JP2015170145A - Program, device, and server for estimating simple sentence symbolizing target sentence - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a program and the like that estimate a concise expression (simple sentence) symbolically expressing a target sentence, thereby enabling a user who encounters the target sentence to intuitively understand a semantic content of the target sentence only at a glance of the concise expression.SOLUTION: The device or server comprises: important word extraction means of extracting a plurality of characteristic important words according to a frequency of appearance of each word in the target sentence; first associative word extraction means of extracting a list of first associative words that cooccurrence-appear with the important words; second associative word extraction means of extracting a group of second associative word of lists that cooccurrence-appear with the simple sentences for each of the simple sentences stored in simple sentence storage means; associative words list matching means of searching the second associative word list, in the group of the second associative word lists for each of the simple sentences, which is the most similar to the first associative word list; and simple sentence estimation means of outputting the simple sentence based on the searched second associative word list as a sentence which symbolizes the target sentence.

Description

  The present invention relates to a technique for generating or selecting a sentence symbolizing the semantic content of 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 adding various features to sentence elements in order to generate a summary with high content coverage and 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 technology, 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.

JP 2011-243166 A

Eiwa Namba and Manabu Okumura, “Automatic text summarization so far”, Information Processing Vol 43, 2002 No.12, [online], [Search on February 11, 2014], Internet <URL: http: // www. ls.info.hiroshima-cu.ac.jp/~nanba/pdf/IPSJ-MGN431203_2002.pdf> Yoichi Kajiura and Emi Ishida, “Annotation of Abstracts for Advanced Academic Paper Search”, Text Annotation Workshop Contest (National Institute of Informatics), 2012, [online], [February 11, 2014 Search], Internet <URL: http: //nlp.nii.ac.jp/tawc/papers/C03_tomiura.pdf> National Institute of Information and Communications Technology, “Japanese WordNet” [online], [Search on February 11, 2014], Internet <URL: http://nlpwww.nict.go.jp/wn-ja/>

  For example, various news articles and blog articles can be intuitively compared to users by comparing them with "quote", "maximum", "proverb", "haiku", "kawayanagi" (hereinafter referred to as "simplified text") as concise expressions. Can be understood. Here, can the inventors of the present application associate a simple sentence as a concise expression instead of a text summary or annotation with the target sentence? I thought. In other words, is it possible to search for “concise expression” that can be sympathized as a person from target sentences such as news articles and blog sentences? I thought. Such a concise expression cannot be obtained by simply combining words appearing in the target sentence. If the concise expression with a high degree of convincing that condenses the semantic content of the target sentence cannot be specified, the user cannot understand and sympathize intuitively at first glance.

  Accordingly, an object of the present invention is to provide a program, an apparatus, and a server that can estimate a concise expression (simple sentence) that symbolically represents a target sentence.

According to the present invention, a program for causing a computer to function to estimate a simple sentence symbolizing a target sentence,
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
An important word extracting means for extracting a plurality of characteristic important words according to the frequency of appearance of each word in the input target sentence;
A first associative word extraction means for extracting a first associative word list that co-occurs with an important word among the associative words stored in the associative word storage means with reference to the corpus sentence storage means;
For each simple sentence stored in the simple sentence storage means, a second associative word list that co-occurs with the simple sentence among the association words stored in the associative word storage means with reference to the corpus sentence storage means Second associative word extracting means for extracting a group of
An associative word list matching means for searching for a second associative word list that is most similar to the first associative word list among a group of second associative word lists for each simple sentence;
The computer is caused to function as simple sentence estimation means for outputting a simple sentence based on the second associative word list searched by the associative word list matching means as a symbol of the target sentence.

According to another embodiment of the program of the present invention,
It is also preferable to make the computer function so that the simple sentences are quotes, maxims, proverbs, haikus, and / or Kawayanagi.

According to another embodiment of the program of the present invention,
The association word list matching means is
For each associated word in the first associated word list in the target sentence X, the semantic distance of each associated word in each second associated word list in the plurality of simple sentences C _{1 to} C _n is searched, and the distance is the shortest. The associative word pair is derived, and the average distance or cumulative distance of the semantic distance of the associative word pair is calculated.
It is also preferable to cause the computer to function so as to select a simple sentence based on the second associative word list having the shortest average distance or cumulative distance with respect to the first associative word list in the target sentence X.

According to another embodiment of the program of the present invention,
Corpus sentence storage means and / or associative word storage means are divided into a plurality of categories,
The first associative word extracting unit and / or the second associative word extracting unit selects a category to be referred to from the corpus sentence storage unit and / or the associative word storage unit according to the input category of the target sentence. It is also preferable to make the computer function.

According to another embodiment of the program of the present invention,
It is also preferable that the simple sentence estimation means causes the computer to function by selecting a predetermined number of simple sentences in ascending order of distance from the target sentence and outputting a plurality of simple sentences as estimation results.

According to another embodiment of the program of the present invention,
The keyword extraction means preferably causes the computer to function so as to extract a keyword using TF-IDF (Term Frequency-Inverse Document Frequency) or DF (Document Frequency). .

According to another embodiment of the program of the present invention,
A target sentence database storing a large amount of target sentences;
A target sentence search means for searching for one or more target sentences using the target sentence database for the input query and outputting the target sentences to the keyword extraction means;
The simple sentence estimation means preferably causes the computer to function so as to output a simple sentence corresponding to each target sentence as a search result.

According to another embodiment of the program of the present invention,
Simple sentence estimation means
Combine each target sentence with a simple sentence and output it as a search result,
Or
It is also preferable to cause the computer to function so that a plurality of simple sentences are output, the user selects any simple sentence, and the target sentence corresponding to the selected simple sentence is output again as a search result.

According to the present invention, a program for causing a computer to function to estimate a target sentence symbolizing a simple sentence,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, the corpus sentence storage means is referred to, and the first association word list group appearing together with the important word is extracted from the association words stored in the association word storage means. Associative word extraction means;
A second associative word list that co-occurs with the simple sentence is extracted from the associative words stored in the associative word storage means with reference to the corpus sentence storage means for the input simple sentence. Associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to the second associative word list among a group of first associative word lists for each target sentence;
The computer is made to function as target sentence estimation means for outputting the target sentence based on the first associative word list searched by the associative word list matching means as a symbol of a simple sentence.

According to the present invention, an apparatus for estimating a simple sentence symbolizing a target sentence,
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
An important word extracting means for extracting a plurality of characteristic important words according to the frequency of occurrence of each word in the target sentence;
A first associative word extraction means for extracting a first associative word list that co-occurs with an important word among the associative words stored in the associative word storage means with reference to the corpus sentence storage means;
For each simple sentence stored in the simple sentence storage means, a second associative word list that co-occurs with the simple sentence among the association words stored in the associative word storage means with reference to the corpus sentence storage means Second associative word extracting means for extracting a group of
An associative word list matching means for searching for a second associative word list that is most similar to the first associative word list among a group of second associative word lists for each simple sentence;
And a simple sentence estimation means for outputting a simple sentence based on the second associative word list searched by the associative word list matching means as a symbol of the target sentence.

According to the present invention, an apparatus for estimating a target sentence symbolizing a simple sentence,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, the corpus sentence storage means is referred to, and the first association word list group appearing together with the important word is extracted from the association words stored in the association word storage means. Associative word extraction means;
A second associative word list that co-occurs with the simple sentence is extracted from the associative words stored in the associative word storage means with reference to the corpus sentence storage means for the input simple sentence. Associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to the second associative word list among a group of first associative word lists for each target sentence;
And a target sentence estimation means for outputting the target sentence based on the first associative word list searched by the associative word list matching means as a symbol of a simple sentence.

According to the present invention, a server that receives a query from a terminal, searches for one or more target sentences corresponding to the query, estimates a simple sentence that symbolizes the target sentence, and returns the simple sentence to the terminal. There,
A target sentence database storing a large amount of target sentences;
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
A target sentence search means for searching for one or more target sentences using a target sentence database in response to the query;
An important word extracting means for extracting a plurality of characteristic important words according to the appearance frequency of each word in the searched target sentence;
A first associative word extraction means for extracting a first associative word list that co-occurs with an important word among the associative words stored in the associative word storage means with reference to the corpus sentence storage means;
For each simple sentence stored in the simple sentence storage means, a second associative word list that co-occurs with the simple sentence among the association words stored in the associative word storage means with reference to the corpus sentence storage means Second associative word extracting means for extracting a group of
An associative word list matching means for searching for a second associative word list that is most similar to the first associative word list among a group of second associative word lists for each simple sentence;
Simple sentence estimation means for outputting a simple sentence based on the second associative word list searched by the associative word list matching means as a search result symbolizing the target sentence;
It is characterized by having.

According to another embodiment of the server of the present invention,
Simple sentence estimation means
Combine each target sentence with a simple sentence and output it as a search result,
Or
It is also preferable to return a plurality of simple sentences to the terminal, receive the simple sentence selected by the user, and return the target sentence corresponding to the simple sentence as a search result.

According to the present invention, a server that receives a simple sentence from a terminal, estimates a target sentence that symbolizes the simple sentence, and returns the target sentence to the terminal,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, the corpus sentence storage means is referred to, and the first association word list group appearing together with the important word is extracted from the association words stored in the association word storage means. Associative word extraction means;
A second associative word list that co-occurs with the simple sentence is extracted from the associative words stored in the associative word storage means with reference to the corpus sentence storage means for the input simple sentence. Associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to the second associative word list among a group of first associative word lists for each target sentence;
And a target sentence estimation means for outputting the target sentence based on the first associative word list searched by the associative word list matching means as a symbol of a simple sentence.

  The program, apparatus, and server of the present invention can estimate a concise expression (simple sentence) that symbolically represents a target sentence. Thereby, the user who touched the target sentence can intuitively understand the semantic content of the target sentence only by looking at the concise expression.

It is a functional block diagram of the simple sentence estimation apparatus in this invention. It is a flowchart of this invention showing the process with respect to object text. It is a flowchart of this invention showing the process with respect to a simple sentence. It is a functional block diagram of the target sentence estimation apparatus which searches a target sentence by inputting a simple sentence. It is a system block diagram which has the simple sentence estimation server in this 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 simple sentence estimation apparatus according to the present invention.

  The simple sentence estimation device 1 inputs a target sentence and estimates a simple sentence that symbolizes the target sentence. The target sentence may be a news article or a blog article. Simple sentences are concise expressions, such as quotes, maxims, proverbs, haiku, and kawayanagi.

It is assumed that a news article on a single topic is input to the simple sentence estimation device 1 as a target sentence, for example, as follows.
-------------------------------------------------- -----------------------------
Title "A child's aunt dies of breast cancer" 2013/5/28 16:56
Movie actress A child (37) 's maternal aunt B child died at age 61 on May 26, 2013 at a hospital in Saitama Prefecture due to breast cancer.
A child announced that a genetic abnormality that increased the risk of breast cancer was found on the 14th, and that both breasts were excised and reconstructed. After child B was diagnosed with breast cancer in 2004, the same genetic abnormality as A child was found. "
-------------------------------------------------- -----------------------------
The simple sentence estimation device 1 can estimate a “say” (simple sentence) that symbolizes this inputted “news article” (target sentence).

  According to FIG. 1, the simple sentence estimation device 1 includes a corpus sentence storage unit 101, an associative word storage unit 102, an important word extraction unit 111, a first associative word extraction unit 112, and a simple sentence storage unit 121. The second associative word extraction unit 122, the associative word list matching unit 131, and the simple sentence estimation unit 132 are included. These functional components can be realized by executing a program that causes a computer installed in the apparatus to function. Also, the processing flow of these functions can be understood as a simple sentence estimation method using an apparatus.

  FIG. 2 is a flowchart of the present invention showing processing for the target sentence.

[Corpus sentence storage unit 101]
The corpus sentence storage unit 101 stores corpus sentences. Corpus means, in a narrow sense, a database in which natural language sentences are structured and accumulated on a large scale, but in a broad sense, it may be distributed sentences such as newspaper news articles and blog articles. . The corpus sentence storage unit 101 may be stored in the simple sentence estimation apparatus 1 or may be capable of searching for another corpus server via a network.

  As another embodiment, it is also preferable that the corpus sentence storage unit 101 stores a large number of corpus sentences according to categories. By changing the corpus sentence group to be referred to according to the category of the target sentence input to the simple sentence estimation device 1, the accuracy of the semantic content in the estimated simple sentence is also increased.

  The corpus sentence to be stored by the corpus sentence storage unit 101 is a sentence that includes both one simple sentence stored in the simple sentence storage unit 121 and at least one associative word stored in the associative word storage unit 102. It is included. The corpus sentence storage unit 101 is a corpus sentence database including a co-occurrence relationship between a simple sentence and an associative word.

According to FIG. 2, for example, the following are stored as corpus sentences.
-------------------------------------------------- -----------------------------
(C1) “In the future, genetic medicine will develop and the disease may disappear.”
(C2) “It ’s the cane that does n’t fall down. It ’s a prevention before you hurt and regret it.”
(C3) “Thank you! This is a ship for the crossing. This measure was good luck.
(C4) “Breast cancer is not scary if examined early”
(C5) “Oh, I was disqualified. It ’s so high in value and evaluation.
(C6) “It is more important to be aware of an incident and diagnose it immediately than to prevent it on its own.”
...
-------------------------------------------------- -----------------------------

[Associative Word Storage Unit 102]
The associative word storage unit 102 stores arbitrary plural associative words. The associative word is merely a group of words, but the content similarity between the target sentence and the simple sentence is derived through the associative word. The associative word storage unit 102 is referred to by both the first associative word extraction unit 112 and the second associative word extraction unit 122, and is one in common.
As another embodiment, it is also preferable that the associative word storage unit 102 stores a “theme associated word” related to the theme and an “emotion associated word” related to human emotion.

According to FIG. 2, the following are set as the associative words, for example.
-------------------------------------------------- -----------------------------
"Inspection""Surgery""Remorse""Treatment""Prevention""Doctor""Illness""Injury" ...
-------------------------------------------------- -----------------------------

[Important word extraction unit 111]
The important word extraction unit 111 extracts a plurality of characteristic important words according to the appearance frequency of each word in the target sentence. The important words explicitly included in the target sentence characterize the target sentence.

  The keyword extraction unit 111 first extracts words from the target sentence by morphological 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.

  Next, the important word extracting unit 111 extracts a plurality of characteristic important words according to the appearance frequency for each word. The simplest method may be simple appearance frequency = number of appearances. Further, the predetermined number of important words may be extracted as the top 100, for example, or the ratio of important words may be extracted as the top 30%, for example.

  As another embodiment, the appearance frequency may be a TF-IDF (Term Frequency-Inverse Document Frequency) value focusing on independent words. 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, as another embodiment, the appearance frequency may be a DF (Document Frequency) value representing the number of target sentences in which the important word appears. For example, for each target sentence, it is the number of important words.

  Furthermore, as another embodiment, when the target sentence is composed of a plurality of paragraphs (paragraphs), generally, the head sentence of each paragraph is often important. Therefore, the word included in the head part of the paragraph can be an important word candidate with its weight ω set to, for example, twice.

According to FIG. 2, it is assumed that, for example, the following important words are extracted from the above-described example of the target sentence.
-------------------------------------------------- -----------------------------
"Actress""Aunt""Hospital""Breastcancer""Risk""Gene"
"Abnormality""Breast""Resection""Surgery""Diagnosis"
-------------------------------------------------- -----------------------------

[First associative word extraction unit 112]
The first associative word extraction unit 112 refers to the corpus sentence storage unit 101, and among the associative words stored in the associative word storage unit 102, a first associative word list that appears together with the important word is displayed. Extract. An associative word is a word that does not appear explicitly in the target sentence but represents the semantic content of the target sentence.

According to FIG. 2, for example, the following corpus sentences are referred to from the above-described example of important words.
-------------------------------------------------- -----------------------------
(C1) “In the future, [gene] medicine will develop and [disease] may disappear”
(C4) “[Breast cancer] is not afraid if [inspection] is done early”
(C6) “It is important to [diagnose] immediately after recognizing an abnormality rather than self-sufficient [prevention]”
-------------------------------------------------- -----------------------------
In C1, an important word “gene” and an associative word “disease” appear together.
In C4, the important word “breast cancer” and the associative word “examination” appear together.
In C6, the important word “diagnosis” and the associated word “prevention” appear together.

As a result, the following first associative word list is extracted.
-------------------------------------------------- -----------------------------
(First association word list) "Disease""Test""Prevention"
-------------------------------------------------- -----------------------------

  FIG. 3 is a flowchart of the present invention showing processing for a simple sentence.

[Simple sentence storage unit 121]
The simple sentence storage unit 121 stores a plurality of simple sentences in advance. A simple sentence is a sentence that intuitively understands the emotional aspect of a person, such as “Quotations”, “Maxims”, “Proverbs”, “Haiku”, and “Kawayanagi”. A simple sentence as a concise expression is very short, and a metaphorical expression is often used. Therefore, extracting the word itself directly included in the simple sentence does not characterize the intention of the simple sentence.

According to FIG. 3, it is assumed that, for example, the following is stored as a simple sentence.
-------------------------------------------------- -----------------------------
"The tip of the tip that doesn't fall down" (Be prepared in advance so that it won't fail)
“Migration ship” (Blessed with people and environment that can help you when you are in trouble)
"Tama ni persimmon" (almost completely close, but has slight defects)
“Monkeys fall from trees” (even the masters of the road sometimes make mistakes)
...
-------------------------------------------------- -----------------------------

[Second associative word extraction unit 122]
For each simple sentence stored in the simple sentence storage unit 121, the second associative word extraction unit 122 refers to the corpus sentence storage unit 101, and among the association words stored in the associative word storage unit 102, the simple sentence The group of the 2nd associative word list which co-occurs and is extracted.

According to FIG. 3, the following is referred to as a corpus sentence, for example.
-------------------------------------------------- -----------------------------
(C2) “It ’s called“ the cane that does n’t fall ”. It ’s“ prevention ”before [injury] and [regret].”
(C3) “[Thanks]! This is [Migrating to the ship]. This [Countermeasure] was good [Luck]?”
(C5) “Ah ~ I was disqualified. That's why [value] and [evaluation] are so high.
-------------------------------------------------- -----------------------------
In C2, a simple sentence “an wand that doesn't fall” and associated words “injury”, “regret”, and “prevention” appear together.
In C3, a simple sentence “Migration boat” and associated words “Thanks” “Countermeasures” “Luck” appear together.
In C5, a simple sentence “Tama ni Aoi” and associative words “disqualification”, “value”, and “evaluation” appear together.

As a result, the following second associative word list is extracted.
-------------------------------------------------- -----------------------------
(Second associative word list)
“A cane that never falls”: “Injury”, “Remorse”, “Prevention”
“Migration ship”: “Thanks” “Countermeasures” “Luck”
“Tama ni 瑕”: “Disqualification” “Value” “Evaluation”
-------------------------------------------------- -----------------------------

  As described above, the corpus sentences in which the simple sentences are cited are extracted from a large number of corpus sentences, and the associative word storage unit 102 is distributed around the appearances of the simple sentences in the corpus sentences. Extract associative words.

  As another embodiment, the corpus sentence storage unit 101 and / or the associative word storage unit 102 are preferably divided into a plurality of categories. In this case, the first associative word extraction unit 112 and / or the second associative word extraction unit 122 is referred from the corpus sentence storage unit 101 and / or the associative word storage unit 102 according to the category of the input target sentence. Select the category you want. When the category (genre) of the target sentence (news article) is related to, for example, “disease”, it is preferable that only corpus sentences and / or simple sentences classified into the genre are extracted.

[Associative Word List Matching Unit 131]
The associative word list matching unit 131 searches the second associative word list that is most similar to the first associative word list from the group of second associative word lists for each simple sentence. Specifically, the following steps are executed. Further, a plurality of second associative word lists similar to the first associative word list may be ranked in order (likelihood order). The higher the similarity between the associative word lists, the higher the affinity between the target sentence and the simple sentence.

First associated word list X of the target sentence x
X: (W _x1 , W _x2 ,..., W _xm )
Group of second associative word list Y for each simple sentence C _{1 to} C _n
Y (C ₁ ): (W ₁₁ , W ₁₂ ,..., W _1m )
Y (C ₂ ): (W ₂₁ , W ₂₂ ,..., W _2m )
...
Y (C _n ): (W _n1 , W _n2 ,..., W _nm )
Here, m is preferably a value of about 3 to 5 in consideration of the reliability of likelihood.

For the similarity between the first associative word list X and each second associative word list Y (C _n ), refer to the knowledge source database in which the distance relationship between words is derived based on the statistical appearance frequency. It is also preferable to do. An example of the knowledge source database is Japanese WordNet (see, for example, Non-Patent Document 3).

First, for each associative word in the first associative word list in the target sentence X, the semantic distance of each associative word in each second associative word list in the plurality of simple sentences C _{1 to} C _n is searched, and the distance The associative word pair with the shortest is derived, and the average distance or cumulative distance of the semantic distance of the associative word pair is calculated.
The distance P between associative words of X and Y (C ₁ ) is calculated as follows.
(S1) For W _x1 , the following distance P _x1 is calculated for each W ₁₁ , W ₁₂ ,..., W _{1 m} .
P (W _x1 , W ₁₁ ), P (W _x1 , W ₁₂ ), ..., P (W _x1 , W _1m )
_Then, the _{W x1,} select the most distance P is short _W 11 _{to W-1 m,} and stores the distance _{P x1.}
(S2) For W _x2 , the following distance P is calculated for each W ₁₁ , W ₁₂ ,..., W _{1 m} .
P (W _x2 , W ₁₁ ), P (W _x2 , W ₁₂ ), ..., P (W _x2 , W _1m )
_Then, the _{W x2,} select the most distance P is short _W 11 _{to W-1 m,} and stores the distance _{P x2.}
...
(Sm) For W _xm , the following distance P is calculated for each W ₁₁ , W ₁₂ ,..., W _{1 m} .
P (W _xm , W ₁₁ ), P (W _xm , W ₁₂ ),..., P (W _xm , W _1m )
_Then, the _{W xn,} select the most distance P is short _W 11 _{to W-1 m,} and stores the distance _{P xn.}
Finally, the average distance _{P A} of the distances _P x1 _{to P xn,} the distance between X and Y _{(C 1).}
Next, an average distance between associative words is calculated for each of Y (C ₂ ) to Y (C _n ) with respect to X.

  Finally, the second associative word list Y (C) having the shortest average distance or cumulative distance (most similar among the associative words) is selected with respect to the first associative word list in the target sentence X. . Note that a predetermined number of Y (C) from the top may be derived as a ranking (in order of likelihood).

"A cane that doesn't fall"
<Target sentence: first associative word list><Simple sentence: second associative word list> Shortest distance
"Inspection""Remorse" 0.080
"Prevention""Prevention" 0
"Illness""Injury" 0.046
[Average distance]: 0.042
"Ship to the cross"
<Target sentence: first associative word list><Simple sentence: second associative word list> Shortest distance
"Inspection""Countermeasures" 0.120
"Prevention""Thanks" 0.925
"Illness""Luck" 0.620
[Average distance]: 0.555
"Tama ni candy"
<Target sentence: first associative word list><Simple sentence: second associative word list> Shortest distance
"Inspection""Evaluation" 0.825
"Prevention""Value" 0.988
"Illness""Disqualification" 0.905
[Average distance]: 0.906

  As another embodiment, instead of the average distance, the cumulative distance in the shortest distance of the top several cases may be used.

[Simple sentence estimation unit 132]
The simple sentence estimation unit 132 outputs a simple sentence based on the second associative word list searched by the associative word list matching unit 131 as a symbol of the target sentence. Moreover, the simple sentence estimation part 132 may select a predetermined number of simple sentences in order with a short distance with respect to an object sentence, and may output several simple sentences as an estimation result.

According to FIGS. 2 and 3, the simple sentence “the tip of the tip that does not fall” with the shortest (most similar) average distance is estimated and displayed.
-------------------------------------------------- -----------------------------
Title "A child's aunt dies of breast cancer" 2013/5/28 16:56
Concise expression: "The cane that never falls"
-------------------------------------------------- -----------------------------

  The user can determine the priority of reading the article by looking at the title and concise expression. The concise expression is displayed like a subtitle. When multiple simple sentences can be extracted as rankings, for example, when the user brings the cursor over the concise expression “the wand that does not fall”, the other concise expressions are also displayed in a list in order of decreasing average distance. It may be displayed.

  As another embodiment, it is also preferable to further indicate to the user a secondary feature such as an icon indicating the polarity (positive / negative) of the content of the retrieved concise expression (simple sentence). The user can intuitively understand the positive / negative sense in addition to the concise expression.

  FIG. 4 is a functional configuration diagram of a target sentence estimation device that searches for a target sentence by inputting a simple sentence.

  According to FIG. 4, as compared with FIG. 1, the target sentence storage unit 110 that stores a plurality of target sentences in advance is provided, but the simple sentence storage unit 121 is not provided. A list of simple sentences (concise expressions) may be clearly indicated to the user, and one simple sentence may be input by user selection. The other functional components execute basically the same processing as in FIG. However, compared with FIG. 1, the process of the following components differs in the point which becomes a structure of a some target sentence and one simple sentence.

The important word extraction unit 111 extracts a plurality of characteristic important words for each target sentence according to the appearance frequency of each word.
The first associative word extraction unit 112 refers to the corpus sentence storage unit 101 for each target sentence, and appears first in association with important words among the association words stored in the associative word storage unit 102. Extract a group of associative word lists.
The second associative word extraction unit 122 refers to the corpus sentence storage unit 101 for the input simple sentence, and appears in association with the simple sentence in the associative words stored in the associative word storage unit 102. A second associative word list is extracted.
The associative word list matching unit 131 searches the first associative word list that is most similar to the second associative word list from the group of first associative word lists for each target sentence.
The target sentence estimation unit 133 outputs the target sentence based on the first associative word list searched by the associative word list matching unit 131 as a symbol of the simple sentence.

  Finally, according to FIG. 4, for example, by inputting the simple sentence “the cane that does not fall”, the news article of the target sentence “A child's aunt, no breast cancer” is retrieved.

  As another embodiment, instead of displaying the target sentence first as the order of presentation to the user, a certain number of simple sentence lists are displayed, and the user is allowed to select an expression of interest from the list. Is also preferable. For example, a concise expression (simple sentence) corresponding to each news article is derived in advance for a new news group, and a list sorted in the appearance frequency is displayed. Thus, the user can browse the news article by selecting the simple sentence while confirming the trend that has been discussed in the event of the day.

  FIG. 5 is a system configuration diagram having a simple sentence estimation server in the present invention.

The simple sentence estimation server 1 receives a query from the terminal 2, and returns a “target sentence and simple sentence set” corresponding to the query to the terminal 2. According to FIG. 5, the simple sentence estimation server 1 includes only the target sentence database 140 and the target sentence search unit 141 as compared with FIG. 1.
The target sentence database 140 stores a large amount of target sentences.
The target sentence search unit 141 searches for one or more target sentences using the target sentence database for the input query. The retrieved target sentence is output to the keyword extraction unit 111.
Finally, the simple sentence estimation unit 132 returns a simple sentence to the terminal 2 in the following two patterns.
(1) A simple sentence is combined with each target sentence and output as a search result.
(2) First, only a plurality of simple sentences corresponding to the target sentence as a query search result are returned to the terminal 2. The user operating the terminal 2 is allowed to select any simple sentence. Then, the target sentence corresponding to the selected simple sentence is returned to the terminal 2 as a search result.

<First Embodiment>
According to such a system, it is possible to search for “a proverb corresponding to a query or a target sentence” as a user interface. For example, it is assumed that the user is trying to select an article that is likely to be interesting in a short time while viewing a list of articles such as news and blogs. At this time, it is possible to display a “say” as a subtitle juxtaposed with the article title.
-------------------------------------------------- -----------------------------
Title "A child's aunt, dies of breast cancer"-> [A cane that never falls]
-------------------------------------------------- -----------------------------
As a user, not only the article title but also a word (concise expression) that helps understanding the topic can be used as a clue to determine the priority of reading the article. As a user, when the concise expression matches well with the content, it is impressed and the article is stored.

<Second Embodiment>
Immediately after receiving the query from the user, the article list is not presented, but an “concise expression” list corresponding to the article is presented as an intermediate stage. The user can select any one concise expression from a plurality of concise expressions. Then, the article corresponding to the concise expression is presented again to the user. For example, a concise expression (a plurality of candidates is also possible) corresponding to each article is derived in advance for a new news group. By displaying the results sorted in the order of their frequency, it is possible to select articles while looking at the trend that has become a topic in the events of the day as a concise expression.

  Both of these embodiments, together with the text information polarity information (positive / negative element determination) according to the prior art, can help to grasp the outline of the article more intuitively.

  As described above in detail, the program, apparatus, and server of the present invention can estimate a concise expression (simple sentence) that symbolically represents a target sentence. Thereby, the user who touched the target sentence can intuitively understand the semantic content of the target sentence only by looking at the concise expression. 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 Simple sentence estimation apparatus, server 101 Corpus sentence memory | storage part 102 Associative word memory | storage part 111 Important word extraction part 112 1st associative word extraction part 121 Simple sentence memory | storage part 122 2nd associative word extraction part 131 Associative word list matching part 132 Simple sentence estimation unit 133 Target sentence estimation part 140 Target sentence database 141 Target sentence search part 2 Terminal

Claims (14)

A program that causes a computer to function to estimate a simple sentence that symbolizes a target sentence,
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
An important word extracting means for extracting a plurality of characteristic important words according to the frequency of appearance of each word in the input target sentence;
First associative word extraction means for referring to the corpus sentence storage means and extracting a first associative word list that co-occurs with the important word from the associative words stored in the associative word storage means When,
For each simple sentence stored in the simple sentence storage means, refer to the corpus sentence storage means, and appear in the association word stored in the associative word storage means together with the simple sentence. Second associative word extracting means for extracting a group of associative word lists;
An associative word list matching means for searching a second associative word list most similar to the first associative word list among the second associative word list group for each simple sentence;
A program for causing a computer to function as simple sentence estimation means for outputting a simple sentence based on a second associative word list searched by an associative word list matching means as a symbol of the target sentence. The program according to claim 1, wherein the computer functions so that the simple sentence is a quote, a saying, a saying, a haiku, and / or a kawayanagi. The associative word list matching means includes:
For each associated word in the first associated word list in the target sentence X, the semantic distance of each associated word in each second associated word list in the plurality of simple sentences C _{1 to} C _n is searched, and the distance is the shortest. The associative word pair is derived, and the average distance or cumulative distance of the semantic distance of the associative word pair is calculated.
The computer is made to function so as to select a simple sentence based on a second associative word list having the shortest average distance or cumulative distance with respect to the first associative word list in the target sentence X. 2. The program according to 2. The corpus sentence storage means and / or the associative word storage means are divided into a plurality of categories,
The first associative word extracting unit and / or the second associative word extracting unit selects a category to be referred to from the corpus sentence storage unit and / or the associative word storage unit according to the input category of the target sentence. The program according to any one of claims 1 to 3, wherein the computer functions as described above. The simple sentence estimation unit selects a predetermined number of simple sentences in order from a short distance to the target sentence, and causes the computer to function so as to output a plurality of simple sentences as estimation results. 5. The program according to any one of 4 above. The important word extraction means causes the computer to function to extract an important word using TF-IDF (Term Frequency-Inverse Document Frequency) or DF (Document Frequency). The program according to any one of claims 1 to 5, characterized in that: A target sentence database storing a large amount of target sentences;
For the input query, using the target sentence database, search for one or more target sentences, the target sentence search means for outputting the target sentence to the keyword extraction means,
The program according to any one of claims 1 to 6, wherein the simple sentence estimation means causes a computer to output a simple sentence corresponding to each target sentence as a search result. The simple sentence estimation means includes:
Output the search results by combining the simple sentences with each target sentence,
Or
A plurality of the simple sentences are output, and the computer is caused to function so that the user selects any one of the simple sentences and outputs the target sentence corresponding to the selected simple sentence as a search result again. The program according to claim 7. A program that causes a computer to function to estimate a target sentence symbolizing a simple sentence,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, reference is made to the corpus sentence storage means, and a group of first association word lists that appear together with the important word is extracted from the association words stored in the association word storage means. First associative word extraction means;
With reference to the corpus sentence storage means for the input simple sentence, a second associative word list appearing together with the simple sentence is extracted from the association words stored in the association word storage means A second associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to a second associative word list among a group of first associative word lists for each target sentence;
A program that causes a computer to function as target sentence estimation means for outputting a target sentence based on a first associative word list searched by an associative word list matching means as a symbol of the simple sentence. A device that estimates a simple sentence symbolizing a target sentence,
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
An important word extracting means for extracting a plurality of characteristic important words according to the frequency of appearance of each word in the target sentence;
First associative word extraction means for referring to the corpus sentence storage means and extracting a first associative word list that co-occurs with the important word from the associative words stored in the associative word storage means When,
For each simple sentence stored in the simple sentence storage means, refer to the corpus sentence storage means, and appear in the association word stored in the associative word storage means together with the simple sentence. Second associative word extracting means for extracting a group of associative word lists;
An associative word list matching means for searching a second associative word list most similar to the first associative word list among the second associative word list group for each simple sentence;
An apparatus comprising: simple sentence estimation means for outputting a simple sentence based on the second associative word list searched by the associative word list matching means as a symbol of the target sentence. A device for estimating a target sentence symbolizing a simple sentence,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, reference is made to the corpus sentence storage means, and a group of first association word lists that appear together with the important word is extracted from the association words stored in the association word storage means. First associative word extraction means;
With reference to the corpus sentence storage means for the input simple sentence, a second associative word list appearing together with the simple sentence is extracted from the association words stored in the association word storage means A second associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to a second associative word list among a group of first associative word lists for each target sentence;
An apparatus comprising: target sentence estimation means for outputting a target sentence based on the first associative word list searched by the associative word list matching means as a symbol of the simple sentence. A server that receives a query from a terminal, searches for one or more target sentences corresponding to the query, estimates a simple sentence that symbolizes the target sentence, and returns the simple sentence to the terminal;
A target sentence database storing a large amount of target sentences;
Simple sentence storage means for storing a plurality of simple sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
In response to the query, target sentence search means for searching for one or more target sentences using the target sentence database;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word in the searched target sentence;
First associative word extraction means for referring to the corpus sentence storage means and extracting a first associative word list that co-occurs with the important word from the associative words stored in the associative word storage means When,
For each simple sentence stored in the simple sentence storage means, refer to the corpus sentence storage means, and appear in the association word stored in the associative word storage means together with the simple sentence. Second associative word extracting means for extracting a group of associative word lists;
An associative word list matching means for searching a second associative word list most similar to the first associative word list among the second associative word list group for each simple sentence;
Simple sentence estimation means for outputting a simple sentence based on the second associative word list searched by the associative word list matching means as a search result symbolizing the target sentence;
The server characterized by having. The simple sentence estimation means includes:
Reply to the terminal as a search result by combining the simple sentence with each target sentence,
Or
The plurality of simple sentences are returned to the terminal, the simple sentences selected by the user are received, and the target sentences corresponding to the simple sentences are newly returned as search results. server. A server that receives a simple sentence from a terminal, estimates a target sentence that symbolizes the simple sentence, and returns the target sentence to the terminal,
A target sentence storage means for storing a plurality of target sentences in advance;
Corpus sentence storage means for storing corpus sentences;
Associative word storage means for storing a plurality of associative words;
Important word extraction means for extracting a plurality of characteristic important words according to the appearance frequency of each word for each target sentence;
For each target sentence, reference is made to the corpus sentence storage means, and a group of first association word lists that appear together with the important word is extracted from the association words stored in the association word storage means. First associative word extraction means;
With reference to the corpus sentence storage means for the input simple sentence, a second associative word list appearing together with the simple sentence is extracted from the association words stored in the association word storage means A second associative word extraction means;
An associative word list matching means for searching a first associative word list most similar to a second associative word list among a group of first associative word lists for each target sentence;
And a target sentence estimating means for outputting a target sentence based on the first associative word list searched by the associative word list matching means as a symbol of the simple sentence.

Images