JP5764052B2 - LINK GENERATION DEVICE, LINK GENERATION METHOD, AND LINK GENERATION PROGRAM - Google Patents

Description

  The present invention relates to a link generation device, a link generation method, and a link generation program.

  Conventionally, there is a technique for embedding a hyperlink for a keyword in a web page so as to be linked with a search result of the keyword by a search engine. Further, a technique for embedding a hyperlink for transferring a keyword to a web page describing the keyword (for example, a web page corresponding to a keyword in an online encyclopedia) is known.

  As an example of automatically embedding hyperlinks, Patent Document 1 discloses a link generation device that detects specific words in a document as keywords in accordance with predefined rules and embeds hyperlinks for those keywords. Has been.

JP 2008-77227 A

  However, in the method of Patent Document 1, hyperlinks are annoying the display of links to keywords that are common to readers. In addition, there is a problem that the uselessness of the user is deteriorated, for example, an unnecessary link induces an erroneous operation on the touch panel (when a hyperlink is adjacent, there is a high possibility that an error is caused by a tap with a finger). Conversely, for a newcomer who has just joined a specific community (such as a bulletin board), it is useful to add a hyperlink to a dictionary site or related site to an unknown word. However, even in this case, the link display for the already learned keyword is similarly obstructive.

  Therefore, the present invention has been made in view of the above problems, and an object of the present invention is to provide a link generation device, a link generation method, and a link generation program that can improve the user-friendliness.

  (1) The present invention has been made in view of the above circumstances, and one aspect of the present invention includes a keyword extraction unit that extracts a keyword from a sentence included in content, a user's degree of participation in a specific community, and each keyword An acquisition unit that acquires a degree of specialization for a specific community, and a link necessity for determining whether to insert a hyperlink for a keyword based on the degree of participation and the degree of specialization acquired by the acquisition unit And a determination unit.

(2) In the link generation device described above, according to one aspect of the present invention, the degree of participation in the specific community is a score determined based on a cumulative value of user participation time in the community. Features.

(3) In the link generation device described above, according to one aspect of the present invention, the degree of specialization to the specific community includes a keyword included in an article posted to the general community and a keyword included in an article posted to the specific community. Are compared, a high score is given to a keyword that appears biased to a specific community.

(4) In the link generation device described above, according to one aspect of the present invention, the link necessity determination unit is configured such that the participation degree is assigned to a specific community when the participation degree is equal to or less than a predetermined participation degree. When the degree of specialization is equal to or higher than a predetermined degree of specialization, a hyperlink to a specific Web content is inserted into the target keyword.

  (5) In the link generation device described above, one aspect of the present invention is characterized in that the link necessity determination unit determines whether or not a link needs to be inserted with reference to a link necessity storage unit.

(6) In the link generation device described above, according to one aspect of the present invention, the link necessity storage unit does not link the user identification information, the information indicating whether the keyword has been clicked, or the like. Information indicating whether or not a word is stored in association with each other.

  (7) In the link generation device described above, according to one aspect of the present invention, the keyword extraction unit extracts a word from a sentence included in content posted in a specific community, and the word extraction An adjacent word extraction unit that extracts adjacent words from the words extracted by the unit, a connected word generation unit that generates a keyword by connecting adjacent words extracted by the adjacent word extraction unit, It is characterized by providing.

  (8) In the link generation device described above, one aspect of the present invention is a rule storage in which a keyword, a rule for connecting the keyword, and a degree of specialization of the keyword to a specific community are stored in association with each other. The connected word generation unit generates a keyword without connecting or connecting two words based on a connection rule stored in the rule storage unit when the two keywords are adjacent to each other. When a bigram connection determination unit and three keywords are adjacent to each other, a keyword is generated based on a connection rule stored in the rule storage unit without connecting three words or connecting or connecting two words. A trigram determination connection unit.

  (9) In the link generation device described above, according to an aspect of the present invention, the trigram connection determination unit applies a connection rule for three words every three words when the number of adjacent words is four or more. Thus, a keyword in which four or more words are connected is extracted.

  (10) In the link generation device described above, one aspect of the present invention is a rule storage in which a keyword, a connection rule for the keyword, and a degree of specialization of the keyword to a specific community are stored in association with each other. The acquisition unit acquires the specialization degree by reading out the specialization degree of the keyword to a specific community from the rule storage unit.

  (11) In the link generation device described above, according to one aspect of the present invention, when two or more words are adjacent to each other, the degree of connection indicating the degree to which the adjacent words are connected is statistically calculated, and the calculation is performed. A rule generation unit is provided that generates a connection rule for connecting words based on the connection degree and stores the generated connection rule in the rule storage unit.

  (12) In the link generation device described above, according to one aspect of the present invention, the rule generation unit calculates a degree of connection when two words are connected and a case where the two words are not connected, and based on the calculated degree of connection, A bigram connection rule generation unit that generates a connection rule for connecting two words, and a degree of connection when three words are connected, when two adjacent words of three words are connected, and when none of the three words are connected And a trigram connection rule generation unit that generates a connection rule for connecting three words based on the calculated connection degree.

  (13) In the link generation device described above, according to one aspect of the present invention, the connectivity is based on an Akaike information amount standard.

  (14) In the link generation device described above, one aspect of the present invention is characterized in that the word extraction unit extracts the word after normalizing the sentence.

  (15) In the link generation device described above, one aspect of the present invention is the user storage unit in which the identification information of the user and the degree of participation of the user in the specific community are stored in association with each other. The acquisition unit acquires the participation degree by reading out the participation degree of the user in the specific community from the user storage unit.

  (16) One aspect of the present invention is a link generation method in which a keyword extraction unit extracts a keyword from a sentence included in content, and an acquisition unit determines a user's degree of participation in a specific community and each Whether the procedure for acquiring the degree of specialization of a keyword to a specific community and the link necessity determination unit insert a hyperlink for the keyword based on the degree of participation and the degree of specialization acquired by the acquisition unit And a procedure for determining whether or not.

  (17) According to one aspect of the present invention, a keyword extraction step for extracting a keyword from a sentence included in content, a degree of participation of a user in a specific community, and a degree of specialization of each keyword in the specific community are acquired by a computer And a link necessity determination step for determining whether to insert a hyperlink for a keyword based on the participation level and the specialization level acquired in the acquisition step. This is a link generation program.

  According to the present invention, user convenience can be improved.

It is a schematic block diagram of the link production | generation apparatus in this embodiment. It is a schematic block diagram of the control part in this embodiment. It is a flowchart which shows an example of the flow of the process by which a connection rule is constructed | assembled. This is a 2 × 2 contingency table for the combination {w _i , w _j }. This is a 2 × 4 contingency table for the combination {w _i , w _j , w _k }. It is an example of the connection rule memorize | stored in the rule memory | storage part. It is a flowchart which shows an example of the flow of a process which inserts a hyperlink in a web page.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. First, the outline | summary of the link production | generation apparatus 1 of this embodiment is demonstrated. The link generation device 1 according to the present embodiment removes and removes hyperlinks to keywords based on the user's community participation elapsed time, the reference frequency of documents distributed in the community, and the degree of community specialization of keywords. Switch.

FIG. 1 is a schematic block diagram of a link generation device 1 in the present embodiment. The link generation device 1 includes an input unit 10, a storage unit 20, a display unit 30, and a control unit 100. The storage unit 20 includes a first first sentence storage unit 21, a second sentence storage unit 22, a rule storage unit 23, a user storage unit 24, and a link necessity storage unit 25.
The first document storage unit 21 stores a general document set D1 in advance. The second sentence storage unit 22 stores a document set D2 of a specific community in advance.

The rule storage unit 23 stores a keyword, a concatenation rule for the keyword, and a degree of specialization of the keyword for a specific community in association with each other.
The user storage unit 24 stores user identification information and the degree of participation of the user in a specific community in association with each other. Here, the participation degree in the specific community is a score determined based on a cumulative value of the user's participation time in the community. The degree of specialization for a specific community is determined by comparing the keywords included in articles posted to the general community with the keywords included in articles posted to the specific community. A high score is given.
The link necessity storage unit 25 stores, for each user, a keyword, information indicating whether or not the user has clicked, and information indicating whether or not the keyword is not linked.

The input unit 10 receives text information indicating text to be added to the text set D <b> 1 stored in the first text storage unit 21, and outputs the received text information to the control unit 100.
In addition, the input unit 10 receives an input of a web page and outputs the received web page to the control unit 100.

The control unit 100 adds the sentence information input from the input unit 10 to the sentence set D <b> 1 stored in the first sentence storage unit 21.
In addition, the control unit 100 reads the sentence set D1 stored in the first sentence storage unit 21, and generates a word connection rule based on the read sentence set D1. Then, the control unit 100 stores the generated word connection rule in the rule storage unit 23. As a result, the rule storage unit 23 stores in advance a connection rule for the words when two or more words are adjacent to each other.

  In addition, the control unit 100 reads the connection rule stored in the rule storage unit 23, and based on the read connection rule, the control unit 100 selects an important sentence included in the sentence from the sentence included in the web page input from the input unit 10. A linking word is generated, and a link is embedded in the generated linking word. Then, the control unit 100 causes the display unit 30 to display the web page in which the link is embedded. As a result, the link generation device 1 displays a web page with a link to the dictionary site for the keyword included in the web page simply by the user inputting the web page. You can know the meaning of the keyword just by opening it.

  FIG. 2 is a schematic block diagram of the control unit 100 in the present embodiment. The control unit 100 includes a keyword extraction unit 104, a rule generation unit 110, an acquisition unit 135, a link necessity determination unit 136, and a link insertion unit 140. Here, the keyword extraction unit 104 includes a word extraction unit 101, an important word extraction unit 102, an adjacent word extraction unit 103, and a connected word generation unit 130.

Next, an outline of processing of each unit will be described.
The keyword extraction unit 104 extracts keywords from sentences included in the content.
When the link generation device 1 generates a connection rule, the word extraction unit 101 reads the sentence set D from the first sentence storage unit 21 and extracts words from sentences included in the sentence set D. Then, the word extraction unit 101 outputs the extracted word to the adjacent word extraction unit 103.
When the link generation device 1 embeds a hyperlink in a keyword included in a web page, the word extraction unit 101 extracts a word from a sentence included in content posted in a specific community, and the extracted word is an important word. The data is output to the extraction unit 102.
Note that the word extraction unit 101 may extract words after normalizing a sentence.

  The keyword extraction unit 102 extracts keywords that are candidates for embedding hyperlinks from the words extracted by the word extraction unit 101, and outputs the extracted keywords to the adjacent word extraction unit 103.

When the link generation device 1 generates a connection rule, the adjacent word extraction unit 103 extracts adjacent words from the words extracted by the word extraction unit 101, and outputs the extracted words to the rule generation unit 110.
The link generation device 1 embeds a hyperlink in a keyword included in a web page, and the adjacent word extraction unit 103 extracts adjacent words from the keywords extracted by the keyword extraction unit 102, and the extracted words are connected words. The data is output to the generation unit 130.

The rule generation unit 110 statistically calculates the degree of connection indicating the degree to which adjacent words are connected to each other, and generates a connection rule for connecting words based on the calculated degree of connection. Here, the degree of connectivity is, for example, a known Akaike information amount standard that is an index for evaluating the goodness of a statistical model. The concatenation rule associates, for example, a condition for concatenating keywords with processing performed by the concatenated word generation unit 130 when the condition is met.
Then, the rule generation unit 110 stores the generated connection rule in the rule storage unit 23. Here, the rule generation unit 110 includes a bigram connection rule generation unit 111 and a trigram connection rule generation unit 112.

The bigram connection rule generation unit 111 calculates a connection degree between two words, and generates a connection rule for connecting two words based on the calculated connection degree.
The trigram connection rule generation unit 112 calculates the connection degree of the three words, and generates a connection rule that connects the three words based on the calculated connection degree.

Based on the linking rules stored in the rule storage unit 23, the linking word generation unit 130 generates a linking word by linking keywords, and outputs the generated linking word to the link insertion unit 140. Here, the connective word generation unit 130 includes a bigram connection determination unit 131 and a trigram connection determination unit 132.
When two keywords are adjacent to each other, the bigram connection determination unit 131 determines whether or not to connect the keywords based on the connection rules stored in the rule storage unit 23.

When three keywords are adjacent to each other, the trigram connection determination unit 132 determines whether to connect three keywords out of the three adjacent keywords based on the connection rule stored in the rule storage unit 23. Decide whether to concatenate two keywords or none.
In addition, when there are four or more keywords adjacent to each other, the trigram connection determination unit 132 extracts the three adjacent keywords from the four or more adjacent keywords. Decide whether to link three keywords, two keywords, or neither.

  The acquisition unit 135 acquires the degree of participation of the user in the specific community and the degree of specialization of each keyword in the specific community. Specifically, for example, the acquisition unit 135 reads from the user storage unit 24 the degree of participation of the user in the specific community corresponding to the user identification information. For example, the acquisition unit 135 reads the degree of specialization of each keyword with respect to a specific community from the rule storage unit 23. Then, the acquisition unit 135 outputs the acquired degree of participation of the user in the specific community and the degree of specialization of each keyword to the specific community to the link necessity determination unit 136.

  The link necessity determination unit 136 determines whether or not to insert a hyperlink to another page for the keyword based on the degree of participation and the degree of specialization acquired by the acquisition unit 135. Specifically, for example, the link necessity determination unit 136, when the participation degree is equal to or less than a predetermined participation degree, the degree of specialization of the target keyword to the specific community is equal to or higher than the predetermined specialization degree. In this case, it is determined to insert a hyperlink to a specific Web content for the target keyword. Here, the target keyword is a generic name of a connected word in which keywords are connected or a keyword that is not connected.

Further, the link necessity determination unit 136 refers to the link necessity storage unit 25 to determine whether or not link insertion is necessary. Specifically, for example, the link necessity determination unit 136 reads, from the link necessity storage unit 25, information indicating whether or not the click corresponding to the target keyword has been completed and information indicating whether or not the keyword is not linked. . The link necessity determination unit 136 determines that the link is not inserted for the target keyword when the information indicating whether or not the keyword does not link is a keyword that does not link.
For example, when the information indicating whether or not the click has been read indicates that the click has been performed, the link necessity determination unit 136 determines that a link is not inserted for the target keyword for a predetermined time.
Then, the link necessity determination unit 136 outputs the determination result to the link insertion unit 140.

  The link insertion unit 140 inserts a hyperlink to a specific Web content with respect to the keyword extracted by the concatenated word generated by the concatenated word generation unit 130 or the adjacent word extraction unit, and displays the web page after the insertion as the display unit 30. To display.

  Next, an outline of processing of the link generation device 1 will be described. The link generation device 1 is. A keyword that is a candidate for a hyperlink is extracted, and then a hyperlink is created for the word. At this time, as a keyword extraction method, a simple method that uses proper nouns or nouns included in the output results of the morphological analyzer as keywords does not guarantee that keywords can be obtained in an optimal form for embedding hyperlinks. There is.

  For example, if a document distributed in a community has a community-specific keyword “connectivity calculation function”, and if this document is analyzed by morpheme, the morpheme such as “concatenation” “degree” “calculation” “function” It will be extracted. Therefore, it is necessary to recognize that they are one keyword that can be combined. One is a method of manually registering such keywords in the morphological analysis dictionary in advance, but it requires management costs such as the need to always register new keywords. In view of this, the link generation device 1 according to the present embodiment uses a large number of past cases to automatically generate combined keywords by a method based on a statistical criterion using the appearance frequency of the combined keywords. Is adopted.

FIG. 3 is a flowchart illustrating an example of a flow of processing for constructing a connection rule.
First, in step S1, the word extraction unit 101 extracts words from sentences included in the sentence set D. The document set D includes a general document set (for example, a document set that can be viewed by anyone such as general news) D1 and a document set of a specific community (for example, a document set belonging to a community in which a specific user participates) D2. Including.

  Specifically, for example, the word extraction unit 101 extracts words from the document set D by using a morphological analyzer. In addition, the word extraction unit 101 performs normalization of the sentence (for example, replacing lowercase English letters with uppercase letters or correcting notation fluctuations) in order to improve the word extraction accuracy of the morphological analyzer, You may extract a word from the text after normalization. Moreover, the word extraction part 101 may normalize the extracted word, after extracting a word.

  Next, in step S2, the adjacent word extraction unit 103 extracts adjacent words from the words extracted by the word extraction unit 101. At this time, the adjacent word extraction unit 103 may extract adjacent words after narrowing down to words corresponding to specific parts of speech such as general nouns and proper nouns, in order to reduce the calculation processing time. Thereby, the adjacent word extraction part 103 acquires word set w = {W1, W2, W3, ...}.

  Next, in step S2, the rule generation unit 110 statistically calculates the degree of connection in order to connect words. Specifically, for example, the rule generation unit 110 calculates the connectivity based on the Akaike information criterion (AIC). Note that the rule generation unit 110 may calculate the connectivity based on a scale such as another information criterion (for example, Bayesian information criterion (BIC)).

Specifically, for example, the rule generation unit 110 extracts two arbitrary words w _i and w _j from the words W = {w1, w2, w3,. Next, the rule generation unit 110 extracts the first and second words from the beginning of the document for each sentence in the document set D. When the adjacent word extraction unit 103 narrows down the parts of speech, the rule generation unit 110 extracts these words when these words are part of speech to be narrowed down. The rule generation unit 110 sets these words as t ₁ and t ₂ . A word concatenation rule is generated using the words w _i and w _j and the words t ₁ and t ₂ .

Here, generation of a two-word connection rule that is a rule for connecting two words by the bigram connection rule generation unit 111 will be described. The bigram connection rule generation unit 111 calculates the following four variables (n ₁₁ , n ₁₂ , n ₂₁ , n ₂₂ ) using t ₁ , t ₂ , w _i, and w _j .
The variable n ₁₁ is the number of combinations where t ₁ is w _i and t ₂ is w _j .
The variable n ₁₂ is the number of combinations where t ₁ is w _i and t ₂ is a word other than w _j .
The variable n ₂₁ is the number of combinations where t ₁ is a word other than w _i and t ₂ is w _j .
The variable n ₂₂ is the number of combinations where t ₁ is a word other than w _i and t ₂ is a word other than w _j .
The variable z is the number of all combinations (n ₁₁ + n ₁₂ + n ₂₁ + n ₂₂ ).

The 2 × 4 contingency table shown in FIG. 4 is created using these values. FIG. 4 is a 2 × 2 contingency table for the combination {w _i , w _j }. Here, the symbol ¬ is a symbol indicating the non-variable following the next, for example, t _{1 =} ¬w _i indicates that t ₁ is other than w _i. In the figure, it is shown that a variable (n ₁₁ , n ₁₂ , n _21, or n ₂₂ ) indicating the number of combinations is determined for each combination of the values of t ₁ and t ₂ described above. For example, the variable n ₁₁ in the first row and first column indicates the number of combinations where t ₁ is w _i and t ₂ is w _j .

The bigram connection rule generation unit 111 generates the following two information amount standards according to the following calculation formula using various values in the 2 × 2 contingency table in FIG. 4.
First, the first information criterion is an information criterion (hereinafter referred to as a dependent information criterion) AIC (DM) of a dependent model (DM). The dependent information criterion AIC (DM) indicates the degree to which the combination of the word w ₁ and the word w ₂ is related to a specific topic x. However, the degree of the dependency information criterion AIC (DM) is more related to a specific topic x as the value is smaller. The bigram connection rule generation unit 111 calculates the subordinate information criterion AIC (DM) according to the following equation (1).

MLL (DM) = n ₁₁ log n ₁₁ + n ₁₂ log n ₁₂ + n ₂₁ log n ₂₁ + n ₂₂ log n ₂₂ −Zlog Z
AIC (DM) = − 2 × MLL (M2) + 2 × 3 (1)

Here, the dependency model (DM) is a model in which the word w _i and the word w _j are connected. In this case, the word w _i and the word w _j are in a dependency relationship.
Next, the second information criterion is the information criterion (IM) of the independent model (hereinafter referred to as the independent information criterion) AIC (IM) AIC (IM) The word w ₁ , the word w ₂ , and the combination of the word w ₁ and the word w ₂ all indicate the degree of irrelevance to the specific topic x. However, the smaller the value of this independent information criterion AIC (IM), the greater the degree of irrelevance to a specific topic x. The bigram connection rule generation unit 111 calculates the independent information criterion AIC (IM) according to the following equation (2).

_{MLL (IM) = (n 11} + n 12) log (n 11 + n 12) + (n 11 + n 21) log (n 11 + n 21) + (n 21 + n 22) log (n 21 + n 22) + (n 12 + N ₂₂ ) log (n ₁₂ + n ₂₂ ) −2 × Zlog Z
AIC (IM) = − 2 × MLL (M1) + 2 × 2 (2)

Here, the independent model (IM) is a model in which the word w _i and the word w _j are not connected. In this case, the word w _i and the word w _j are not dependent. Here, the dependent information criterion AIC (DM) and the independent information criterion AIC (IM) are collectively referred to as the connectivity of two words.
It should be noted that “... + 2 × 3” in the equation (1) of the dependent information criterion AIC (DM) and “... + 2 × 2” in the equation (2) of the independent information criterion AIC (IM) are respectively “... + 2 × log (3) ”and“... + 2 × log (2) ”, the subordinate model information criterion BIC (IM) and the independent model information criterion BIC ( DM) is obtained.

The bigram concatenation rule generation unit 111 calculates two values of the dependent information criterion AIC (DM) and the independent information criterion AIC (IM), and then includes a concatenation rule including two elements, a condition and a processing, by the following processing. (Hereinafter also referred to as a rule).
The bigram connection rule generation unit 111 generates the following rule when AIC (DM) <AIC (IM) holds, that is, when the dependent information criterion AIC (DM) is smaller than the independent dependent information criterion AIC (IM). The rule is that when t1 is w _i and t2 is w _j , the word w _i and the word w _j are connected. Here, the condition is “when t1 is w _i and t2 is w _j ”, and the process is “concatenate the word w _i and the word w _j ”.

  After completing the process for generating the word connection rule, the bigram connection rule generation unit 111 extracts the second and third words from the beginning of the document and performs the same process. Thereafter, the bigram connection rule generation unit 111 repeats the word until it reaches the end of the sentence, even by one word at a time. When the bigram connection rule generation unit 111 reaches the end of the sentence, it reads the next document in the document set D and performs the same processing. These processes are repeated until the entire sentence in the document set D is processed.

  The above process is a process for generating a two-word connection rule when two words are given, but the trigram connection rule generation unit 112 receives three words when three other adjacent words are given. Generate a concatenation rule. Specifically, it is as follows.

The trigram connection rule generation unit 112 extracts arbitrary three words w _i , w _j, and w _k from the words W = {w ₁ , w ₂ , w ₃ ,. Next, the trigram connection rule generation unit 112 extracts the first, second, and third words from the top of the document for each sentence in the document set D. When the trigram connection rule generation unit 112 narrows down the part of speech using the adjacent word extraction function, if all three words are adjacent and are part of speech to be narrowed down, those words are used. To extract. Let these words be t ₁ , t ₂ and t ₃ . The trigram connection rule generation unit 112 calculates the following eight variables using the words t ₁ , t _2, and t ₃ and the words w _i , w _j, and w _k .

The variable n ₁₁ is the number of combinations where t ₁ is w _i , t ₂ is w _j , and t ₃ is w _k .
The variable n ₁₂ is the number of combinations where t ₁ is w _i , t ₂ is a word other than w _j , and t ₃ is w _k .
The variable n ₁₃ is the number of combinations where t ₁ is w _i , t ₂ is w _j , and t ₃ is a word other than w _k .
The variable n ₁₄ is the number of combinations where t ₁ is w _i , t ₂ is a word other than w _j , and t ₃ is a word other than w _k .
The variable n ₂₁ is the number of combinations where t ₁ is a word other than w _i , t ₂ is w _j , and t ₃ is w _k . n ₂₂ is the number of combinations where t ₁ is a word other than w _i , t ₂ is a word other than w _j , and t ₃ is w _k .
The variable n ₂₃ is the number of combinations where t ₁ is a word other than w _i , t ₂ is w _j , and t ₃ is a word other than w _k .
The variable n ₂₄ is the number of combinations where t ₁ is a word other than w _i , t ₂ is a word other than w _j , and t ₃ is a word other than w _k .
The variable z is the number of all combinations: (n ₁₁ + n ₁₂ + n ₁₃ + n ₁₄ + n ₂₁ + n ₂₂ + n ₂₃ + n ₂₄ ).

The 2 × 4 contingency table shown in FIG. 5 is created using these values. FIG. 5 is a 2 × 4 contingency table for the combination {w _i , w _j , w _k }. Here, the symbol ∧ means “and”. For example, (t ₂ = w _j ) ∧ (t ₃ = w _k ) indicates that t ₂ is w _j and t ₃ = w _k. Show. In the 2 × 4 contingency table of the same figure, variables (n ₁₁ , n ₁₂ , n ₁₃ , n ₁₄ , n ₂₁ , n ₂₂₎ indicating the number of combinations for each combination of the values of t ₁ , t _2, and t ₃ described above. , N ₂₃ or n ₂₄ ) is determined. For example, the variable n ₁₁ is the number of combinations when t ₁ is w _i , t ₂ = w _j and t ₃ = w _k .

The trigram connection rule generation unit 112 generates four models according to the following calculation formula using various values in the 2 × 4 contingency table. First, the trigram connection rule generation unit 112 calculates an information amount criterion (hereinafter referred to as a first information amount criterion) AIC (M1) of the first model M1 according to the following equation. Here, the first model M1 is a model in which the word w _i and the word w _j are connected, but the word w _j and the word w _k are not connected. In this case, the word w _i and the word w _j are in a dependency relationship.

_{MLL (M1) = (n 11} + n 12) log (n 11 + n 12) + (n 13 + n 14) log (n 13 + n 14) + (n 21 + n 22) log (n 21 + n 22) + (n 23 + N ₂₄ ) log (n ₂₃ + n ₂₄ ) + (n ₁₁ + n ₁₃ + n ₂₁ + n ₂₃ ) log (n ₁₁ + n ₁₃ + n ₂₁ + n ₂₃ ) + (n ₁₂ + n ₁₄ + n ₂₂ + n ₂₄ ) log (n ₁₂ + n _{14 21} + n ₂₃ ) -2 × ZlogZ
AIC (M1) = − 2 × MLL (M1) + 2 × 4 (3)

Next, the trigram connection rule generation unit 112 calculates an information amount criterion (hereinafter referred to as a second information amount criterion) AIC (M2) of the second model M2 according to the following equation. Here, the second model M2 is a model in which the word w _j and the word w _k are connected, but the word w _i and the word w _j are not connected. In this case, the word w _j and the word w _k are dependent.

_{MLL (M2) = (n 11} + n 12 + n 13 + n 14) log (n 11 + n 12 + n 13 + n 14) + (n 21 + n 22 + n 23 + n 24) log (n 21 + n 22 + n 23 + n 24) + ( _{n 11 + n 12) log (} n 11 + n 12) + (n 12 + n 22) log (n 12 + n 22) + (n 13 + n 23) log (n 13 + n 13) + (n 14 + n 24) log (n ₁₄ + n ₂₄ ) -2 × Zlog Z
AIC (M2) = − 2 × MLL (M2) + 2 × 4 (4)

Next, the trigram connection rule generation unit 112 calculates an information amount criterion (hereinafter referred to as a third information amount criterion) AIC (M3) of the third model M3 according to the following equation. Here, the third model M3 is a model in which the word w _i , the word w _j, and the word w _k are connected. In this case, the word w _i , the word w _j, and the word w _k are in a dependency relationship.

_{MLL (M3) = n 11 logn} 11 + n 12 logn 12 + n 13 logn 13 + n 14 logn 14 + n 21 logn 21 + n 22 logn 22 + n 23 log n 23 + n 24 logn 24 -ZlogZ
AIC (M3) = − 2 × MLL (M3) + 2 × 7 (5)

Next, the trigram connection rule generation unit 112 calculates an information amount criterion (hereinafter referred to as a fourth information amount criterion) AIC (M0) of the 0th model M0 according to the following equation. Here, the fourth model M0 is a model in which the word w _i , the word w _j, and the word w _k are not connected. In this case, the word w _i , the word w _j, and the word w _k are all in independent events.

MLL (M0) = (n ₁₁ + n ₁₂ + n ₁₃ + n ₁₄ ) log (n ₁₁ + n ₁₂ + n ₁₃ + n ₁₄ ) + (n ₁₁ + n ₁₂ + n ₂₁ + n ₂₂ ) log (n ₁₁ + n ₁₂ + n ₂₁ + n ₂₂ ) _{n 11 + n 13 + n 21} + n 23) log (n 11 + n 13 + n 21 + n 23) + (n 21 + n 22 + n 23 + n 24) log (n 21 + n 22 + n 23 + n 24) + (n 13 + n 14 + n 23 + N ₂₄ ) log (n ₁₃ + n ₁₄ + n ₂₃ + n ₂₄ ) + (n ₁₂ + n ₁₄ + n ₂₂ + n ₂₄ ) log (n ₁₂ + n ₁₄ + n ₂₂ + n ₂₄ ) −3 × ZlogZ
AIC (M0) = − 2 × MLL (M0) + 2 × 3 (6)

Next, the trigram connection rule generation unit 112 acquires a rule by the following processing. The rule consists of two parts: condition and processing. The trigram connection rule generation unit 112 forms the following rule when the first information amount criterion AIC (M1) is the smallest among the four information amount criteria. The rule is to perform a process of “linking word w _i and word w _j ” under the condition “t1 is w _i , t2 is w _j , and t3 is w _k ”. is there.

Next, the trigram connection rule generation unit 112 forms the following rule when the second information amount criterion AIC (M2) is the smallest among the four information amount criteria. The rule is to perform a process of “linking word w _j and word w _k ” under the condition “t1 is w _i , t2 is w _j , and t3 is w _k ”. is there.

Next, the trigram connection rule generation unit 112 forms the following rule when the third information amount criterion AIC (M3) is the smallest among the four information amount criteria. The rule is a process of “concatenating word w _i , word w _j, and word w _k ” under the condition “t1 is w _i , t2 is w _j , and t3 is w _k ”. Is to do.

  Next, when the 0th information criterion AIC (M0) is the smallest among the four information criterions, the trigram connection rule generator 112 ends without generating a rule. Here, the first information criterion AIC (M1), the second information criterion AIC (M2), the third information criterion AIC (M3), and the zero information criterion AIC (M0) are collectively referred to. This is called the connectivity of three words.

  After completing the above processing, the trigram connection rule generation unit 112 extracts the second, third, and fourth words from the beginning of the document, and performs the same processing. Thereafter, the trigram connection rule generation unit 112 repeats this process until the end of the sentence is reached while even one word at a time. When the trigram connection rule generation unit 112 reaches the end of the sentence, the trigram connection rule generation unit 112 next reads the document and performs the same processing as described above. The trigram connection rule generation unit 112 repeats these processes until the entire sentence in the document set D is processed.

  In step S4, the rule generation unit 110 causes the rule storage unit 23 to store the rule obtained in step S3. Specifically, for example, the rule generation unit 110 assigns an ID to each rule obtained in step S3, and then stores the ID, the condition, and the process in association with each other in the rule storage unit 23.

  In step S <b> 5, the rule generation unit 110 gives a flag to the word combination rule that appears specifically for the specific community. Specifically, the rule generation unit 110 compares the word combination rule obtained from the document set D2 of a specific community (for example, community A) with the word combination rule obtained from the general document set D1 (general). To do. Then, the rule generation unit 110 extracts word combination rules that appear only in the document set D2 (community A) of a specific community. And the rule production | generation part 110 sets the flag of specific community specialization (for example, community A specialization) in the extracted word coupling rule in the rule memory | storage part 23. FIG. Accordingly, the rule generation unit 110 can store the keyword, the rule for connecting the keyword, and the degree of specialization of the keyword with respect to the specific community in the rule storage unit 23 in association with each other. Above, the process of this flowchart is complete | finished.

FIG. 6 is an example of a connection rule stored in the rule storage unit 23. In the figure, a set of ID, condition, and processing is shown. For example, when the ID is 1, the condition is “t ₁ is w _i , t ₂ is w _j , t ₃ is w _k ”, and the process is “word w _i and word w _j It is shown that it is “connected”.

FIG. 7 is a flowchart showing an example of the flow of processing for inserting a hyperlink into a web page. This figure shows the flow of processing after the connection rule is constructed in the rule storage unit 23 in FIG.
First, in step S6, the word extraction unit 101 extracts words from the input web page. Here, it is assumed that no hyperlink is inserted in the word on the web page. The word extraction unit 101 extracts words using, for example, a morphological analyzer. At this time, in order to improve the word extraction accuracy of the morphological analyzer, the word extraction unit 101 normalizes the sentence (for example, replaces lowercase English letters with uppercase letters or corrects notation fluctuations) before extraction. ) May be performed.

  Next, in step S7, the keyword extraction unit 102 extracts keywords that are candidates for embedding hyperlinks from the document. Specifically, for example, the keyword extraction unit 102 selects only words corresponding to a specific part of speech (for example, general nouns or proper nouns). The important word extraction unit 102 may extract an important word using TF-IDF (Term Frequency-Inverse Document Frequency) or the like.

  Next, in step S8, the adjacent word extraction unit 103 extracts keywords that are adjacent to each other from the keywords extracted by the keyword extraction unit 102, and sets these as candidates for combination.

  Next, when there are two adjacent words (bi-gram) or three words (tri-gram) among the adjacent keywords, the connected word generation unit 130 determines whether to combine these words. Determine. Then, when it is determined that the words are combined, the connected word generation unit 130 connects the words to generate a connected word. Thereby, the link insertion part 140 can insert a hyperlink with respect to an appropriate word. Specifically, the connected word generation unit 130 connects words according to the following procedure.

In step S9-1, the process of the bigram connection determination unit 131 will be described using a case where t ₁ is w _i and t ₂ is w _j as an example. At this time, the bigram connection determination unit 131 refers to the rules regarding t ₁ and t ₂ stored in the rule storage unit 23. At that time, the rule that the condition is “t ₁ is w _i and t ₂ is w _j ” and the process is “concatenate the word w _i and the word w _j ” is stored in the rule storage unit 23. In this case, the bigram connection determination unit 131 connects the word w _i and the word w _j .

The processing of the trigram connection determination unit 132 will be described by taking as an example a case where t ₁ is w _i , t ₂ is w _j and t ₃ is w _k in step S9-2. The trigram connection determination unit 132 refers to the rules regarding t ₁ , t _2, and t ₃ stored in the rule storage unit 23. At that time, the condition is “t ₁ is w _i , t ₂ is w _j , and t ₃ is w _k ”, and the process is “concatenate word w _i , word w _j and word w _k ”. Is stored in the rule storage unit 23, the trigram connection determination unit 132 connects the word w _i , the word w _j, and the word w _k .

  In addition, when four or more words are continuous, the trigram connection determination unit 132 applies the rule of three words (tri-gram) to the three words from the beginning among the consecutive four or more words. . Then, the trigram connection determination unit 132 successively applies the rules of three words (tri-gram) while shifting backward one word at a time. That is, when the number of adjacent words is four or more, the trigram connection determination unit 132 extracts a keyword in which four or more words are connected by applying a connection rule for three words every three words. Thereby, the trigram connection determination unit 132 determines whether to connect the words each time the rule of three words (tri-gram) is applied. When the trigram connection determination unit 132 determines to connect the words, the trigram connection determination unit 132 connects the words.

  Next, in step S10, the link necessity determination unit 136 determines whether to insert a hyperlink for the target keyword. When the target keyword is generated by applying a word combination rule specialized for the community, the link necessity determination unit 136 determines to insert a link into the target keyword only when a newcomer of the community is browsing. . On the other hand, when the target keyword is generated by applying a word combination rule that is not specific to the community, the link necessity determination unit 136 determines that a link is inserted into the target keyword for all members of the community.

  Here, the link necessity determination unit 136 determines whether or not a new community is entered based on the activity history in the community. More specifically, the link necessity determination unit 136 determines whether or not it is a new entry of the community based on the participation duration obtained from (viewing, utterance (posting), etc.) and / or the frequency of occurrence thereof. Specifically, for example, the link necessity determination unit 136 determines that the participation duration is equal to or less than a predetermined threshold time, and determines that the community is new, and otherwise determines that the community is not new. To do.

Next, processing for a link to a keyword once displayed to the user will be described. When the link to the keyword is clicked once, the link necessity determination unit 136 determines that the user has temporarily stored, and does not set the link thereafter. However, the link necessity determination unit 136 presents the link again after a predetermined time (for example, n days) or more, and if the link is not clicked, the keyword is regarded as a word that does not link, and the link necessity storage unit 25. Remember me. In this case, the link generation device 1 assumes that the new user understands the keyword.
Accordingly, the link necessity storage unit 25 holds, as a set of data (one record), for each user, a keyword, information indicating whether or not the user has clicked, and information indicating whether or not the keyword is not linked. .

Next, when the link necessity determination unit 136 determines in step S11 that a hyperlink is to be inserted into the target keyword (YES in step S10), the link insertion unit 140 inserts a hyperlink into the target keyword. To do.
Specifically, for example, the link insertion unit 140 inserts a hyperlink to which a search engine query is assigned. More specifically, the link insertion unit 140 holds a query template for a search engine in advance, incorporates a character string indicating a query of a linked word or a keyword that is not linked into the query template for a search engine, and Generate a link.

  For example, the link insertion unit 140 adds a character string such as “/ search? Q =” <search target> ”after the site URL of a commercial search engine. Here, it is assumed that the URL of a commercial search engine service (eg, service name is examplesearch) is http://examplesearch.com.

  When the target keyword is “National Diet Building”, the link insertion unit 140 generates a hyperlink “http://examplesearch.com/search?q=National Diet Building”. And the link insertion part 140 embeds the produced | generated hyperlink in the keyword in a web page. For example, in the case of HTML (hypertext markup language), the link insertion unit 140 surrounds a word with a tag <a href> </a>, so that “today, in the National Diet Building” href = ”http://examplesearch.com/search?q=In the Parliament”> In the Parliament </a> The link insertion unit 140 causes the display unit 30 to display the web page in which the hyperlink is embedded.

When the link insertion unit 140 stores a word and a hyperlink to a web page indicating the meaning of the word in advance, the hyperlink corresponding to the input connected word or the keyword that is not connected. May be read out. Then, the link insertion unit 140 may embed the read hyperlink at the position of the word on the web page.
When the link necessity determination unit 136 determines not to insert a hyperlink for the target keyword (NO in step S10), the link generation device 1 causes the display unit 30 to display a web page.
Above, the process of this flowchart is complete | finished.

<Effect>
As described above, the link generation device 1 according to the present embodiment determines whether or not to insert a link for a keyword based on an activity history in the community (for example, an elapsed time of participation in the community). As a result, the link generation device 1 can prevent a veteran user who has been in the community for a long time from joining the community from linking to a keyword specific to the community. Can be prevented. As a result, the link generation device 1 can improve user convenience.

  In addition, the link generation device 1 can link a keyword specific to a community to a new user who has a short elapsed time in participation in the community, thereby improving the convenience of the new user. Can be improved. As a result, the link generation device 1 can improve user convenience.

  Moreover, the link production | generation apparatus 1 judges that the user memorize | stored temporarily, when the link inserted in the keyword once was clicked with respect to the user, and does not set a link after that. Then, the link generation device 1 presents the link again after a predetermined time, and stores the link in the link necessity storage unit 25 as a keyword that does not link if the link is not clicked. In that case, since the link generation device 1 considers that the user has understood the keyword and removes the keyword hyperlink, the user's erroneous operation can be prevented. As a result, the link generation device 1 can improve user convenience.

  Further, the link generation device 1 of the present embodiment generates a word connection rule from the sentence set D in advance, and stores the generated connection rule in the rule storage unit 23. And the link production | generation apparatus 1 extracts a word from the input web page, and extracts a keyword among the extracted words. When the extracted keywords are adjacent to each other, the link generation device 1 refers to a connection rule stored in advance in the rule storage unit 23 and determines whether or not to connect these keywords.

When the link generation device 1 determines that the keywords are to be connected, the link generation device 1 generates a connected word obtained by connecting the keywords as a keyword. And the link production | generation apparatus 1 embeds the hyperlink to the web page explaining the meaning of the keyword with respect to the produced | generated keyword.
Thereby, the user can know the meaning of the word by opening the hiberlink embedded in the keyword included in the web page. Therefore, the link generation device 1 can reduce the time and labor required to know the meaning of the words included in the web page.

  In addition, since the connection rule is determined based on the degree of connection, the link generation device 1 can generate a keyword suitable for embedding a hyperlink. Here, keywords suitable for embedding hyperlinks are, for example, keywords that are often seen by users of the Internet, and keywords that are highly likely to be registered in a dictionary site such as Wikipedia (registered trademark).

  When the user periodically inputs the latest newspaper article or the latest article on the Internet to the link generation apparatus 1, the link generation apparatus 1 stores those articles in the document set D <b> 1 stored in the first sentence storage unit 21. Is added. Thereby, the link production | generation apparatus 1 can produce | generate the connection rule with respect to the newest keyword contained in those articles. As a result, it is not necessary for the user himself to extract the latest keyword and input it to the link generation device 1 in order to generate the connection rule, so that the link generation device 1 can reduce the user's processing effort.

  In addition, when a web page including the latest keyword is input, the link generation device 1 can extract the latest keyword and embed a hyperlink in the extracted latest keyword. Thus, the user can easily know the meaning of the latest keyword by opening the hyperlink of the latest keyword.

  Also, the link generation device 1 can compare at a time whether one word is a keyword, two words are a keyword, or three words are a keyword, and the keyword with the highest probability is selected with high accuracy. it can.

Note that a system including a plurality of devices may process each process of the link generation device 1 of this embodiment in a distributed manner by the plurality of devices.
Further, by recording a program for executing each process of the link generation device 1 of the present embodiment on a computer-readable recording medium, causing the computer system to read and execute the program recorded on the recording medium. The various processes described above related to the link generation device 1 may be performed.

  Here, the “computer system” may include an OS and hardware such as peripheral devices. Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW system is used. The “computer-readable recording medium” means a flexible disk, a magneto-optical disk, a ROM, a writable nonvolatile memory such as a flash memory, a portable medium such as a CD-ROM, a hard disk built in a computer system, etc. This is a storage device.

  Further, the “computer-readable recording medium” refers to a volatile memory (for example, DRAM (Dynamic) in a computer system serving as a server or a client when a program is transmitted via a network such as the Internet or a communication line such as a telephone line. Random Access Memory)) that holds a program for a certain period of time is also included. The program may be transmitted from a computer system storing the program in a storage device or the like to another computer system via a transmission medium or by a transmission wave in the transmission medium. Here, the “transmission medium” for transmitting the program refers to a medium having a function of transmitting information, such as a network (communication network) such as the Internet or a communication line (communication line) such as a telephone line. The program may be for realizing a part of the functions described above. Furthermore, what can implement | achieve the function mentioned above in combination with the program already recorded on the computer system, and what is called a difference file (difference program) may be sufficient.

  As mentioned above, although embodiment of this invention was explained in full detail with reference to drawings, the concrete structure is not restricted to this embodiment, The design etc. of the range which does not deviate from the summary of this invention are included.

DESCRIPTION OF SYMBOLS 1 Link production | generation apparatus 10 Input part 20 Storage part 21 1st sentence storage part 22 2nd sentence storage part 23 Rule storage part 24 User storage part 25 Link necessity storage part 30 Display part 100 Control part 101 Word extraction part 102 Key word extraction unit 103 Adjacent word extraction unit 104 Keyword extraction unit 110 Rule generation unit 111 Bigram connection rule generation unit 112 Trigram connection rule generation unit 130 Link word generation unit 131 Bigram connection determination unit 132 Trigram connection determination unit 135 Acquisition unit 136 Link necessity determination unit 140 Link insertion unit

Claims (17)

A keyword extraction unit that extracts keywords from sentences included in the content;
An acquisition unit for acquiring the degree of participation of a user in a specific community and the degree of specialization of each keyword in a specific community;
A link necessity determination unit that determines whether to insert a hyperlink for a keyword based on the degree of participation and the degree of specialization acquired by the acquisition unit;
A link generation apparatus comprising:   The link generation apparatus according to claim 1, wherein the participation degree in the specific community is a score determined based on a cumulative value of a user's participation time in the community.   The degree of specialization to the specific community is high with respect to keywords that appear biased in the specific community when comparing keywords included in articles posted to the general community and keywords included in articles posted to the specific community. The link generation device according to claim 1, wherein a score is given.   When the degree of participation is equal to or less than a predetermined degree of participation and the degree of specialization of the target keyword to a specific community is greater than or equal to a predetermined degree of specialization, the link necessity determination unit 4. The link generation device according to claim 1, further comprising: determining to insert a hyperlink to a specific Web content. 5.   The link generation apparatus according to any one of claims 1 to 4, wherein the link necessity determination unit determines whether or not a link insertion is necessary with reference to a link necessity storage unit.   The link necessity storage unit stores, for each user, a keyword, information indicating whether or not the keyword has been clicked, and information indicating whether or not the keyword is not linked. The link generation device according to claim 5, characterized in that: The keyword extraction unit
A word extraction unit that extracts words from sentences included in content posted in a specific community;
An adjacent word extraction unit that extracts adjacent words from the words extracted by the word extraction unit;
A concatenated word generation unit that generates a keyword by concatenating adjacent words extracted by the adjacent word extraction unit;
The link generation device according to any one of claims 1 to 6, further comprising: A rule storage unit in which a keyword, a connection rule of the keyword, and a degree of specialization of the keyword to a specific community are associated and stored;
The connective word generation unit
When the two keywords are adjacent to each other, a bigram connection determination unit that generates a keyword without connecting or connecting two words based on a connection rule stored in the rule storage unit;
A trigram connection determination unit that generates keywords without connecting or connecting three words or connecting two words based on a connection rule stored in the rule storage unit when three of the keywords are adjacent to each other; ,
The link generation device according to claim 7, further comprising:   The trigram connection determination unit extracts a keyword in which four or more words are connected by applying a connection rule of three words for every three words when the number of adjacent words is four or more. Item 9. The link generation device according to Item 8. A rule storage unit in which a keyword, a connection rule of the keyword, and a degree of specialization of the keyword to a specific community are associated and stored;
The link according to any one of claims 1 to 7, wherein the acquisition unit acquires the specialization degree by reading out the specialization degree of the keyword to a specific community from the rule storage unit. Generator.   When two or more words are adjacent to each other, the degree of connection indicating the degree to which the adjacent words are connected is statistically calculated, and a connection rule for connecting words is generated based on the calculated degree of connection, and the generation The link generation device according to any one of claims 8 to 10, further comprising a rule generation unit that stores the connected rule in the rule storage unit. The rule generator is
A bigram connection rule generating unit for calculating a connection degree when two words are connected and when not connecting, and generating a connection rule for connecting two words based on the calculated connection degree;
Calculate the degree of connection when three words are connected, when two adjacent words of the three words are connected, and when none of the three words are connected, and connect the three words based on the calculated degree of connection A trigram concatenation rule generator for generating rules;
The link generation device according to claim 11, comprising:   The link generation apparatus according to claim 12, wherein the connectivity is based on an Akaike information amount standard.   The link generation device according to claim 7, wherein the word extraction unit extracts the word after normalizing the sentence. A user storage unit that stores the identification information of the user and the degree of participation of the user in a specific community in association with each other;
The link according to any one of claims 1 to 14, wherein the acquisition unit acquires the participation degree by reading out the participation degree of the user in a specific community from the user storage unit. Generator. A link generation method of a link generation device including an input unit, a storage unit, and a control unit ,
An input procedure in which the input unit receives input of content;
A keyword extraction procedure in which the control unit extracts a keyword from a sentence included in the content input from the input unit ;
An acquisition procedure for reading out the degree of participation and the degree of specialization from the storage unit in which the control unit stores a degree of participation in a specific community of a user and a degree of specialization of each keyword in a specific community;
A link necessity determination procedure for determining whether to insert a hyperlink for a keyword based on the participation level and the specialization level read from the storage unit by the control unit ;
A link generation method characterized by comprising:
On the computer,
A keyword extraction step for extracting keywords from sentences contained in the content;
An acquisition step for acquiring the degree of participation of the user in the specific community and the degree of specialization of each keyword in the specific community;
A link necessity determination step for determining whether or not to insert a hyperlink for a keyword based on the participation degree and the specialization degree acquired by the acquisition step;
Link generation program to execute

Images