a recording medium that stores information such as data and programs by electrical, magnetic, optical, mechanical, or chemical action and can be read from a computer or the like. Say. In such a medium, elements constituting a computer such as a CPU and a memory may be provided to cause the CPU to execute a program.

  Examples of such a recording medium that can be removed from a computer or the like include a flexible disk, a magneto-optical disk, a CD-ROM, a CD-R / W, a DVD, a DAT, an 8 mm tape, and a memory card. .

  Moreover, there are a hard disk drive, a ROM, and the like as a recording medium fixed to a computer or the like.

10 Information processing system
100 server device
110 Service Department
120 Collection Department
130 Calculation unit
140 recommendation section
200 storage device
210 Frequency DB
211 Word appearance frequency table
212 Tag appearance frequency table
213 Tag word co-occurrence frequency table
214 Tag Usage History Table
220 Tag Usage Scale DB
221 Tag usage scale table
230 Tag Term Co-occurrence Scale DB
231 Tag word co-occurrence scale table
300 User terminal
1000 Information processing device
1002 CPU
1004 memory
1006 Storage unit
1008 Input section
1010 Output unit
1012 Communication Department

Claims (3)

A collection means for collecting documents including tags with specific symbols and character strings;
Extraction means for extracting the words included in each collected document, the tags, and combinations of tags and words included in the same document from the collected documents;
Tag word co-occurrence scale indicating the degree of co-occurrence between each tag and each word in the same document based on the word extracted by the extraction means, the tag, the combination of the tag and word contained in the same document, and the number of documents Calculating means for each word and tag combination;
A recommendation means for receiving a document, extracting words included in the received document, and calculating a recommendation score for each tag for the received document based on a tag word co-occurrence scale for all the extracted words;
A tag recommendation device comprising: The collecting means collects a document including a tag with a specific symbol and a character string and a date and time when the document is posted,
The calculation means calculates a tag usage scale indicating a usage level of the tag for each tag based on the date and time when the document including the tag and the tag is posted;
The recommendation means calculates a recommendation score for each tag for the received document based on the tag word co-occurrence scale and the tag usage scale.
The tag recommendation device according to claim 1. The collecting means collects a document including a tag with a specific symbol and a character string and additional information related to the document,
The extraction means extracts a tag included in each collected document, the additional information, and a combination of a tag and additional information related to the same document from each collected document and additional information,
The calculation means calculates the tag included in each collected document, the additional information, the combination of the tag and additional information regarding the same document, and the number of documents, and the tag and each additional information regarding the same document. A tag additional information co-occurrence scale indicating the degree of co-occurrence is calculated for each combination of word and additional information,
The recommendation means receives a document and additional information, and calculates a recommendation score for each tag for the received document based on a tag additional information co-occurrence scale related to the received additional information.
The tag recommendation device according to claim 1 or 2.

Images