JP5980708B2 - Kanji estimation program, apparatus, server, and method for estimating kanji symbolizing target sentence - Google Patents

Description

  The present invention relates to a technique for automatically generating a summary from a target sentence.

  In recent years, with the progress of text summarization technology, a long target sentence described in a Web page can be compressed in a short time and viewed with a browser. When the summary is further advanced, such target sentences can be condensed into sentences and keywords. However, the text summarization technique retains a logical group of meanings, and usually only generates sentences such as paragraphs.

  In the text summarization technique, words are extracted from target sentences, and sentences containing these words are connected to generate a summary sentence. Recently, there are some which summarize a plurality of target sentences, but in any case, they are generally reconstructed as logical sentences (see, for example, Non-Patent Document 1). In addition, there is a technique for assigning various features to sentence elements in order to generate a summary with high content coverage and high readability (see, for example, Patent Document 1). On the other hand, there is also a technique for easily searching by applying various annotations to a target sentence (for example, see Non-Patent Document 2). According to this technique, various attributes are given from the word distribution and formal characteristics in the target sentence, and humans rarely refer to the annotated information itself.

  In order for humans to understand only the summary text without reading the target text that is the original text, the summary text needs to have a certain length of text, and it is necessary to reflect the intention of the original text . However, it is generally very difficult to meet the required level. On the other hand, there is a technique for presenting titles and table-of-contents information displayed in a search result list or the like, but according to this technique, the information is only intermediate stage clue information for indicating the link destination.

The text summary and annotation have the following relationship.
Text summary
(Characteristics) It is difficult to generate summary sentences in which a plurality of sentences are logically reconstructed with a certain length.
It takes time to understand the summary text (from the user's perspective).
[Annotation]
(Characteristics) Create a summary sentence basically in sentence units.
Use summary text as a clue to search rather than direct reference (from user perspective).

JP 2011-243166 A

Eiwa Namba and Manabu Okumura, “Automatic text summarization so far”, Information Processing Vol43, 2002 No.12, [online], [February 11, 2013 search], Internet <URL: http: // www. ls.info.hiroshima-cu.ac.jp/~nanba/pdf/IPSJ-MGN431203_2002.pdf> Yoichi Kajiura and Emi Ishida, “Annotations of Abstracts for Advanced Academic Paper Search”, Text Annotation Workshop Contest (National Institute of Informatics), 2012, [online], [February 11, 2013 Search], Internet <URL: http: //nlp.nii.ac.jp/tawc/papers/C03_tomiura.pdf> `` Kanji two-character idioms table created using 2965 kanji JIS '', Bulletin of Faculty of Education, Nagoya University, Vol.44, pp.243-299, 1997. "Quotation Navi", [online], [Search February 11, 2013], Internet <URL: http://www.meigennavi.net/>

  For example, as an annual news, the degree of attention to the vote result of “this year's kanji” representing the social aspect of the year is high. Here, the inventors of the present application thought that by further proceeding with text summarization, the target sentence could be symbolized to a single “Kanji” character. That is, I thought that even target sentences such as news articles and comment sentences written by humans could be symbolically expressed in a single “Kanji” that can be sympathized.

  Then, an object of this invention is to provide the Chinese character estimation program, apparatus, server, and method which can estimate the single Chinese character which symbolizes a target sentence.

According to the present invention, a kanji estimation program for causing a computer to function to estimate a kanji symbolizing a target sentence,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying topic words into kanji units and associating appearance frequencies of the predetermined number of topic words for each kanji;
The computer is caused to function as a kanji estimation means for estimating a predetermined number of kanji characters having a high appearance frequency as a kanji symbolizing the target sentence.

According to another embodiment of the kanji estimation program of the present invention,
It is also preferable to make the computer function so that the appearance frequency is a TF-IDF (Term Frequency-Inverse Document Frequency) value.

According to another embodiment of the kanji estimation program of the present invention,
It is also preferable that the computer function so that the appearance frequency is a DF (Document Frequency) value representing the number of target sentences that have appeared.

According to another embodiment of the kanji estimation program of the present invention,
The kanji estimation means uses a Japanese dictionary information that stores the number of idioms including the kanji for each kanji, and estimates a predetermined number of higher kanji characters with a high appearance frequency and a small number of idioms as a kanji symbolizing the target sentence < It is also preferable to make the computer function like this.

According to another embodiment of the kanji estimation program of the present invention,
Kanji estimation means
Adjustment point = (appearance frequency / number of topic words) / √ (number of idioms)
Appearance frequency: Appearance frequency of words containing the kanji
Number of topic words: The number of topic words including the kanji in the predetermined number of topic words
Number of idioms: Estimate a kanji higher than a predetermined number with a high adjustment point calculated by the number of idioms including the kanji based on national language dictionary information as a kanji symbolizing the target sentence. It is also preferable to make the computer function.

According to another embodiment of the kanji estimation program of the present invention,
A plurality of target sentence storage means for storing the target sentences separately based on predetermined conditions,
In addition to the appearance frequency for each word, the topic word extraction unit preferably causes the computer to function so as to extract a predetermined number of topic words that appear characteristically by comparing different target sentence storage units.

According to another embodiment of the kanji estimation program of the present invention,
A similar word accumulating unit that stores a second word that is similar to the first word and that is composed of only kanji characters;
Using the similar word storage means, the computer functions as a topic word conversion means that outputs a second word similar to the topic word (first word) output from the topic word extraction means to the kanji classification means as a topic word It is also preferable that

According to another embodiment of the kanji estimation program of the present invention,
And further having a word storage means for storing kanji and quote text in association with each other,
It is also preferable to cause the computer to function so as to output the kanji output from the kanji estimation means and the noun text corresponding to the kanji using the noun accumulation means.

According to another embodiment of the kanji estimation program of the present invention,
It is also preferable to cause the computer to function so that the target sentence is a comment sentence posted to the comment release server by an unspecified number of third parties.

According to the present invention, an apparatus for estimating a kanji symbolizing a target sentence,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying topic words into kanji units and associating appearance frequency of a predetermined number of topic words with each kanji,
It has a kanji estimation means for estimating a kanji of a predetermined number higher in appearance frequency as a kanji symbolizing the target sentence.

According to the present invention, a comment publishing server that publishes comment text posted by an unspecified number of third parties, a terminal that is based on a user operation, and a network that can communicate with the terminal, and estimate kanji that symbolizes the target text A kanji estimation server that
Target sentence collection means for collecting comment sentences as target sentences from the comment public server,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying topic words into kanji units and associating appearance frequency of a predetermined number of topic words with each kanji,
It has a kanji estimation means for estimating a kanji of a predetermined number higher in appearance frequency as a kanji symbolizing the target sentence.

According to the present invention, using a device, a kanji estimation method for estimating a kanji symbolizing a target sentence,
A first step of extracting words from the target sentence by morphological analysis;
A second step of extracting a characteristic predetermined number of topic words (words) according to the appearance frequency of each word;
A third step of classifying the topic words into kanji units and associating the appearance frequency of a predetermined number of topic words for each kanji;
And a fourth step of estimating a predetermined number of kanji characters having a high appearance frequency as kanji characters that symbolize the target sentence.

  According to the kanji estimation program, apparatus, server, and method of the present invention, it is possible to estimate one kanji character that symbolizes the target sentence.

It is a functional block diagram of the Chinese character estimation apparatus in this invention. It is the table | surface which sorted "kanji" by the kanji estimation part of this invention. It is a system configuration diagram in the present invention. It is a sequence diagram in the present invention.

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

  FIG. 1 is a functional configuration diagram of a Chinese character estimation apparatus according to the present invention.

  According to FIG. 1, the kanji estimation device 1 includes a target sentence storage unit 10, a morpheme analysis unit 11, a topic word extraction unit 12, a kanji classification unit 13, a kanji estimation unit 14, and a similar word storage unit 15. , A topic word conversion unit 16, a kanji output unit 17, and a quote storage unit 18. These functional components can be realized by executing a kanji estimation program that causes a computer mounted on the server to function.

[Target sentence storage unit 10]
The target sentence storage unit 10 stores a large number of target sentences based on predetermined conditions. Here, a set of target sentences based on the predetermined condition is output to the morphological analysis unit 11. “A set of target sentences” is a set in a temporal / spatial range having some meaning, and is intended to extract a single kanji character symbolizing the set.

The target sentence storage unit 10 may store, for example, the following target sentences.
(1) Target texts related to news articles Collect major news articles from sites that contain genres such as politics and economy.
(2) Target texts related to the online community Collect, for example, posts to the movie review community.
(3) Target sentences related to twitter (registered trademark) Collect tweets, for example, within a certain time range (for example, one day) of an individual.

  For example, from a target sentence related to a news article for one year, a single “Kanji” character representing the social aspect of the year can be extracted. Moreover, it is not a wide object such as one-year social conditions, and may be limited to a short time such as one day or one week in terms of time, or spatially a specific region or a specific community in the country. You may limit to such a range. In addition, kanji that symbolizes the day can be estimated from the tweets of the individual for one day. A user who is in contact with such a “kanji” character can intuitively understand the atmosphere and sense in the time and space in which the target sentence occurred from the atmosphere and sense of the kanji.

[Morphological analyzer 11]
The morpheme analysis unit 11 inputs the target sentence from the target sentence storage unit 10. Here, by inputting a set of target sentences within a certain range, it is possible to estimate “kanji” that symbolizes the entire target sentences as a result. The morpheme analysis unit 11 extracts words from these target sentences by morpheme analysis. Here, “morphological analysis” refers to a technique for dividing a sentence into meaningful words and discriminating parts of speech and contents using a dictionary. The “morpheme” means the smallest meaningful unit (including a word) among sentence elements.

[Topic word extraction unit 12]
The topic word extraction unit 12 extracts a predetermined number of characteristic topic words according to the appearance frequency of each word. The simplest method may be simple appearance frequency = number of appearances. The predetermined number may be, for example, the top 100 topic words.

  Here, the appearance frequency may be a TF-IDF (Term Frequency-Inverse Document Frequency) value. TF-IDF refers to an index for identifying how characteristic a word that appears in a sentence is. TF (term frequency) represents the number of times the word appears in the sentence, and IDF (inverse document frequency) represents the natural logarithm of the number of sentences including the sentence in the entire corpus. Then, TF value × IDF value becomes the TF-IDF value of the word in the sentence. A word having a higher TF-IDF value is recognized as having a higher topicality (important).

  Furthermore, the appearance frequency may be a DF (Document Frequency) value representing the number of target sentences in which the topic word appears. For example, when each target sentence is each news article, the DF value is the number of articles including the topic word. For example, when each target sentence is a tweet, the DF value is the number of tweets including the topic word.

  The topic word extraction unit 12 may extract a predetermined number of topic words that appear characteristically by comparing different target sentence storage units 10 in addition to the appearance frequency of each word. For example, a target sentence storage unit that stores news articles of 2011 as target sentences, a target sentence storage unit that stores news articles of the past several years as target sentences, and a target sentence storage that stores news articles of this year as target sentences. Part. At this time, there are also topic words with high appearance frequency every year. Therefore, for example, it is preferable to extract only topic words unique to the current year as topic words of the year. Specifically, it can be recognized as a topic word only when the DF value (number of articles) in which the topic word appears is prominently higher than other average years. That is, a topic word that appears every year is not a topic word representing that year.

[Kanji classification part 13]
The kanji classifying unit 13 classifies the topic word into kanji units and associates the appearance frequency of the topic word with each kanji. As described above, the appearance frequency itself may be used, or a DF value or a TF-IDF value may be used. For example, the DF value for each Chinese character is derived as follows.
Topic "Earthquake" (DF value = 30)->
DF value of earthquake derived from “earthquake disaster” + = 30
DF value of disaster derived from “earthquake disaster” + = 30
Similarly, for "earthquake", from "earthquake""seismicintensity" ...
About "disaster", from "disaster""disaster" ...
Since the DF value corresponding to each topic word to be derived is obtained,
By summing them
-> Earthquake DF = 267
-> Disaster DF value = 234

  As another embodiment, the kanji division unit 13 may output only kanji including a predetermined radical to the kanji estimation unit 14. For example, only a kanji character including the radical “heart” may be output. Many Kanji characters that include the radical “heart” represent human psychology and emotions. Thus, by specifying the radical of the Chinese character output from the Chinese character classification unit 13, the Chinese character can be estimated according to the specific purpose to be recognized by the user.

[Kanji estimation unit 14]
The kanji estimation unit 14 estimates a predetermined number of upper kanji characters having a high appearance frequency as kanji symbols representing the target sentence.

  FIG. 2 is a table in which “kanji” are sorted by the kanji estimation unit of the present invention.

  According to FIG. 2, the top 20 kanji characters that can be said to be “this year's kanji” are represented from the target sentences related to news articles for one year. According to FIG. 2, it is sorted in descending order based on the DF value of each Chinese character. For example, it is also preferable that only topic words that are a predetermined threshold times or more than the average appearance frequency (DF value: number of articles) in the past several years are estimated as kanji characters for this year. The predetermined threshold value can be adjusted as a parameter, and about 1.5 to 2.0 times is appropriate, and may be 1.7 times, for example.

  On the other hand, extracting the upper Kanji characters using only the DF value or the TF-IDF value means that the original characteristics of the Kanji characters are not reflected. As a next step, it is conceivable to sort the kanji in descending order in consideration of the “word-making ability (or semantic diversity)” of the kanji. As the word-making ability, the number of idioms that contain a single kanji character is used.

  Therefore, as another embodiment, the kanji estimation unit 14 multiplies the appearance frequency by using the kanji dictionary information in which the number of idioms including the kanji is stored, so that the kanji having a smaller number of idioms is sorted higher. Is also preferable. According to the existing national language dictionary, the “number of idioms (different number)” of kanji idioms using commonly used basic kanji can be known (see, for example, Non-Patent Document 3). As a result, the kanji can be sorted to reflect the diversity of kanji formation. In other words, it is estimated that the higher the ratio of topic words (including idioms) to the number of idioms including a certain Chinese character, the higher the degree of attention to the Chinese character.

  As the tendency of the higher-order kanji estimated as “this year's kanji” in FIG. 2, not only the appearance frequency but also the degree related to the formation of various words and the degree of emotional meaning are high. That is, it means that the kanji that is related to the formation of various words and has a high degree of emotional meaning attracts more attention than the actual appearance frequency. Therefore, according to the present invention, these attributes are quantified to adjust the sorting based on the appearance frequency.

As a specific example of the preceding paragraph, the following parameters are used as adjustment parameters for sorting kanji.
(1) Number of idioms (of topic word) in the kanji
(2) Number of articles including topic words in the kanji
(3) b / a = average number of appearances per topic word (4) number of idioms including the kanji (basic measure of semantic diversity of kanji)
And the kanji estimation part 14 sorts kanji by the adjustment point calculated by the following formula | equation.
Adjustment point = (appearance frequency / number of topic words) / √ (number of idioms)

  According to FIG. 2, the kanji characters are sorted in the DF order in the left column, and the kanji characters are sorted in the adjustment point order in the right column. According to the adjustment points, in 2011, kanji characters are estimated in the order of “earthquake”-> “disaster”-> “electricity”-> “cover”->. When the topic word is a proper noun, a value obtained by multiplying the adjustment point assigned to the kanji into which the topic word is divided by 0.1 is given.

[Similar word storage unit 15]
The similar word accumulation | storage part 15 matches and memorize | stores the 2nd word comprised only with the Chinese character similar to a 1st word. For example, the correspondence is as follows.
"Olympic Games"<->"OlympicGames"

[Topic Word Conversion Unit 16]
The topic word conversion unit 16 uses the similar word accumulation unit 15 to output a word similar to the topic word output from the topic word extraction unit 12 to the kanji classification unit 13 as a topic word. Here, if the topic word is Katakana, it is converted into a “kanji word” indicating the same object. According to the above example, the topic word “Olympic” is converted to “Olympic”. The converted “Golden Olympics” is output to the Chinese character classifying unit 13 and is divided into the Chinese characters “Gold” and “Woman”.

  FIG. 3 is a system configuration diagram according to the present invention.

  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. According to FIG. 3, the miniblog server 2 is further connected to the Internet. An unspecified number of third-party comment contributors can use the terminal 4 to post comment texts to the miniblog server 2 and to view other person's comment texts. The terminal 4 may be a mobile terminal or a smartphone, for example.

  According to the present invention, the Chinese character estimation server 1 is further connected to the Internet. The kanji estimation server 1 collects a large number of comment sentences (target sentences) within a certain range from the miniblog server 2. Then, the kanji estimation server 1 estimates “kanji” (one character) related to these comment sentences. The estimated “kanji” is transmitted to the user-operated terminal 3 so that the user can recognize it.

  According to FIG. 3, compared with FIG. 1, it has further the target sentence collection part 17, the Chinese character transmission part 18, and the noun accumulation | storage part 19. FIG.

[Target sentence collection unit 17]
For example, the target sentence collection unit 17 collects comment sentences (target sentences) from the miniblog server 3. These comment sentences are output to the target sentence storage unit 10. For example, when a posted sentence of twitter (registered trademark) is used as a target sentence, for example, tweets for a certain time range of an individual, for example, one day can be collected.

[Kanji transmitter 18]
The kanji transmitting unit 18 transmits the estimated upper kanji characters to the terminal 3. Here, the predetermined number may be one Chinese character or a plurality of Chinese characters. A user who operates the terminal 3 can know the atmosphere and feeling of the kanji.

As a specific embodiment, a case will be described in which a post sentence to the movie review community is a target sentence. By estimating a single kanji character from those target sentences, it is possible to know the atmosphere and feeling of the movie. For example, in the case of a musical movie, it is assumed that there are 120 postings in one week for the community.
It is assumed that the topic word extraction unit 12 has extracted the following topic words.
"Youth""France""Love""Revolution"
Next, the topic word conversion unit 16 converts the topic words as follows.
"France"->"Buddha"
Next, the Chinese character classification unit 13 classifies the Chinese characters as follows.
"Blue""Year""Buddha""Love""Emotion""Leather""Life"
And the kanji estimation part 14 estimates as follows the kanji sorted by the upper rank.
“Love” “Leather” “Buddha” “Blue” “Emotion”…
Finally, when there is one kanji character that is clearly indicated to the user, only “love” is output.

[Quote Accumulation Unit 19]
The quote accumulation unit 19 accumulates kanji and quote sentences in association with each other. For example, it is assumed that they are associated as follows.
Kanji "Error"-> Quote "Monkey falls from the tree"
Here, it is assumed that the kanji transmitting unit 18 tries to transmit the kanji “wrong”. At this time, referring to the quote storage unit 18, the quote “monkey also falls from the tree” is transmitted together with the Chinese character “false”. This allows the user to know not only kanji but also various words such as haiku and kawayanagi that are deeply linked to the kanji.
As another embodiment, the kanji transmitting unit 18 may transmit the extracted kanji to the quote server and receive the quote text corresponding to the kanji (see, for example, Non-Patent Document 4).

  FIG. 4 is a sequence diagram in the present invention.

(S10) The kanji estimation server 1 collects the target text (see the target text collection unit 17 described above). The collected target sentences are stored in the target sentence storage unit 10.
(S11) A word is extracted from the target sentence by morphological analysis (see the morphological analysis unit 11 described above).
(S12) A characteristic predetermined number of topic words (words) are extracted according to the appearance frequency for each word (see the topic word extraction unit 12 described above). For example, a topic word such as katakana is converted into a topic word composed of kanji (see the topic word conversion unit 16 described above).
(S13) The topic words are divided into kanji units, and the appearance frequency of the topic word is associated with each kanji (see the above-described kanji classification unit 13).
(S14) Predetermined upper kanji characters having a high appearance frequency are estimated as kanji characters that symbolize the target sentence (see the kanji character estimation unit 14 described above).
(S15) Here, together with the estimated Chinese character, a quotation corresponding to the Chinese character may be searched (see the above-described quote accumulation unit 19).
(S16) Finally, the kanji estimation server 1 transmits the kanji (or its quote) symbolizing the target sentence to the terminal 3 based on the user operation.

  As described above in detail, according to the kanji estimation program, apparatus, server, and method of the present invention, it is possible to estimate a single kanji character that symbolizes the target sentence. It is possible to estimate a single kanji character that symbolizes (summarizes) a large number of posted texts of people who occur in various time and space ranges. This allows the user to intuitively understand and sympathize with the atmosphere and sensation that the Chinese character has. In this respect, it differs greatly from text summarization and annotation in the prior art.

  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 Kanji estimation server 10 Target sentence accumulation | storage part 11 Morphological analysis part 12 Topic word extraction part 13 Kanji classification part 14 Kanji estimation part 15 Similar word accumulation part 16 Topic word conversion part 17 Target sentence collection part 18 Kanji transmission part 19 Nominal accumulation part 2 Mini-blog server 3 Terminal 4 Comment author's terminal

Claims (12)

A kanji estimation program that causes a computer to function to estimate kanji that represents a target sentence,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying the topic words into kanji units and associating the appearance frequency of the predetermined number of topic words for each kanji;
A kanji estimation program for causing a computer to function as kanji estimation means for estimating a predetermined number of kanji characters having a high appearance frequency as a kanji symbolizing the target sentence.   The kanji estimation program according to claim 1, wherein the computer functions so that the appearance frequency is a TF-IDF (Term Frequency-Inverse Document Frequency) value.   2. The kanji estimation program according to claim 1, wherein the computer functions so that the appearance frequency is a DF (Document Frequency) value representing the number of target sentences that have appeared. The kanji estimation means uses the national language dictionary information storing the number of idioms including the kanji for each kanji, and uses the kanji that symbolizes the target sentence as the kanji that is higher in frequency and has a lower number of idioms. Estimate as
Kanji estimation program according to any one of claims 1 to 3, characterized in that to function by Uni computer. The kanji estimation means includes
Adjustment point = (appearance frequency / number of topic words) / √ (number of idioms)
Appearance frequency: Appearance frequency of words containing the kanji
Number of topic words: Number of topic words including the kanji in the predetermined number of topic words
Number of idioms: Estimates a kanji character higher by a predetermined number as a kanji symbolizing the target sentence in descending order of adjustment points calculated by the number of idioms including the kanji character based on the national language dictionary information <br / The computer program of claim 4 , wherein the computer functions as follows. A plurality of target sentence storage means for storing the target sentences separately based on predetermined conditions,
The topic word extracting means causes the computer to function to extract a predetermined number of topic words that appear characteristically by comparing different target sentence storage means in addition to the appearance frequency for each word. The kanji estimation program according to any one of claims 1 to 5. A similar word accumulating unit that stores a second word that is similar to the first word and that is composed of only kanji characters;
Topic word conversion means for outputting, to the kanji classification means, a second word similar to the topic word (first word) output from the topic word extraction means by using the similar word accumulation means. 7. The Chinese character estimation program according to claim 1, wherein the computer is caused to function as: And further having a word storage means for storing kanji and quote text in association with each other,
8. The computer according to claim 1, wherein the computer is caused to function so as to output the kanji output from the kanji estimation means and the quote text corresponding to the kanji using the noun accumulation means. The kanji estimation program according to item.   The kanji estimation according to any one of claims 1 to 8, wherein the target sentence is made to function as a comment sentence posted to a comment release server by an unspecified number of third parties. program. A device that estimates kanji that symbolizes the target sentence,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying the topic words into kanji units and associating the appearance frequency of the predetermined number of topic words for each kanji;
An apparatus for estimating kanji, comprising: a kanji estimation means for estimating a predetermined number of kanji characters having a high appearance frequency as a kanji symbolizing the target sentence. It is a comment release server that publishes comment sentences posted by an unspecified number of third parties, and a kanji estimation server that can communicate via a terminal and a network based on user operations and estimates kanji that symbolize the target sentences. And
Target sentence collection means for collecting comment sentences as target sentences from the comment public server,
Morphological analysis means for extracting words from the target sentence by morphological analysis;
Topic word extraction means for extracting a predetermined number of characteristic topic words (words) according to the appearance frequency of each word;
Kanji classification means for classifying the topic words into kanji units and associating the appearance frequency of the predetermined number of topic words for each kanji;
A kanji estimation server comprising kanji estimation means for estimating a predetermined number of upper kanji characters having a high appearance frequency as a kanji symbolizing the target sentence. A kanji estimation method for estimating a kanji symbolizing a target sentence using a device,
A first step of extracting words from the target sentence by morphological analysis;
A second step of extracting a characteristic predetermined number of topic words (words) according to the appearance frequency of each word;
A third step of classifying the topic words into kanji units and associating the appearance frequency of the predetermined number of topic words with each kanji;
A kanji estimation method, comprising: a fourth step of estimating a predetermined number of kanji characters having a high appearance frequency as kanji symbols representing the target sentence.

Images