JP5711689B2 - Topic word extraction device, topic word extraction method, and program - Google Patents

Description

  The present invention relates to a topic word extraction device, a topic word extraction method, and a program for extracting a topic word from information transmitted for communication between users via the Internet.

  In recent years, communication between users via the Internet by e-mail, blog, SNS, TWITTER (registered trademark), and the like has increased explosively. The information transmitted by e-mail, blog, etc. for communication between users via the Internet shows the user's emotional expression and consumption trend, and also has an advertising effect as a word of mouth. Therefore, it is important to collect and analyze information transmitted for communication between users via the Internet. Therefore, it is more and more important to extract the words that are the topic (hereinafter referred to as topic words) from the information transmitted for communication between users via the Internet, indicating the characteristics of the content of the transmitted information. It has become.

  For example, as a conventional example of extracting document keywords, there is a tf-idf method as a method of extracting document keywords by a statistical method (see Non-Patent Document 1). This “tf-idf method” is a method in which a large number of words appear in a processing target document based on the frequency of occurrence of a word and the number of documents in which the word appears, and appear in other documents. In this method, a high importance level is calculated for a few words, and keywords are extracted based on the calculated importance level.

Gerard Salton, Michael J. et al. McGill “Introduction to Modern Information Retrieval”, McGraw-Hill, 1983.

  However, information transmitted for communication between users via the Internet is close to a conversation sentence, and there is a feature that words for facilitating communication are frequently used regardless of the topic. Therefore, if a word with a large number of appearances is used as a topic word, words for facilitating communication are included, and words other than the true topic word are also extracted as topic words. A topic word extracted from information transmitted for communication between users via the Internet using the “tf-idf method” has a problem that accuracy is poor and cannot be used for analysis as it is.

  Accordingly, the present invention has been made in view of the above-described problems, and a topic word extraction device and a topic word extraction method for extracting a true topic word from information transmitted for communication between users via the Internet. And to provide a program.

  The present invention proposes the following matters in order to solve the above problems. In addition, in order to make an understanding easy, although the code | symbol corresponding to embodiment of this invention is attached | subjected and demonstrated, it is not limited to this.

  (1) The present invention is a topic word extraction device for extracting a topic word from communication information transmitted for communication between users via the Internet, and one or more pieces of communication information storing the communication information Communication information acquisition means (for example, equivalent to the communication information acquisition unit 110 in FIG. 1) that acquires communication information for a predetermined period from the storage server, and morphological analysis is performed on text information included in the acquired communication information to extract words. Morphological analysis means (for example, equivalent to the morphological analysis unit 120 in FIG. 1) and importance calculation means for calculating the importance for each word extracted by the morpheme analysis means (for example, in the importance calculation unit 130 in FIG. 1) Equivalent) and the communication information acquisition means Unique user number acquisition means (for example, FIG. 1) that acquires the number of unique users who transmitted communication information including each word in the text information for each word extracted by the morphological analysis means with reference to the communication information The unique user number acquisition unit 140), the unique user number acquired by the unique user number acquisition unit, and the number of communication information acquired by the communication information acquisition unit Weighting coefficient calculating means for calculating a weighting coefficient (for example, equivalent to the weighting coefficient calculating section 150 in FIG. 1), importance for each word calculated by the importance calculating means, and calculated by the weighting coefficient calculating means From the words extracted by the morpheme analysis means based on the weighting coefficient, A topic word extraction device is provided that includes a topic word extraction means (for example, equivalent to the topic word extraction unit 160 in FIG. 1).

  According to this invention, the communication information acquisition means acquires communication information for a predetermined period from one or more communication information storage servers that store communication information. The morpheme analysis unit performs morpheme analysis on text information included in the acquired communication information, and extracts words. The importance calculation means calculates the importance for each word extracted by the morpheme analysis means. The unique user number acquisition means refers to the communication information acquired by the communication information acquisition means, and acquires, for each word extracted by the morphological analysis means, the number of unique users who transmitted communication information including each word in the text information. To do. The weighting coefficient calculation means calculates a weighting coefficient that takes into account the unique user based on the number of unique users acquired by the unique user number acquisition means and the number of communication information acquired by the communication information acquisition means. The topic word extraction unit is configured to extract a topic word from the word extracted by the morpheme analysis unit based on the importance for each word calculated by the importance calculation unit and the weighting coefficient calculated by the weighting coefficient calculation unit. Extract. Therefore, by extracting topic words based on the importance considering the number of unique users from words of text information included in communication information transmitted for communication between users via the Internet, It is possible to extract true topic words that many users are talking about in communication between users via the Internet.

  (2) The present invention relates to the topic word extraction device according to (1), wherein the communication information includes retransmission information indicating whether or not the communication information is retransmitted communication information transmitted by another user, and the communication information Communication retransmitting communication information transmitted by the other user based on the retransmission information for each word extracted by the morphological analysis unit with reference to the communication information acquired by the acquisition unit Re-sending communication information number acquiring means (e.g., equivalent to the re-sending communication information number acquiring unit 210 in FIG. 3) for acquiring the number of re-sending communication information indicating the number of information. Re-transmission community acquired by means of acquiring communication information A topic word extraction device that calculates a weighting factor that takes into account the degree of propagation of communication information based on the number of application information and the number of communication information acquired by the communication information acquisition means .

  According to this invention, the communication information includes retransmission information indicating whether or not the communication information transmitted by another user is retransmitted. The retransmitted communication information number acquiring means refers to the communication information acquired by the communication information acquiring means, and for each word extracted by the morpheme analyzing means, communication transmitted by another user based on the retransmitted information. The number of retransmitted communication information indicating the number of communication information retransmitting information is acquired. The weighting coefficient calculating means weights considering the propagation degree of communication information based on the number of retransmitted communication information acquired by the number of retransmitted communication information acquiring means and the number of communication information acquired by the communication information acquiring means. Calculate the coefficient. Therefore, by extracting topic words based on the importance taking into account the degree of propagation of communication information, it is possible to extract the true topic words propagating between users in communication between users via the Internet. .

  (3) The present invention relates to the topic word extraction device according to (1) or (2), wherein the communication information storage server manages the link relationship between the users, and the link relationship between the users from the communication information storage server. For each word extracted by the morpheme analyzing means with reference to the communication information obtained by the link relation obtaining means (e.g., corresponding to the link relation obtaining unit 220 in FIG. 3) and the communication information obtaining means. The unique user who transmitted the communication information including each word is acquired, and based on the acquired unique user and the link relationship acquired by the link relationship acquiring unit, the other unique user acquired Get the number of linked users to get the number of linked users linked to unique users (E.g., corresponding to the link user number acquisition unit 230 in FIG. 3), and the weighting coefficient calculation means includes the link user number acquired by the link user number acquisition means and the unique user number acquisition means. On the basis of the number of acquired unique users, a topic word extraction device is proposed that calculates a weighting factor that takes into account the link relationship between the users.

  According to this invention, the communication information storage server manages the link relationship between users. The link relationship acquisition means acquires a link relationship between users from the communication information storage server. The link user number acquisition means acquires a unique user who has transmitted communication information including each word for each word extracted by the morphological analysis means with reference to the communication information acquired by the communication information acquisition means. Based on the unique user and the link relationship acquired by the link relationship acquisition means, the number of link users linked to other unique users among the acquired unique users is acquired. The weighting coefficient calculation means calculates a weighting coefficient that considers the link relationship between users based on the number of link users acquired by the link user number acquisition means and the number of unique users acquired by the unique user number acquisition means. . Therefore, by extracting the topic words based on the importance considering the link relationship between users, it is possible to extract the topic words that are the topic among the users having the link relationship.

  (4) Although the present invention is used to facilitate communication between users via the Internet for the topic word extraction device of (1) to (3), the content of text information included in the communication information An important word for extracting an important word from a word extracted by the morpheme analyzing means by using a previously prepared dictionary that stores a word that is not an important word representing the feature of Extraction means (e.g., equivalent to the keyword extraction unit 310 in FIG. 4), the importance calculation means calculates the importance for each keyword extracted by the keyword extraction means, and the topic word extraction means Is calculated by the importance calculating means based on the importance for each important word calculated by the importance calculating means and the weighting coefficient calculated by the weighting coefficient calculating means. It has proposed a topic word extraction device and extracting the topic word from the main language.

According to the present invention, the keyword extraction unit is used for facilitating communication between users via the Internet, but stores words that are not important words representing the characteristics of the contents of the text information included in the communication information. Using a dictionary prepared in advance, important words are extracted by removing words stored in the dictionary from words extracted by the morphological analysis means. The importance level calculating means calculates the importance level for each important word extracted by the important word extracting means. The topic word extraction means uses the importance words calculated by the importance calculation means based on the importance for each important word calculated by the importance calculation means and the weighting coefficient calculated by the weighting coefficient calculation means. To extract. Therefore, it is used to facilitate communication between users via the Internet, but words that are not important words representing the characteristics of the content of communication can be excluded from topic words.
(5) According to the present invention, in the topic word extraction device of (4), the dictionary stores a demonstrative pronoun dictionary that stores demonstrative pronouns (for example, equivalent to the demonstrative pronoun dictionary 321 in FIG. 4) and words used for greetings. Including a greeting dictionary (for example, equivalent to the greeting dictionary 322 in FIG. 4) and a time-dependent word dictionary (for example, equivalent to the time-dependent word dictionary 323 in FIG. 4) for storing words related to the time. We have proposed a featured topic word extraction device.

  According to the present invention, the dictionary includes an indicating pronoun dictionary that stores demonstrative pronouns, a greeting dictionary that stores words used for greetings, and a time-dependent word dictionary that stores words related to the time of each time. Therefore, by removing words related to pronouns, greetings, and time clauses, which are words used to facilitate communication between users via the Internet, from words extracted from text information included in communication information It is possible to extract important words representing the characteristics of communication contents.

  (6) The present invention relates to the topic word extracting device of (5), wherein the important word extracting means stores the words stored in the demonstrative pronoun dictionary and the greeting dictionary from the words extracted by the morpheme analyzing means. A key word candidate is extracted, and for each word of the extracted key word candidate, a combination of the word time period and the word specified based on the transmission date and time of communication information of text information including the word The topic word extracting device is characterized in that it is determined whether or not is stored in the time-dependent word dictionary and a word not stored in the time-based word dictionary is extracted as an important word.

  According to this invention, the important word extracting unit extracts the important word candidates from the words extracted by the morpheme analyzing unit, excluding the words stored in the demonstrative pronoun dictionary and the greeting dictionary, and the extracted important words For each word candidate word, it is determined whether or not the combination of the time and word of the word specified based on the transmission date and time of communication information including the word is stored in the time-dependent word dictionary, and the time-dependent word dictionary Extract words that are not stored as important words. Therefore, before using the word-by-period dictionary, you can reduce the number of words that specify the time by removing words stored in the pronoun dictionary and greeting dictionary, and as a result, extract important words efficiently. can do.

  (7) In the topic word extraction device according to (5) or (6), the word used for the greeting includes a word used for chatting before entering the main topic in conversation or at the end of the conversation. We have proposed a topic word extraction device characterized by this.

  According to the present invention, the word used for greeting includes the word used for chatting before entering the main topic in conversation or at the end of conversation. Therefore, by including words in the greeting dictionary in a broad sense, the word used for information related to each other before the main subject, the weather, irrelevant chats before and after, and chats that occur when the conversation ends, It is used to facilitate communication between users via the Internet. By removing words that are not important words, it is possible to extract important words that represent the characteristics of the content of communication.

  (8) The present invention proposes a topic word extraction device (5) to (7), wherein the time is a season, a day of the week, or a time zone.

  According to the present invention, the time is the season, day of the week, and time zone. Therefore, it is possible to extract important words representing the characteristics of the contents of communication, excluding words related to seasons, days of the week, and time zones, which are often used to facilitate communication between users via the Internet.

  (8) The present invention comprises communication information acquisition means, morpheme analysis means, importance calculation means, unique user number acquisition means, weighting coefficient calculation means, and topic word extraction means for communication between users via the Internet. A topic word extraction method in a topic word extraction device for extracting a topic word from communication information transmitted to the communication information, wherein the communication information acquisition unit is configured to receive a predetermined value from one or more communication information storage servers storing the communication information. A first step (for example, step S1 in FIG. 2) for acquiring communication information of a period, and the morpheme analyzing means morphologically analyze text information included in the communication information acquired in the first step, Second step to extract For example, step S2) in FIG. 2, the third step (for example, step S3 in FIG. 1) in which the importance calculation unit calculates the importance for each word extracted by the morpheme analysis unit, and the unique A unique user who has transmitted the communication information including each word in the text information for each word extracted in the third step with reference to the communication information acquired in the first step. A fourth step of acquiring a number (for example, step S4 in FIG. 2), the weighting coefficient calculating means, the number of unique users acquired in the fourth step, and the communication acquired in the first step. 5th step (for example, step S5 of FIG. 2) which calculates the weighting coefficient which considers a unique user based on the number of information And the topic word extraction means is extracted in the third step based on the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. A topic word extraction method characterized by including a sixth step (for example, step S6 in FIG. 2) for extracting a topic word from a word is proposed.

  According to this invention, first, in the first step, the communication information acquisition means acquires communication information for a predetermined period from one or more communication information storage servers that store communication information. Next, in the second step, the morpheme analyzing means performs morphological analysis on the text information included in the communication information acquired in the first step, and extracts words. Next, in a third step, the importance calculation means calculates the importance for each word extracted by the morpheme analysis means. Next, in the fourth step, the unique user number acquisition means refers to the communication information acquired in the first step and includes each word in the text information for each word extracted in the third step. Get the number of unique users who sent communication information. Next, in the fifth step, the weighting coefficient calculation means considers unique users based on the number of unique users acquired in the fourth step and the number of communication information acquired in the first step. A weighting coefficient is calculated. Next, in the sixth step, the topic word extracting means performs the third step based on the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. Extract topic words from the extracted words. Therefore, by extracting topic words from the words of text information included in communication information transmitted for communication between users via the Internet based on the importance considering the unique user, In the communication between the users, it is possible to extract the true topic words that many users are talking about.

  (9) The present invention comprises communication information acquisition means, morpheme analysis means, importance calculation means, unique user number acquisition means, weighting coefficient calculation means, and topic word extraction means for communication between users via the Internet. A program for causing a computer to execute a topic word extraction method in a topic word extraction device that extracts a topic word from communication information transmitted to a computer, wherein the communication information acquisition means stores the communication information. A first step of acquiring communication information for a predetermined period from the communication information storage server (for example, step S1 in FIG. 2), and a text included in the communication information acquired by the morpheme analyzing unit in the first step. information A second step (for example, step S2 in FIG. 2) of analyzing the morphological element and extracting a word; and a third step of calculating the importance level for each word extracted by the morpheme analyzing means. Step (for example, step S3 in FIG. 1) and the unique user number acquisition means refer to the communication information acquired in the first step, and for each word extracted in the third step, A fourth step (for example, step S4 in FIG. 2) for acquiring the number of unique users who have transmitted communication information including a word in the text information, and the weighting coefficient calculation means is the unique number acquired in the fourth step. Based on the number of users and the number of pieces of communication information acquired in the first step, a weighting factor for calculating a weighting factor considering a unique user is calculated. 5 (for example, step S5 in FIG. 2), and the topic word extraction means determines the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. Based on this, a sixth step (for example, step S6 in FIG. 2) for extracting a topic word from the word extracted in the third step and a program for causing a computer to execute are proposed.

  According to this invention, first, in the first step, the communication information acquisition means acquires communication information for a predetermined period from one or more communication information storage servers that store communication information. Next, in the second step, the morpheme analyzing means performs morphological analysis on the text information included in the communication information acquired in the first step, and extracts words. Next, in a third step, the importance calculation means calculates the importance for each word extracted by the morpheme analysis means. Next, in the fourth step, the unique user number acquisition means refers to the communication information acquired in the first step and includes each word in the text information for each word extracted in the third step. Get the number of unique users who sent communication information. Next, in the fifth step, the weighting coefficient calculation means considers unique users based on the number of unique users acquired in the fourth step and the number of communication information acquired in the first step. A weighting coefficient is calculated. Next, in the sixth step, the topic word extracting means performs the third step based on the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. Extract topic words from the extracted words. Therefore, by extracting topic words from the words of text information included in communication information transmitted for communication between users via the Internet based on the importance considering the unique user, In the communication between the users, it is possible to extract the true topic words that many users are talking about.

According to the present invention, at least one of the number of unique users, the degree of information propagation, and the link relationship between users among the words of the text information included in the communication information transmitted for communication between users via the Internet. By extracting the topic words based on the importance taking into account one, it is possible to extract the true topic words that many users are talking about in communication between users via the Internet.

It is a figure which shows the function structure of the hot topic extraction apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart of the topic word extraction process which concerns on the 1st Embodiment of this invention. It is a figure which shows the function structure of the hot topic extraction apparatus which concerns on the 2nd Embodiment of this invention. It is a figure which shows the function structure of the hot topic extraction apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart which shows an example of the extraction process of the important word which concerns on the 3rd Embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. Note that the components in the present embodiment can be appropriately replaced with existing components and the like, and various variations including combinations with other existing components are possible. Therefore, the description of the present embodiment does not limit the contents of the invention described in the claims.

<First Embodiment>
A first embodiment of the present invention will be described with reference to FIGS. 1 and 2.

<Functional configuration of topic word extraction device>
FIG. 1 is a diagram showing a functional configuration of a topic word extraction device 100 according to the first embodiment of the present invention. The topic word extraction device 100 is connected to the communication information storage server 10 via a communication network.

  The communication information accumulation server 10 accumulates and manages communication information transmitted for communication between users via the Internet. Here, the communication information is, for example, e-mail, post information to social networking services such as blogs, SNS and TWITTER, and instant messages. The communication information of the present embodiment includes text information indicating the content of communication, date and time when communication information was transmitted (hereinafter referred to as transmission date and time), and information that uniquely identifies the user who transmitted the communication information (for example, user ID and user name).

  As shown in FIG. 1, the topic word extraction device 100 includes a communication information acquisition unit 110, a morpheme analysis unit 120, an importance calculation unit 130, a unique user number acquisition unit 140, a weighting coefficient calculation unit 150, and a topic word extraction unit 160. Is provided.

  The communication information acquisition unit 110 acquires communication information from the communication information storage server 10 for a predetermined period. For example, the communication information is acquired from the communication information storage server 10 by using an API provided by a provider providing the communication information storage server. Here, the predetermined period can be arbitrarily set by an analyst who analyzes the communication information, and can be set to the most recent one month, last 12 months, or the like.

  The morphological analysis unit 120 performs morphological analysis on the communication information acquired by the communication information acquisition unit 110 and extracts words. Note that the morpheme analysis unit 120 may extract a specific part of speech that is set in advance from words that have been analyzed by the morpheme analysis unit 120.

  The importance calculation unit 130 calculates the importance of each word extracted by the morphological analysis unit 120. As a method of calculating importance, there is a method of calculating tf-idf (Term Frequency Inverse Document Frequency) and using the score as importance.

  The unique user number acquisition unit 140 first extracts communication information of text information including each word from the communication information acquired by the communication information acquisition unit 110 for each word extracted by the morphological analysis unit 120. And the unique user number acquisition part 140 counts and acquires the number of unique users for every word based on the user information of the extracted communication information. Here, the number of unique users is the number of unique users who have transmitted communication information of text information including words, and the same user can transmit communication information including the same words once and all times This is the net number of users who used the word.

  For each word extracted by the morphological analysis unit 120, the weighting coefficient calculation unit 150 first determines the number of communication information including each word from the communication information acquired by the communication information acquisition unit 110 (hereinafter referred to as communication information number). Count and get. Next, for each word extracted by the morpheme analyzer 120, the weighting coefficient calculator 150 calculates the number of unique users acquired by the number of unique users acquired by the number of pieces of communication information acquired as shown in Equation 1. The divided quotient is calculated as a weighting coefficient α. The weighting coefficient α has a small value when the number of unique users is small, that is, a word that is frequently used by a specific user, whereas the weighting coefficient α is large, that is, the number of unique users is large, that is, the frequency of use of a specific user is low. In the case of a word used by a unique user, the value becomes large.

  The topic word extraction unit 160 multiplies the importance calculated by the importance calculation unit 130 by the weighting coefficient α calculated by the weighting coefficient calculation unit 150, and based on the importance obtained by multiplying the weighting coefficient α, the morpheme A topic word is extracted from the words extracted by the analysis unit 120. For example, the topic word extraction unit 160 extracts words whose importance level multiplied by the weighting coefficient α is equal to or higher than a preset threshold value as topic words. The topic word extraction unit 160 may arrange the words extracted by the morpheme analysis unit 120 in descending order of importance multiplied by the weighting coefficient α, and may use a predetermined number of words from the top as topic words.

  By using a value obtained by multiplying the importance calculated by tf-idf or the like with a weighting coefficient α taking into account the unique user as the importance used when extracting the topic word, the importance of a word frequently used by a specific user is used. On the other hand, the importance of words used by many unique users can be increased. As a result, true topic words used by many users can be extracted in communication between users via the Internet.

<Topic word extraction processing flow>
FIG. 2 is a diagram showing a topic word extraction processing flow according to the first embodiment of the present invention.

  First, in step S <b> 1, the communication information acquisition unit 110 acquires communication information from the communication information accumulation server 10.

  Next, in step S2, the morphological analysis unit 120 performs morphological analysis on the text information included in the communication information acquired in step S1, and extracts words.

  Next, in step S3, the importance calculation unit 130 calculates the importance of the word extracted in step S2.

  Next, in step S4, the unique user number acquisition unit 140 counts and acquires the number of unique users who have transmitted the communication information of the text information including the words for each word extracted in step S2.

  Next, in step S5, the weighting coefficient calculation unit 150 uses a quotient obtained by dividing the number of unique users acquired in step S4 by the number of communication information acquired in step S1 as a weighting coefficient α in consideration of the unique users. calculate.

  Next, in step S6, the topic word extraction unit 160 multiplies the importance calculated in step S3 by the weighting coefficient α calculated in step S5, and based on the importance obtained by multiplying the weighting coefficient α, A topic word is extracted from the words extracted in step S2.

  As described above, according to the present embodiment, the importance calculated by tf-idf or the like is unique from the words included in the text information of communication information transmitted in communication between users via the Internet. By extracting the topic words based on the importance obtained by multiplying the weighting coefficient α considering the number of users, the true topic words that many users are talking about are extracted in communication between users via the Internet. be able to.

<Second Embodiment>
A second embodiment of the present invention will be described with reference to FIG. In addition, the topic word extraction apparatus in this embodiment extracts a topic word in consideration of the propagation degree of a word and the link relation of a user who uses the word. In addition, about the component which attaches | subjects the same code | symbol as 1st Embodiment, since it has the same function, the detailed description is abbreviate | omitted.

  FIG. 3 is a diagram showing a functional configuration of the topic word extraction device 200 according to the second embodiment of the present invention. The topic word extraction device 200 is connected to the communication information storage server 10 via a communication network.

  In the present embodiment, the communication information stored and managed by the communication information storage server 10 includes the date and time when the communication information was transmitted (hereinafter referred to as the transmission date and time) and information that uniquely identifies the user who transmitted the communication information ( For example, retransmission information is included together with a user ID and a user name. Here, the retransmission information is information indicating whether or not the text information of the communication information includes all or a part of the text information of the communication information of other users. It is information indicating whether or not there is.

  As shown in FIG. 1, the topic word extraction apparatus 200 includes a communication information acquisition unit 110, a morpheme analysis unit 120, an importance calculation unit 130, a unique user number acquisition unit 140, a retransmitted communication information number acquisition unit 210, and a link relationship acquisition. Unit 220, link user number acquisition unit 230, weighting coefficient calculation unit 151, and topic word extraction unit 161.

  The number of retransmitted communication information acquisition unit 210 is the number of communication information retransmitting communication information transmitted by other users based on the retransmitted information included in the communication information acquired by communication information acquiring unit 110. (Hereinafter referred to as the number of retransmitted communication information).

  The link relationship acquisition unit 220 acquires a link relationship between users from the communication information storage server 10. For example, the information is that user B has a link relationship with user A. Note that it is arbitrary how many levels of link relations are acquired, but it is desirable to have up to the second and third levels.

  The link user number acquisition unit 230 first extracts a unique user who has transmitted communication information including a word for each word, based on the user information of the extracted communication information. Next, the link user number acquisition unit 230 divides the extracted unique user into a link user and a linked user for each word based on the link relationship acquired by the link relationship acquisition unit 220. Here, the link user is a user who has linked to other users, and is, for example, a follower of TWITTER. On the other hand, the linked user is a user who has a link relationship with another user, for example, a person who is followed by TWITTER. And the link user number acquisition part 230 counts and acquires the number of link users (henceforth a link user number) for every word.

  First, the weighting coefficient calculation unit 151 searches for communication information including each word from the communication information acquired by the communication information acquisition unit 110 for each word extracted by the morpheme analysis unit 120, and determines the number of communication information for each word. get. Next, for each word extracted by the morphological analysis unit 120, the weighting coefficient calculation unit 151 retransmits the number of communication information acquired by the retransmission communication information number acquisition unit 210 as shown in Equation 2. A quotient obtained by dividing the number of communication information is calculated as a weighting coefficient β. The weighting coefficient β is small when the number of retransmitted communication information is small, that is, when the communication information is not propagated so much, while the value of the retransmitted communication information is large, that is, the communication information is wide. In the case of a propagating word, the value is large.

  Further, the weighting coefficient calculation unit 151 uses the number of unique users acquired by the unique user number acquisition unit 140 for each word extracted by the morpheme analysis unit 120 and the number of link users acquisition unit 230 as shown in Equation 3. A quotient obtained by dividing the number of acquired link users is calculated as a weighting coefficient γ. The weighting factor γ has a large value in the case of a word having a large number of link users, that is, a word that is frequently used among users in a link relationship, while the number of link users is small, that is, in a link relationship. In the case of a word that is frequently used except between users, the value is small.

  The topic word extraction unit 161 multiplies the importance calculated by the importance calculation unit 130 by the weighting coefficient α and the weighting coefficient β calculated by the weighting coefficient calculation unit 151, and the weighted coefficients α and β multiplied by the importance Based on the degree, the topic word is extracted from the words extracted by the morphological analysis unit 120. For example, the topic word extraction unit 161 extracts a word whose importance level multiplied by the weighting coefficients α and β is equal to or higher than a preset threshold value as a topic word. The topic word extraction unit 161 may arrange the words extracted by the morpheme analysis unit 120 in descending order of importance multiplied by the weighting coefficients α and β, and may use a predetermined number of words as the topic words from the top.

  As the importance used when extracting the topic word, a value obtained by multiplying the importance calculated by tf-idf or the like with the weighting coefficient α considering the unique user and the weighting coefficient β considering the word propagation degree is used. This reduces the importance of words that are not propagated between users, while increasing the importance of words that are widely propagated between users. As a result, it is possible to extract true topic words that are used by many users and are widely propagated in communication between users via the Internet.

  Further, the topic word extraction unit 161 multiplies the importance calculated by the importance calculation unit 130 by the weighting coefficient α and the weighting coefficient γ calculated by the weighting coefficient calculation unit 151, and the weighting coefficients α and γ are multiplied. Based on the importance, the topic word is extracted from the words extracted by the morphological analysis unit 120. For example, the topic word extraction unit 161 extracts a word whose importance level multiplied by the weighting coefficients α and γ is equal to or more than a preset threshold value as a topic word. The topic word extraction unit 161 may arrange the words extracted by the morpheme analysis unit 120 in descending order of importance multiplied by the weighting coefficients α and γ, and may use a predetermined number of words as the topic words from the top.

  As the importance used when extracting a topic word, a value obtained by multiplying the importance calculated by tf-idf or the like with a weighting coefficient α considering a unique user and a weighting coefficient γ considering a link relation between users is used. Accordingly, it is possible to reduce the importance of words that are frequently used except between users in a link relationship, while increasing the importance of words that are frequently used between users in a link relationship. As a result, in communication between users via the Internet, it is a word that is used by many users and is frequently used among users who are linked, that is, a word that is talked about by some users. From this, it is possible to extract words that are likely to become topics after being propagated between users as topic words.

  Note that the topic word extraction unit 161 may use the importance obtained by multiplying both the weighting coefficient β and the weighting coefficient γ together with the weighting coefficient α for topic word extraction.

  As described above, according to the present embodiment, the importance calculated by tf-idf or the like from words included in the text information of communication information transmitted in communication between users via the Internet, By extracting topic words based on the importance obtained by multiplying the weighting coefficient β taking into account the degree of propagation of write information together with the weighting coefficient α taking into account the unique user, it is propagated between users in communication between users via the Internet. It is possible to extract true topic words.

  Furthermore, among the words included in the text information of the communication information transmitted in the communication between users via the Internet, the importance calculated by tf-idf etc. is used together with the weighting coefficient α considering the unique user. By extracting the topic word based on the importance obtained by multiplying the weighting coefficient γ considering the link relationship of the word, it is a word that has become a topic in some users in communication between users via the Internet, From this, it is possible to extract words that are propagated between users and become a topic as topic words.

<Third Embodiment>
A third embodiment of the present invention will be described with reference to FIGS.

<Functional configuration of topic word extraction device>
FIG. 4 is a diagram showing a functional configuration of a topic word extraction device 300 according to the third embodiment of the present invention. The topic word extraction device 300 is connected to the communication information storage server 10 via a communication network.

  As shown in FIG. 4, the topic word extraction device 300 includes a communication information acquisition unit 110, a morpheme analysis unit 120, an importance calculation unit 132, a unique user number acquisition unit 142, a weighting coefficient calculation unit 152, a topic word extraction unit 162, An important word extraction unit 310 and a dictionary storage unit 320 are provided.

  The keyword extraction unit 310 extracts the keyword by removing the words stored in the dictionary storage unit 320 prepared in advance from the words morphologically analyzed by the morphological analysis unit 120. The important word extraction process will be described later.

  The dictionary storage unit 320 stores a dictionary prepared in advance. The dictionary stores words that are used to facilitate communication between users via the Internet, but are not important words representing the characteristics of the content of communication. In addition, an important word is a word showing the characteristic of the content of communication.

  In the present embodiment, the dictionary storage unit 320 is a word used to facilitate communication between users via the Internet, and is used as a pronoun or greeting that is a word that does not become an important word in communication. A pronoun pronoun dictionary 321, a greeting dictionary 322, and a time-dependent word dictionary 323 that store words and time-related words, respectively, are stored. The dictionary stored in the dictionary storage unit 320 can be added and deleted.

  The demonstrative pronoun dictionary 321 stores demonstrative pronouns. For example, when the text information is Japanese, he, girlfriend, and the like are stored.

  The greeting dictionary 322 stores words used for greetings, such as hello, goodbye and thank you.

  Note that the words used for greetings may include words used for chatting before entering the main topic in conversation or at the end of conversation. Since the communication information is transmitted for communication between users via the Internet, the text information is close to a conversational sentence. For this reason, it is common for the written information to include information about each other, the weather, irrelevant chats before and after the main topic, and chats that occur at the end of the conversation before the main topic. is there. The words included in these chats are words that are used to facilitate communication between users via the Internet, and are not important words in communication information, as in the case of Hello, Goodbye, etc. These words are also included in the greeting dictionary 322, so that words used for smoothing communication information can be excluded from important words.

  The word-by-time dictionary 323 stores words related to the time for each time. Here, time is a time zone, a day of the week, a season, or the like. The hourly word dictionary 323 stores words such as “morning” and “breakfast” related to the morning, for example, in association with the time zone “morning”, and associates with the day of the week “Sunday” in Sunday. Stores the related words “Sunday” and “Holiday”.

  Here, an example of the important word extraction processing by the important word extraction unit 310 using the pronoun pronoun dictionary 321, the greeting dictionary 322, and the time-dependent word dictionary 323 stored in the dictionary storage unit 320 will be described with reference to FIG. 5. I will explain.

  First, in step S <b> 11, the important word extraction unit 310 removes words stored in the demonstrative pronoun dictionary 321 from the words analyzed by the morphological analysis unit 120.

  Next, in step S12, the important word extraction unit 310 removes words stored in the greeting dictionary 322 from the words remaining as a result of step S11. Here, the word remaining as a result of step S12 is set as an important word candidate. Note that the order of step S11 and step S12 may be reversed.

  Next, in step S <b> 13, the keyword extraction unit 310 extracts one word from the keyword candidates. The extracted word is excluded from the important word candidates.

  Next, in step S14, the important word extraction unit 310 identifies the time period of the word extracted in step S13 based on the posting date and time of the writing information including the word extracted in step S13.

  Next, in step S15, the important word extraction unit 310 determines whether or not the word extracted in step S13 in association with the time clause specified in step S14 is stored in the hourly word dictionary 323. If it is stored in the time-dependent word dictionary 323 (YES), the word extracted in step S13 is not extracted as an important word, and the process proceeds to step S17. On the other hand, if it is not stored in the hourly word dictionary 323 (NO), the process proceeds to step S16.

  In step S16, the keyword extraction unit 310 includes the word extracted in step S13 in the keyword.

  In step S <b> 17, the keyword extraction unit 310 determines whether keyword candidates remain. If important word candidates remain (YES), the process returns to step S13. On the other hand, if no important word candidates remain (NO), the process ends.

  In this way, by using words stored in the demonstrative pronoun dictionary and greeting dictionary before using the time-dependent word dictionary, it is possible to reduce the number of words that specify the time, and as a result, efficiently Words can be extracted.

  The unique user number acquisition unit 140 first extracts communication information including each important word from the communication information acquired by the communication information acquisition unit 110 for each important word extracted by the important word extraction unit 310. And the unique user number acquisition part 140 counts and acquires the number of unique users based on the user information of the extracted communication information for every important word. Here, the number of unique users is the number of unique users who have transmitted communication information including important words, and the same user can collectively transmit communication information including the same important words as one transmission. This is the net number of users who used the important words.

  The weighting coefficient calculation unit 152 first counts and acquires the number of pieces of communication information including each important word from the communication information acquired by the communication information acquisition unit 110 for each important word extracted by the important word extraction unit 310. Next, the weighting coefficient calculation unit 152 obtains the number of unique users for each important word extracted by the important word extraction unit 310 with the acquired number of communication information as shown in the equation 1 shown in the first embodiment. A quotient obtained by dividing the number of unique users acquired by the unit 142 is calculated as a weighting coefficient α. The weighting coefficient α has a small value in the case of an important word with a small number of unique users, that is, a high usage frequency of a specific user, while the number of unique users is large, that is, the usage frequency of a specific user is low. For important words used by many unique users, the value increases.

  The topic word extraction unit 162 multiplies the importance calculated by the importance calculation unit 130 by the weighting coefficient α calculated by the weighting coefficient calculation unit 152, and based on the importance obtained by multiplying the weighting coefficient α by the importance A topic word is extracted from the important words extracted by the word extraction unit 310. For example, the topic word extraction unit 160 extracts, as a topic word, an important word whose importance level multiplied by the weighting coefficient α is equal to or greater than a preset threshold value. The topic word extraction unit 160 may arrange the important words extracted by the keyword extraction unit 310 in descending order of importance multiplied by the weighting coefficient α, and may use a predetermined number of important words as the topic words from the top. .

  As described above, according to the present embodiment, it is used to facilitate communication between users via the Internet, but from important words excluding non-important words representing the characteristics of the content of communication. , Tf-idf, etc., by extracting the topic word based on the importance obtained by multiplying the importance calculated by the weighting coefficient α taking into account the unique user, the truth that many users of social media are talking about Topic words can be extracted.

  The processing of the topic word extraction device is recorded on a computer-readable recording medium, and the program recorded on the recording medium is read and executed by each device constituting the topic word extraction device. A word extraction device can be realized. The computer system here includes an OS and hardware such as peripheral devices.

  Further, the “computer system” includes a homepage providing environment (or display environment) if a WWW (World Wide Web) system is used. 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.

  The embodiments of the present invention have been described in detail with reference to the drawings. However, the specific configuration is not limited to the embodiments, and includes designs and the like that do not depart from the gist of the present invention.

DESCRIPTION OF SYMBOLS 10 Communication information storage server 100 Topic word extraction apparatus 110 Communication information acquisition part 120 Morphological analysis part 130 Importance calculation part 140 Unique user number acquisition part 150 Weighting coefficient calculation part 160 Topic word extraction part

Claims (10)

A topic word extraction device that extracts topic words from communication information transmitted for communication between users via the Internet,
Communication information acquisition means for acquiring communication information for a predetermined period from one or more communication information storage servers storing the communication information;
Morphological analysis means for analyzing text information included in the acquired communication information and extracting words;
Importance calculating means for calculating importance for each word extracted by the morpheme analyzing means;
The number of unique users that refers to the communication information acquired by the communication information acquisition means and acquires the number of unique users who have transmitted the communication information including each word in the text information for each word extracted by the morphological analysis means Acquisition means;
A weighting coefficient calculating means for calculating a weighting coefficient considering a unique user based on the number of unique users acquired by the unique user number acquiring means and the number of communication information acquired by the communication information acquiring means;
Topic word extraction for extracting a topic word from words extracted by the morpheme analyzing means based on the importance for each word calculated by the importance calculating means and the weighting coefficient calculated by the weighting coefficient calculating means Means,
A topic word extraction device. The communication information includes retransmission information indicating whether or not the communication information transmitted by another user is retransmitted,
Referring to the communication information acquired by the communication information acquisition means, for each word extracted by the morphological analysis means, the communication information transmitted by the other user is retransmitted based on the retransmission information. Re-sending communication information number acquisition means for acquiring the number of re-sending communication information indicating the number of communication information being
The weighting coefficient calculating means determines the propagation degree of communication information based on the number of retransmitted communication information acquired by the retransmitted communication information number acquiring means and the number of communication information acquired by the communication information acquiring means. The topic word extraction device according to claim 1, wherein a weighting coefficient to be considered is calculated. The communication information storage server manages a link relationship between the users,
Link relationship acquisition means for acquiring a link relationship between users from the communication information storage server;
With reference to the communication information acquired by the communication information acquisition unit, for each word extracted by the morpheme analysis unit, a unique user who has transmitted the communication information including each word is acquired, and the acquired unique user and Based on the link relationship acquired by the link relationship acquisition unit, a link user number acquisition unit that acquires the number of link users linked to other unique users among the acquired unique users;
With
The weighting factor calculating means weights considering the link relationship between the users based on the number of link users acquired by the number of link user acquisition means and the number of unique users acquired by the number of unique user acquisition means. The topic word extraction device according to claim 1, wherein a coefficient is calculated. Using a dictionary prepared in advance to store words that are not important words representing the characteristics of the content of text information included in the communication information, but used to facilitate communication between users via the Internet, An important word extraction unit for extracting an important word by removing a word stored in the dictionary from the word extracted by the morphological analysis unit;
The importance calculating means calculates importance for each important word extracted by the important word extracting means;
The topic word extracting unit calculates the importance calculated by the importance calculating unit based on the importance for each important word calculated by the importance calculating unit and the weighting coefficient calculated by the weighting coefficient calculating unit. The topic word extraction device according to claim 1, wherein a topic word is extracted from a word.   The dictionary includes an indicating pronoun dictionary that stores demonstrative pronouns, a greeting dictionary that stores words used for greetings, and a time-dependent word dictionary that stores words related to time periods for each time zone. 4. The topic word extraction device according to 4. The important word extraction means extracts important word candidates from the words extracted by the morphological analysis means, except for the words stored in the demonstrative pronoun dictionary and the greeting dictionary,
Whether for each word of the extracted important word candidate, the combination of the word and the word specified based on the transmission date and time of the communication information of the text information including the word is stored in the time-dependent word dictionary 6. The topic word extraction apparatus according to claim 5, wherein a word that is not stored in the time-dependent word dictionary is extracted as an important word.   The topic word extraction device according to claim 5 or 6, wherein the word used for the greeting includes a word used for a chat exchanged before entering the main topic in a conversation or at the end of the conversation.   The topic word extracting device according to claim 5, wherein the time is a season, a day of the week, and a time zone. Communication information acquisition means, morphological analysis means, importance calculation means, unique user number acquisition means, weighting coefficient calculation means, and topic word extraction means, from communication information transmitted for communication between users via the Internet A topic word extraction method in a topic word extraction device for extracting a topic word,
A first step in which the communication information acquisition means acquires communication information for a predetermined period from one or more communication information storage servers storing the communication information;
A second step in which the morphological analysis means performs a morphological analysis on text information included in the communication information acquired in the first step, and extracts a word;
A third step in which the importance calculation means calculates importance for each word extracted by the morpheme analysis means;
The unique user number acquisition means refers to the communication information acquired in the first step and transmits communication information including each word in the text information for each word extracted in the third step. A fourth step of acquiring the number of unique users;
The weighting coefficient calculating means calculates a weighting coefficient that takes into account a unique user based on the number of unique users acquired in the fourth step and the number of communication information acquired in the first step. 5 steps,
The topic word extraction unit is configured to extract the word extracted in the third step based on the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. A sixth step of extracting topic words;
A topic word extraction method characterized by including Communication information acquisition means, morphological analysis means, importance calculation means, unique user number acquisition means, weighting coefficient calculation means, and topic word extraction means, from communication information transmitted for communication between users via the Internet A program for causing a computer to execute a topic word extraction method in a topic word extraction device for extracting a topic word,
A first step in which the communication information acquisition means acquires communication information for a predetermined period from one or more communication information storage servers storing the communication information;
A second step in which the morphological analysis means performs a morphological analysis on text information included in the communication information acquired in the first step, and extracts a word;
A third step in which the importance calculation means calculates importance for each word extracted by the morpheme analysis means;
The unique user number acquisition means refers to the communication information acquired in the first step and transmits communication information including each word in the text information for each word extracted in the third step. A fourth step of acquiring the number of unique users;
The weighting coefficient calculating means calculates a weighting coefficient that takes into account a unique user based on the number of unique users acquired in the fourth step and the number of communication information acquired in the first step. 5 steps,
The topic word extraction unit is configured to extract the word extracted in the third step based on the importance for each word calculated in the third step and the weighting coefficient calculated in the fifth step. A sixth step of extracting topic words;
A program that causes a computer to execute.

Images