JP2013134738A - Apparatus, program and method for tagging keyword to position information on the basis of multiple comment sentences - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a device and the like capable of tagging real-time keywords to an event position in which a special trend different from the ordinary can be observed, on the basis of multiple comment sentences.SOLUTION: A tagging apparatus includes: comment collection means for collecting comment sentences including a posting time and position information from a blog server; event position detecting means for detecting an event position in which predetermined number or more comment sentences are concentrated in a predetermined position range, using the multiple comment sentences posted in a predetermined time zone; comment keyword extraction means for extracting words from a group of the comment sentences for every event position, counting an appearance frequency for every word, and extracting a comment keyword the appearance frequency of which becomes a prescribed number or more; and tagging means for associating the comment keyword as a tag with every event position.

Description

  The present invention relates to a technique for tagging position information with a keyword.

  In recent years, comment texts from an unspecified number of third parties have become active over sites such as blogs (Web log) and mini blogs (for example, twitter (registered trademark)) via the Internet. Has been sent to. These comment sentences are characterized not only by sending them but also by discussing common topics. This gives the feeling that the topic is shared with many users. For example, according to twitter, a contributor sends a comment sentence including a unique hash tag, and a viewer can browse the entire comment sentence including the hash tag. For example, when a local event such as a festival is held in a certain area, the comment text can be transmitted using a hashtag linked to the festival. On the other hand, the viewer can browse the entire tweets about the festival by browsing using this hashtag.

  On the other hand, for example, there is a technology that can discover a local event using tweets with geotags in twitter (see, for example, Non-Patent Document 1). The geo-tagged tweets means that current position information is automatically added to a comment sentence (tweet) using a GPS (Global Positioning System) function. According to this technology, it is possible to detect a region where a special trend different from usual is observed for a local event such as a festival where many people are active.

Tatsuya Fujisaka, Ryu Li, Kazutoshi Kakutani, “Discovering regional events based on regional normal conditions using collective experience knowledge of Twitter users”, WebDBForum 2010, [online], [December 27, 2011 search], Internet <URL: http://www.shse.u-hyogo.ac.jp/sumiya/japanese/kenkyuu_jyusyou.html>

  However, according to the technique described in Non-Patent Document 1, although the event position can be detected, the contents of the event cannot be detected. That is, the event name, category (sports, fireworks, accidents, etc.), and detailed contents (event related Web information, etc.) cannot be detected. In particular, in the case of tweets with geotags, only the position information is added to the comment text, and the content of the event is not specified. As a result, comment sentences based on events need to be collected by hashtags. However, comment text collected by hashtags can be posted from the terminal of a user who does not stay in the area, and the nature of the comment text is different from the comment text posted from the terminal of a user who actually stays in that area. Come.

  Therefore, an object of the present invention is to provide an apparatus, a program, and a method for tagging a keyword in real time at an event position where a special trend different from usual is seen based on a large number of comment sentences.

According to the present invention, a tagging device that accumulates comment text including post time and position information posted from an unspecified number of third parties,
Event position detecting means for detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
Comment keyword extracting means for extracting a word from a group of comment sentences for each event position, counting the appearance frequency for each word, and extracting a comment keyword whose appearance frequency is a predetermined number or more;
There is provided tagging means for associating a comment keyword as a tag for each event position.

  According to another embodiment of the tagging device of the present invention, the event position detecting means is configured such that the number of users who have posted these comment sentences is equal to or less than a predetermined ratio value with respect to the total number of comment sentences for each event position. In this case, it is also preferable to narrow down by excluding the event position.

  According to another embodiment of the tagging apparatus of the present invention, the event position detecting means extracts only the comment text including the position information from the many comment text posted in the predetermined time zone, and the posting It is also preferable to narrow down by excluding comment sentences in which a sentence claimed by the person himself is not described.

  According to another embodiment of the tagging device of the present invention, the event position detecting means excludes the event position when there is no word having the appearance frequency equal to or higher than a predetermined threshold value for each event position. It is also preferable to narrow down.

  According to another embodiment of the tagging device of the present invention, the event position detecting means detects an event position using a DBSCAN (Density-Based Spatial Clustering) algorithm with position information included in the comment text as an element. It is also preferable.

  According to another embodiment of the tagging apparatus of the present invention, the map information indicating the event position stored by the tagging means and the comment keyword to be described in the event position portion on the map are operated by the user operation. It is also preferable to further have a map information transmitting means for transmitting to a possible terminal.

  According to another embodiment of the tagging apparatus of the present invention, the comment keyword extracting means extracts a word from a plurality of comment sentences by morphological analysis, and also uses a TF-IDF (Term Frequency-Inverse Document Frequency). It is also preferable to extract a characteristic word as a comment keyword based on (reverse appearance frequency).

According to another embodiment of the tagging device of the present invention,
The comment text is obtained from the blog (Web log) server or the bulletin board site server.
It is also preferable to further include comment collecting means for collecting comment text from the blog server or bulletin board site server.

According to another embodiment of the tagging device of the present invention,
The blog server is a Twitter (registered trademark) server of a mini blog server,
The comment keyword extracting means preferably extracts the hash tag as a comment keyword.

According to another embodiment of the tagging device of the present invention,
Comment text re-search means for further searching comment text including a hash tag using a twitter (registered trademark) server,
It is also preferable that the comment text acquired by the comment keyword re-search means is acquired by the comment text collection means.

According to the present invention, there is a tagging program for causing a computer mounted on an apparatus that accumulates comment text including post time and position information posted from an unspecified number of third parties to function,
Comment collecting means for collecting comment text including posting time and position information from the blog server;
Event position detecting means for detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
Comment keyword extracting means for extracting a word from a group of comment sentences for each event position, counting the appearance frequency for each word, and extracting a comment keyword whose appearance frequency is a predetermined number or more;
The computer is caused to function as tagging means for associating a comment keyword as a tag for each event position.

According to the present invention, there is a tagging method in a device that accumulates comment text including post time and position information posted from an unspecified number of third parties,
A first step of collecting comment text including a posting time and position information from a blog server;
A second step of detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
A third step of extracting a word from a group of comment sentences for each event position, counting an appearance frequency for each word, and extracting a comment keyword having an appearance frequency equal to or greater than a predetermined number;
And a fourth step of associating a comment keyword as a tag for each event position.

  According to the tagging apparatus, program, and method of the present invention, it is possible to tag a keyword in real time at an event position where a special trend different from usual is seen based on a large number of comment sentences.

It is a system configuration diagram in the present invention. It is a functional block diagram of the tagging server in this invention. It is a flowchart in an event detection part. It is the screen figure which displayed the keyword in the event position on the map in this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

  FIG. 1 is a system configuration diagram according to the present invention.

  According to FIG. 1, a blog (Web log) server (or bulletin board site server) 2 that publishes comment text posted by an unspecified number of third parties is connected to the Internet. The blog server 2 is a mini blog server such as a twitter server.

  Here, according to FIG. 1, it is assumed that a regional event is held in a specific region. A regional event is a region where a special trend different from usual is observed, that is, a region where people are concentrated only during the day and time like a festival. Many users are staying at this event position, and each user can post a comment text to the miniblog server 2 using the terminal 3 that the user owns. The terminal 3 is movable with the user and is a mobile terminal such as a mobile phone or a smartphone.

  Moreover, according to FIG. 1, the tagging server (apparatus) 1 which comprises this invention is further connected to the internet. The tagging server 1 collects a large number of comment sentences from the miniblog server 2. Of course, communicating with the mini-blog server 2 is not an essential component, and the tagging device (server) 1 may be one in which a large number of comment texts are stored in advance. Then, the tagging server 1 tags a keyword in real time at an event position where a special trend different from usual is seen based on these comment sentences. Other users can access the tagging server 1 using the terminal 4 that they own, and browse the event position and its keyword in association with each other.

  FIG. 2 is a functional configuration diagram of the tagging server in the present invention.

  According to FIG. 2, the tagging server 1 includes a communication interface unit 10 connected to the Internet, a comment text collection unit 111, an event position detection unit 112, a comment keyword extraction unit 113, a tagging unit 114, The event location storage unit 115, the map information transmission unit 116, and the comment keyword re-search unit 117 are included. These functional components are realized by executing a program that causes a computer mounted on the server to function.

[Comment text collection unit 111]
The comment text collection unit 111 collects comment text including the posting time and position information from the miniblog server (for example, twitter server) 2. These comment sentences are tweets with geotags, and include user id (user_id), transmission time (created_at), location information (loc_lat, loc_lng) (latitude / longitude information), and text (texts). Here, it is also preferable that the comment sentence collection unit 111 collects only comment sentences posted in a predetermined time period. This is because local events are held only during a certain period of time. Of course, instead of the comment text collection unit 111, a comment text storage unit that stores a large number of comment texts in advance may be used.

The position information in the comment text is expressed as follows, for example.
"I'm at (place name) http://t.co/BNw9NbL2"
“@XXX http://t.co/3acyFI3K”
"Touch (location name)! Http://t.co/4a7dabK"

  Moreover, it is also preferable that the comment sentence collection part 111 uses StreamingAPI with respect to a twitter server, for example. The Twitter API is basically a pull type, whereas the Streaming API allows a client to open a connection to a twitter server, send a request, and maintain the connection until disconnected. Meanwhile, the server can continue to send the response (posted comment text) to the client (tagging server 1) in real time. Further, according to the Streaming API, it is possible to set to collect only comment texts including position information or only comment texts including specific keywords. Then, the comment text collected by the comment text collection unit 111 is output to the event position detection unit 112.

[Event position detection unit 112]
The event position detection unit 112 detects an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period.

  FIG. 3 is a flowchart in the event detection unit.

(S31) Event position candidate detection (position information clustering)
Assume that the number of posts is concentrated in the area where the event is held. Therefore, first, the event position detection unit 112 plots the latitude / longitude information added to the comment text on a plane and clusters the positions to detect an event position (crowded area) where the comment text concentrates. . For this clustering, for example, DBSCAN (Density-Based Spatial Clustering) can be used. According to k-means, the number of clusters k needs to be given in advance, whereas according to DBSCAN, a relatively intuitive cluster can be extracted without giving the number of clusters in advance.

  DBSCAN is a clustering method for extracting clusters having an arbitrary shape that is not hyperspherical, and derives directly density-rechable at two points. From a certain seed point, the density reachable relationship is directly traced, and the reachable maximal set is extracted as one cluster.

  According to DBSCAN, two parameters are used: Eps (epsilon, distance) and MinPts (minimum points). For example, Eps = 0.0005 and MinPts = 10 may be given in advance. Further, these parameters may be changed according to a predetermined time period of positioning or a result of final address location estimation.

(S32) Exclusion of position information having a predetermined number of postings on a daily basis Next, the event position detecting unit 112 has a predetermined number of postings on a daily basis from among the candidate event positions in S31. Filter by excluding location information. As a place where the number of postings is always large, for example, there is “station”. A place like a station is not an event location where a special trend different from usual can be seen. For this reason, positions with a large number of postings on a daily basis are excluded from event position candidates.

(S33) Exclusion of event positions where only the number of users equal to or less than a predetermined ratio is present The event position detection unit 112 determines that the number of users who posted these comment sentences is equal to or less than a predetermined ratio value for each event position. If it is, the event position is excluded and narrowed down. In such an event position, the same user has posted comment texts many times, and is often not an event position where many users gather. Therefore, such event position candidates are excluded. For example, when the number of posts in the area is less than half (when the number of users in the area where 10 comment sentences are posted is 4 or less), this event position is excluded from the candidates.

(S34) Exclusion of comment text in which the sentence claimed by the poster itself is not described The event position detection unit 112 describes the text claimed by the poster himself among a large number of comment sentences posted in a predetermined time period. Exclude comment text that has not been posted. For example, it is often configured only with phrases specific to the location information registration site, such as “I'm at (location)” and “touch (location)!”. For this reason, only comment sentences with such phrases and comment sentences in which the sentences claimed by the posters are not described are excluded. This is because such comment text does not include a keyword suitable for specifying the local event.

The comment text of “x” does not describe the text claimed by the poster, and is excluded by S33.
(Comment text 1) ○ I arrived. The previous students were A size. @Tokyo Metropolitan Nishishinjuku Building http://t.co/3acyFI3K
(Comment 2) ○ I went to OSC 2011 Tokyo / Fall # osc11tk (@ Tokyo Metropolitan Government Nishi-Shinjuku Building w / 3 others) http://t.co/BNw9NbL2
(Comment Sentence 3) × I'm at Tochō Nishi-Shinjuku Building, 28th floor http://t.co/v66BR2Iv
(Comment text 4) ○ Tsuita (@ Tocho Nishi-Shinjuku Building w / 4 others) http://t.co/RZkmuSG8 ',
(Comment Sentence 5) x I'm at Toshin Nishishinjuku Building w / @hirokazu_nishi http://t.co/HdCZoVD4 ',
(Comment 6) ○ Participated in # osc11tk. No, but it's a large campus. (@Tokyo Metropolitan Nishi Shinjuku Building w / 7 others) http://t.co/V1YAzqZg
(Comment text 7) x I'm at Tokyo Metropolitan Government Nishi-Shinjuku Building w / @ minky0 http://t.co/5aYa34CN
(Comment Sentence 8) x I'm at Tokyo Metropolitan Government Nishi-Shinjuku Building D2-1-1, City C) w / 10 others http://t.co/Uxcc0IrB

(S35) Exclusion of event position in which a keyword with an appearance frequency equal to or higher than a predetermined threshold is not detected The event position detection unit 112 determines that the event is detected when there is no word with the appearance frequency equal to or higher than the predetermined threshold for each event position. Filter by excluding position. When it is an event position, it is considered that a keyword having an appearance frequency of a certain level or more is always detected according to the contents of the event. That is, the keyword is often appropriate for expressing the contents of the event. On the other hand, if a keyword with an appearance frequency of a certain level or more is not detected, it is considered that a specific event is not held. Therefore, such event position candidates are excluded.

[Comment keyword extraction unit 113]
The comment keyword extraction unit 113 extracts words from the group of comment sentences for each event position, counts the appearance frequency for each word, and extracts a comment keyword whose appearance frequency is a predetermined number or more. The extracted comment keyword is output to the tagging unit 114.

  The comment keyword extraction unit 113 first extracts words from a plurality of comment sentences by morphological analysis. “Morphological analysis” refers to a technique in which sentences are divided into meaningful words and the part of speech and contents are discriminated using a dictionary. The “morpheme” means the smallest unit having meaning among the elements of the sentence.

  Here, for example, an open source morphological analysis engine “MeCab” can be used for the morphological analysis. This engine has a hierarchical part-of-speech system and can also analyze morpheme part-of-speech. For each morpheme, parts of speech such as “noun”, “proprietary noun”, “organization”, “region”, “general”, etc. are also output. However, according to the present invention, it is preferable to exclude place names and addresses from the comment keywords. This is because the place name itself does not represent the contents of the event.

  Next, a characteristic keyword is extracted as a comment keyword by TF-IDF (Term Frequency-Inverse Document Frequency). TF-IDF is a technique that weights each word, expresses a document from a query in a vector space, and ranks the documents based on the similarity between the query and the query. The higher the ranked value, the more important keywords are recognized.

  Moreover, it is preferable that the comment keyword extraction unit 113 further extracts a hash tag (#) as a comment keyword for a comment sentence collected from, for example, a twitter server. The hash tag is used to link comment sentences on a common topic.

[Tagging unit 114]
The tagging unit 114 associates comment keywords as tags for each event position. Then, the event position (latitude / longitude information) + comment keyword is output to the event position storage unit 115.

[Event position storage unit 115]
The event position storage unit 115 stores an event position (latitude / longitude information) + a comment keyword.

[Map information transmission unit 116]
The map information transmission unit 116 transmits map information and a comment keyword drawn at an event position included in the map to the terminal 3 that can be operated by the user.

  FIG. 4 is a screen diagram in which keywords are displayed at event positions on a map according to the present invention.

  According to FIG. 4, characteristic comment keywords are displayed at event positions on the map. Here, the comment keywords “# osc11tk”, “OSC2011 Tokyo / Fall”, and “University A” are displayed at the event location of the Tokyo Metropolitan Government Office Nishi-Shinjuku based on the comment text of FIG. By viewing this map, the viewer can intuitively recognize that an event based on the comment keyword is held at the location of the Tokyo Metropolitan Government Office Nishi-Shinjuku.

[Comment keyword re-search unit 117]
The comment keyword re-search unit 117 further searches using the twitter (registered trademark) server using the comment keyword output from the comment keyword extraction unit 113 as a key. The comment keyword includes, for example, a hash tag (#) and an event proper noun, and a comment sentence including these is re-searched.

  For example, in the case of twitter, there is only about 0.5% of the comment text with geotag (2010 statistics in Japan). On the other hand, there are many comment sentences to which position information is not added despite being posted from the event position. Therefore, the comment keyword re-search unit 117 further searches for comment sentences including the comment keyword, and these comment sentences are output to the comment sentence collection unit 111. As a result, the number of comment sentences that can be collected by the comment sentence collecting unit 111 can be increased. A characteristic keyword based on the event position is further extracted from the comment text to which position information is not added.

For example, according to the twitter server, it is possible to search for comment sentences (tweets) including keywords specified as follows. For example, the keyword "OSC2011
When searching for tweets containing “Tokyo / Fall”, the following URL (Uniform Resource Locator) is specified.
http://search.twitter.com/search.atom?phrase="OSC2011 Tokyo / Fall "

For example, according to twitter, it is also possible to collect only comment text tweets with a hash tag (#) added. In the case of twitter, the topic can be made common by using a hash tag. For example, when searching for a tweet to which the hash tag “# osc11tk” is added, the following URL is specified.
http://search.twitter.com/search.atom?q=%23osc11tk
You can also search for tweets for up to one week by specifying the date.

  As another embodiment, it is also preferable to detect a user who has posted the comment sentence from the comment sentence from which the comment keyword is extracted, and to re-search the comment sentence posted by the user. The end time of the event can also be extracted from the entire comment text of the user. For example, keywords such as “end”, “closed”, and “close” can be detected, and the posting time can be extracted. Based on the end time extracted in this way, the item (event position-keyword group) stored in the event position storage unit 115 can also be deleted.

Further, as another embodiment, it is preferable to further add a category representing the meaning of the comment keyword on the map. According to FIG. 4, although “OSC2011 Tokyo / Fall” is described, for example, “OSC2011
“Tokyo / Fall” and the category are also described. In order to realize such a description, it is necessary to create an event-related word list prepared in advance for each event. For example, “OSC-> society” is stored in the event-related word list, and the category “society” is searched by including “OSC”. Thus, by adding a category in addition to the comment keyword, the viewer can be made to understand more intuitively.

  As described above in detail, according to the tagging apparatus, program and method of the present invention, real-time keywords are tagged at event positions where special trends different from usual are observed based on a large number of comment sentences. can do. For a specific area, it is possible to detect what is happening in the area from the posted text related to a predetermined event transmitted by the poster and assign the keyword to the position. The viewer can intuitively recognize what event is being held at the position by browsing the map in which the comment keyword is associated with the position (event position).

  Various changes, modifications, and omissions of the above-described various embodiments of the present invention can be easily made by those skilled in the art. The above description is merely an example, and is not intended to be restrictive. The invention is limited only as defined in the following claims and the equivalents thereto.

DESCRIPTION OF SYMBOLS 1 Tagging server 10 Communication interface part 111 Comment text collection part 112 Event position detection part 113 Comment keyword extraction part 114 Tagging part 115 Event position memory | storage part 116 Map information transmission part 117 Comment keyword re-search part 2 Miniblog server 3 Contributor Terminal 4 Reader terminal

Claims (12)

A tagging device that accumulates comment sentences including posting time and position information posted from an unspecified number of third parties,
Event position detecting means for detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
Comment keyword extraction means for extracting a word from a group of comment sentences for each event position, counting the appearance frequency for each word, and extracting a comment keyword with an appearance frequency equal to or greater than a predetermined number;
A tagging device comprising tagging means for associating the comment keyword as a tag for each event position.   The event position detecting means excludes the event position and narrows down the event position when the number of users who have posted these comment sentences is equal to or less than a predetermined ratio value with respect to the total number of comment sentences for each event position. The tagging apparatus according to claim 1.   The event position detecting means extracts only comment sentences including position information from a large number of comment sentences posted in a predetermined time zone, and extracts comment sentences that do not contain sentences claimed by the posters themselves. 3. The tagging device according to claim 1, wherein the tagging device is narrowed down by exclusion.   4. The event position detection unit according to claim 1, wherein, when there is no word having the appearance frequency equal to or higher than a predetermined threshold for each event position, the event position detection unit excludes the event position and narrows down the event position. 5. The tagging apparatus according to any one of the above.   5. The event position detection means detects the event position using a DBSCAN (Density-Based Spatial Clustering) algorithm with position information included in the comment text as an element. 5. The tagging apparatus according to claim 1.   Map information transmitting means for transmitting the map information indicating the event position stored by the tagging means and the comment keyword to be described in the part of the event position on the map to a user operable terminal. The tagging device according to claim 1, wherein the tagging device is provided. The comment keyword extraction means extracts a word from a plurality of comment sentences by morphological analysis, and extracts a characteristic word from the comment keyword by TF-IDF (Term Frequency-Inverse Document Frequency). The tagging device according to claim 1, wherein the tagging device is extracted as The comment text is obtained from a blog (Web log) server or a bulletin board site server,
8. The tagging apparatus according to claim 1, further comprising comment collection means for collecting comment text from the blog server or bulletin board site server. The blog server is a mini blog server twitter (registered trademark) server,
9. The tagging apparatus according to claim 8, wherein the comment keyword extraction unit extracts a hash tag as the comment keyword. Comment keyword re-search means for further searching the comment text including the hash tag using the twitter (registered trademark) server,
The tagging apparatus according to claim 9, wherein the comment text acquired by the comment keyword re-search means is acquired by the comment text collection means. A tagging program that allows a computer mounted on a device that accumulates comment text including post time and position information posted from an unspecified number of third parties to function,
From the blog server, a comment collecting means for collecting comment text including posting time and position information;
Event position detecting means for detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
Comment keyword extraction means for extracting a word from a group of comment sentences for each event position, counting the appearance frequency for each word, and extracting a comment keyword with an appearance frequency equal to or greater than a predetermined number;
A tagging program that causes a computer to function as tagging means for associating the comment keyword as a tag for each event position. A tagging method in a device that accumulates comment text including post time and position information posted from an unspecified number of third parties,
A first step of collecting comment text including posting time and position information from the blog server;
A second step of detecting an event position where a predetermined number or more of comment sentences are concentrated in a predetermined position range using a large number of comment sentences posted in a predetermined time period;
A third step of extracting words from a group of comment sentences for each event position, counting the appearance frequency for each word, and extracting a comment keyword having an appearance frequency equal to or greater than a predetermined number;
And a fourth step of associating the comment keyword as a tag for each event position.

Images