JP2009251957A - Interest information specification system, interest information specification method, and program for interest information specification - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide an interest information specification system for specifying feature words which appear at random in a certain period as interest information. <P>SOLUTION: An appearance frequency calculation means 971 calculates a feature word appearance frequency as the use frequency of a content shown by feature words for each feature word by referring to a feature word history including feature words showing features of the content used by a person or a group and a use time when each content whose features are shown by the feature words by the person or the group. An appearance interval calculation means 972 calculates an appearance time interval as the use time interval of content shown by the feature words for each feature word by referring to the feature word history. A feature word evaluation means 973 refers to the feature word appearance frequency and the appearance time interval, and calculates the evaluation value of the feature words according to the dissociation of the distribution of the appearance time interval and the probability distribution of a model for each feature word. A feature word specification means 974 specifies the feature words based on the feature value. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to an interest information identification system, an interest information identification method, and an interest information identification program, and more particularly to an interest information identification system, an interest information identification method, and an interest information identification program for identifying a regular interest of a person or a group. .

  In a system for recommending content such as documents, music, and videos, there is a system that supports the recommendation or search of content suitable for a user by capturing the user's interest based on the user's past browsing history and viewing history of the content. .

  In such a system, information representing the user's interest is specified using the number of times each content is browsed by the user and a feature word representing a feature of the content browsed by the user.

  For example, consider a case where there is a document 1 whose feature words are “proposal” and “network”, and a document 2 whose feature words are “proposal” and “security”. If a user A views document 1 twice and document 2 three times, it is considered that “suggestion” appears five times, “network” appears twice, and “security” appears three times. . Assuming that the number of appearances represents the strength of interest in each feature word of user A, the strengths of interest in “proposal”, “network”, and “security” are “5”, “2”, “3” Can be expressed. From the strength of this interest, it can be seen that the user A is more interested in the “proposal” than the “network” and “security”.

  Furthermore, an information recommendation method has been proposed that makes it possible to recommend content in consideration of user interest transitions (see, for example, Patent Document 1). In the method described in Patent Document 1, short-term item-oriented information of a user is calculated based on the user's operation history and content metadata information, and temporal interest transition is considered based on the information. The user item-oriented information is calculated. Then, based on the information and the metadata information of the content, the content recommendation considering the temporal interest transition of the user is performed.

  Patent Document 2 describes an information processing apparatus that generates preference information for short-term and long-term users. In the device described in Patent Document 2, whether the update of the preference information data indicating the user's preference proceeds rapidly or slowly is determined by the preference element parameter. The apparatus described in Patent Literature 2 calculates a preference addition value as a product of an operation preference value parameter and a preference element parameter, and adds the preference addition value to the preference value. As a result, if the preference element parameter is set to be small, the preference value does not increase so much by one viewing and operation, and long-term preference can be expressed. In addition, by increasing the preference element parameter, the preference value is greatly increased by a single viewing and operation, and thus a short-term preference can be expressed.

  Patent Document 3 describes a latent needs inference device that provides information that a user is potentially interested in as recommendation information using a keyword obtained by a user's operation. In the apparatus described in Patent Document 3, keywords having a strong overall characteristic of “high occurrence frequency”, “large difference between maximum and minimum occurrence times”, and “small standard deviation of occurrence interval” Identify.

  Non-Patent Document 1 describes that the exponential distribution is known as the distribution of time (waiting time) until an event determined under a certain condition occurs.

  Patent Document 4 discloses an image search system that calculates a similarity score for an input image of a registered image by comparing the distribution of the image feature amount of the input document image with the distribution of the image feature amount of the registered image. Are listed.

Japanese Patent Laying-Open No. 2005-208896 (paragraph 0028) JP 2005-86472 A (paragraphs 0060, 0061, 0074-0080) Republished patent WO01 / 039118 (6th page, 13th page) JP 2007-172077 A (paragraphs 0021, 0023) Matsubara Nozomi, “Introductory Stochastic Process”, 5th edition, Tokyo Book Co., Ltd., May 10, 2007, p. 46

  It is not possible to specify the regular interest of a person or group simply by counting the number of appearances of feature words representing the features of the content according to the number of times of browsing the content. For example, it is assumed that a certain user is regularly interested in matters represented by a characteristic word “B”. Then, it is assumed that the user browses a large number of documents having the characteristic word “A” only for the purpose of investigation during a certain period, and then stops browsing the documents having the characteristic word “A”. In this case, the number of appearances of the feature word “B” exceeds the number of appearances of the feature word “A” even though the document related to the feature word “A” is temporarily browsed for the purpose of investigation. During the period, it is determined that the user has a strong interest in the feature word “A”.

  Here, the constant interest is not an interest that occurs only for a short period of time, but an interest that a person or group has for a long period of time.

  In the invention described in Patent Document 3, a keyword having a strong overall characteristic of “high occurrence frequency”, “large difference between maximum and minimum occurrence times”, and “small standard deviation of occurrence interval” By identifying, the user's potential interest is estimated. However, as a typical appearance mode of feature words indicating a constant interest, there is an appearance mode of “occurring randomly within a certain period”, and it is preferable that such feature words can be extracted more effectively.

  In addition, when extracting feature words with properties such as “the difference between the maximum and minimum occurrence times is large” and “the standard deviation of the occurrence interval is small”, the feature words are taken into account when considering the length of the appearance interval. More preferably, can be extracted. For example, if it is a feature word that is extracted specially for a feature word that appears every day for a long period of time, or a feature word that repeats every few days even if it is not every day, It is preferable that information indicating interest can be extracted.

  Therefore, an object of the present invention is to provide an interest information specifying system, an interest information specifying method, and an interest information specifying program that can specify feature words that appear randomly within a certain period as interest information.

  Further, according to the present invention, when specifying a feature word having a large difference between the maximum value and the minimum value of the occurrence time and a small standard deviation of the occurrence interval, the feature word may be specified in consideration of the length of the appearance interval. An object of the present invention is to provide an interest information identification system, an interest information identification method, and an interest information identification program.

  The interest information identification system of the present invention is an interest information identification system for identifying interest information representing the interest of a person or group, wherein the feature word representing the characteristics of the content used by the person or group, Appearance frequency calculation means for obtaining, for each feature word, a feature word appearance frequency, which is the use frequency of the content represented by the feature word, with reference to a feature word history including the use time of each content whose feature is represented by the word And, referring to the feature word history, for each feature word, refer to the appearance interval calculation means for obtaining the appearance time interval that is the use time interval of the content represented by the feature word, the feature word appearance frequency and the appearance time interval, For each word, feature word evaluation means that obtains the evaluation value of the feature word according to the difference between the distribution of the appearance time intervals and the probability distribution as a model, and the feature word based on the evaluation value Characterized in that it comprises a feature word specifying means for.

The interest information identification system of the present invention is an interest information identification system for identifying interest information representing the interest of a person or group, wherein a feature word representing the characteristics of content used by the person or group, and a person or group Appearance frequency for obtaining the feature word appearance frequency, which is the use frequency of the content represented by the feature word, for each feature word with reference to the feature word history including the use time of each content whose feature is represented by the feature word A calculation means, an appearance interval calculation means for obtaining an appearance time interval that is a use time interval of content represented by the feature word for each feature word with reference to the feature word history, and a standard deviation of the appearance time interval for each feature word The average value is calculated, the standard deviation is STDEV, the average value is AVE, and the first use time and the last use time of the content represented by the feature word are respectively T {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} is calculated by using the parameter β, where _last and T ₀ are T and the characteristic word history derivation target period is T. Thus, a feature word evaluation unit that obtains an evaluation value of a feature word, sets the evaluation value of the feature word that is equal to or less than a predetermined number of times the feature word appears, and a feature word that identifies the feature word based on the evaluation value And a word specifying means.

  The interest information specifying method of the present invention is an interest information specifying method for specifying interest information representing the interest of a person or group, wherein a feature word representing a feature of content used by the person or group and a person or group Appearance frequency for obtaining the feature word appearance frequency, which is the use frequency of the content represented by the feature word, for each feature word with reference to the feature word history including the use time of each content whose feature is represented by the feature word Referring to the calculation step, the feature word history, the appearance interval calculation step for obtaining the appearance time interval that is the use time interval of the content represented by the feature word for each feature word, and the feature word appearance frequency and the appearance time interval For each feature word, a feature word evaluation step for obtaining an evaluation value of the feature word in accordance with a deviation between the distribution of appearance time intervals and the model probability distribution, and based on the evaluation value Characterized in that it comprises a characteristic word specifying step of specifying the symptoms word.

The interest information specifying method of the present invention is an interest information specifying method for specifying interest information representing the interest of a person or group, wherein a feature word representing a feature of content used by the person or group and a person or group Appearance frequency for obtaining the feature word appearance frequency, which is the use frequency of the content represented by the feature word, for each feature word with reference to the feature word history including the use time of each content whose feature is represented by the feature word A calculation step, an appearance interval calculation step for obtaining an appearance time interval that is a use time interval of content represented by the feature word for each feature word with reference to the feature word history, and a standard deviation of the appearance time interval for each feature word The average value is calculated, the standard deviation is STDEV, the average value is AVE, and the first use time and the last use time of the content represented by the feature word are respectively T {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} is calculated by using the parameter β, where _last and T ₀ are T and the characteristic word history derivation target period is T. Thus, a feature word evaluation step for obtaining an evaluation value of a feature word, setting the evaluation value of the feature word that is equal to or less than a predetermined number of times of appearance of the feature word to a predetermined value, and a feature word specifying the feature word based on the evaluation value A word specifying step.

  The interest information specifying program of the present invention is an interest information specifying program mounted on a computer for specifying interest information representing the interest of a person or group, and the feature of the content used by the person or group on the computer. Frequency of use of the content represented by the feature word for each feature word with reference to the feature word history including the feature word representing the character and the use time when each person or group represented the feature by the feature word. Appearance frequency calculation processing for obtaining a feature word appearance frequency, appearance frequency calculation processing for obtaining an appearance time interval that is a use time interval of content represented by the feature word for each feature word with reference to the feature word history, and feature word appearance Refer to the frequency and appearance time interval, and for each feature word, evaluate the evaluation value of the feature word according to the difference between the distribution of the appearance time interval and the model probability distribution. Mel characteristic word evaluation processing, and characterized in that to execute the characteristic word specifying process for specifying a characteristic word based on the evaluation value.

The interest information specifying program of the present invention is an interest information specifying program mounted on a computer for specifying interest information representing the interest of a person or group, and the feature of the content used by the person or group on the computer. Frequency of use of the content represented by the feature word for each feature word with reference to the feature word history including the feature word representing the character and the use time when each person or group represented the feature by the feature word. Appearance frequency calculation process for obtaining a feature word appearance frequency, and an appearance interval calculation process for obtaining an appearance time interval that is a use time interval of content represented by the feature word for each feature word with reference to the feature word history, for each feature word Then, the standard deviation and the average value of the appearance time intervals are calculated, the standard deviation is STDEV, the average value is AVE, and the content of the content represented by the feature word is calculated. Each _T last use time and the end of use time of the _{T 0,} the derivation period of characteristic word history when is T, by using the parameter _{β, {(T last -T 0} ) / T} · e By calculating ^{(−β · STDEV · AVE)} , a feature word evaluation value is obtained, and a feature word evaluation value in which the feature word appearance frequency is equal to or less than a predetermined number of times is determined, and And a feature word specifying process for specifying a feature word based on the evaluation value.

  According to the present invention, feature words that appear randomly within a certain period can be specified as interest information.

  Further, according to the present invention, the evaluation value can be determined such that the difference between the maximum value and the minimum value of the occurrence time is large, and the value is increased as the standard deviation and the average value of the appearance time interval are smaller. It can be adjusted whether or not feature words with a short time interval are easily extracted with priority.

  Embodiments of the present invention will be described below with reference to the drawings.

Embodiment 1. FIG.
FIG. 1 is a block diagram illustrating an example of an interest information specifying system according to the first embodiment of this invention. The interest information identification system according to the first embodiment includes an extraction target setting unit 100, an access history storage unit 200, a content management unit 300, a feature word history generation unit 400, an appearance frequency calculation unit 500, and an appearance interval calculation. Means 600, feature word evaluation means 700, feature word selection means 800, and interest information presentation means 900 are provided.

  First, the access history and meta information stored in advance by the interest information identification system of the present invention and the feature word history generated from the information will be described.

  The interest information specifying system stores an access history and meta information of each content in advance.

  The access history is a usage history including content identification information, use time of the content, and identification information of a person who uses the content (hereinafter referred to as a user) or a group. In the access history, content identification information, content use time, and user or group identification information using the content are associated with each other. Note that the usage mode of the content is not particularly limited, and the usage time of the content may be any of the time when the user or group browses the content, the viewing time, the downloading time, and the like. Further, the unit of content use time is not particularly limited. For example, the use time may be expressed in hours, minutes and seconds as a second unit, or may be expressed in dates as a day unit. Alternatively, it may be expressed in hours or weeks. Further, the identification information of the user and the group to which the user belongs may be associated with the identification information of the content and the use time. Hereinafter, a case where a user name is used as user identification information and a group name (for example, a department name) is used as group identification information will be described as an example.

  FIG. 2 is an explanatory diagram illustrating an example of an access history. In the access history illustrated in FIG. 2, the date (use time), the document ID (content identification information), the user name, and the user's department name are associated with each other. For example, in the access history illustrated in FIG. 2, the data in the first row is “September 1, 2007” and the user “USER1” who belongs to the department “SECTION1” downloaded the document with the document ID “ID001”. It is shown that.

  Content meta information is additional information of content and is prepared for each content. The meta information includes content identification information and a feature word representing the feature of the content, and the content identification information and the feature word are associated with each other. Further, the meta information may include other information (for example, content name, content creator, creation date and time) in addition to content identification information and feature words. FIG. 3 is an explanatory diagram illustrating an example of meta information. In the meta information illustrated in FIG. 3, a document ID (content identification information), a document name (content name), and a feature word are associated with each other. For example, in the meta information of the document ID “ID001” illustrated in FIG. 3, the document name of the document ID “ID001” is “XX proposed material”, and “security, ubiquitous, network” is a characteristic word for the document. It shows that it is stipulated.

  The feature word history includes a feature word that represents a feature of content used by a certain user (or group), and a use time at which each user whose feature is represented by the feature word is used by the user (or the group). It is information to include. In the feature word history, feature words and usage times are associated with each other. FIG. 4 is an explanatory diagram illustrating an example of a feature word history. In the example illustrated in FIG. 4, for example, content having “security” as a feature word is “2007/09/01”, “2007/09/01”, “2007/09/02” by a certain user (or group). ”Indicates that it was used. The feature word history is generated from the access history and meta information.

  The interest information identification system of the present invention generates a feature word history from the access history and meta information, and calculates an evaluation value representing the degree of steady interest of the user or group for each feature word. FIG. 5 is an explanatory diagram illustrating an example of the evaluation value of each feature word. In the example illustrated in FIG. 5, the evaluation value for the feature word “security” is “0.1”, and the evaluation value for the feature word “ubiquitous” is “1.0”. If the evaluation value is larger, the user's interest is stronger. This indicates that a user (or a group) is most interested in the “network”.

  An example of content is a document (electronic document), but the content is not limited to a document. For example, the content may be a WEB page such as a home page or an electronic bulletin board, location-related information such as store information or sightseeing information, television broadcast or radio broadcast program information, video content, music content, book information, or the like. Hereinafter, a case where the content is a document (electronic document) will be described as an example.

Next, each component of this embodiment is demonstrated.
The extraction target setting unit 100 sets a user or group to be a target for specifying a feature word as interest information representing a regular interest. For example, the extraction target setting unit 100 outputs a screen that prompts the user to enter a user name or group name indicating the user or group, and when the user name or group name is input to the screen, the extraction target setting unit 100 specifies the user name or group name. The user or group to be played may be determined as a specific target of constant interest. FIG. 6 is an example of a user name or group name input screen. The screen illustrated in FIG. 6 includes a radio button 401 for designating either user name input or group name input, and an input field 402 for inputting a user name or group name. The extraction target setting unit 100 displays the input screen illustrated in FIG. 6. When either the user name input or the group name input is designated and the name is input to the input field 402, the input name is displayed. Determine as user name or group name.

  Alternatively, the extraction target setting unit 100 may use a cookie or the like to determine a user or group specified by a previously input user name or group name as a target of constant interest or from another system. The user name or group name may be automatically taken over, and the user or group specified by the user name or group name may be determined as the target of regular interest.

  The access history storage unit 200 stores an access history. For example, when the document (content) stored in the content management unit 300 is accessed (that is, used), the access history storage unit 200 uses the document identification information, the use time, and the document. It is only necessary to associate the user name of the user and add it as an access history. As already described, the group name to which the user belongs may also be included in the access history. The manner in which the access history storage unit 200 acquires the access history is not particularly limited. For example, an access history created externally may be input to the access history storage unit 200, and the access history storage unit 200 may store the access history.

  Further, when a user name or group name is designated, the access history storage means 200 accesses the document identification information (hereinafter referred to as document ID) and the use time corresponding to the designated user name or group name. Search from history.

  The content management unit 300 stores a document (content) in association with meta information of the document. The meta information includes a document ID and a feature word representing the feature of the document. The meta information may include other information such as a document name. The document name may be used as the document ID. Further, the content management unit 300 may search for a document based on a document ID, a document name, or the like.

  The feature word history generation unit 400 refers to the access history stored in the access history storage unit 200 and the meta information of the document stored in the content management unit 300 to generate a feature word history. The feature word history generation unit 400 specifies a feature word representing the feature of the content used by the user or group set in the extraction target setting unit 100, and performs processing for associating the use time of the content with the feature word Thus, a feature word history is generated.

  For example, it is assumed that the access history storage unit 200 stores the access history illustrated in FIG. 2, and the content management unit 300 stores the meta information illustrated in FIG. The access history illustrated in FIG. 2 is an access history during the period from September 1, 2007 to September 7, 2007. Further, it is assumed that the extraction target setting unit 100 sets “USER1”. An example of feature word history generation processing in this case will be described.

  First, the feature word history generation unit 400 acquires the document ID and the use time corresponding to the user name or group name set in the extraction target setting unit 100. For example, the access history storage unit 200 is searched. In this example, the feature word history generation unit 400 acquires “ID001”, “ID002”, “ID003”, and “ID005” as document IDs corresponding to “USER1” (see FIG. 2). In addition, “2007/09/01” is acquired as the use time of the document of “ID001”. The use times of other documents are acquired in the same manner.

  Further, the feature word history generation unit 400 extracts a feature word associated with the document ID from the meta information for each document ID. Then, the feature word history generation unit 400 associates the feature word with the use time associated with the same document ID. For example, the feature word history generation unit 400 extracts the feature word “security, ubiquitous, network” corresponding to “ID001” from the meta information stored in the content management unit 300 for the document ID “ID001”. Each feature word is associated with the use time “2007/09/01” associated with “ID001”. That is, “2007/09/01” is associated with “security”, “ubiquitous”, and “network”. Similar processing is performed for the other document IDs “ID002”, “ID003”, and “ID005”. FIG. 4 shows the feature word history obtained as a result of the above processing. Since the access history is a history in the period from September 1, 2007 to September 7, 2007, the derivation target period of the feature word history is from September 1, 2007 to September 7, 2007. It is a period.

  The appearance frequency calculation unit 500 refers to the feature word history generated by the feature word history generation unit 400 and obtains the feature word appearance frequency of each feature word. The feature word appearance frequency is the use frequency (number of uses) of each content whose feature is represented by the feature word. The appearance frequency calculation means 500 may refer to the feature word history, count the number of times of use associated with the feature word for each feature word, and use the count result as the feature word appearance frequency. Hereinafter, the feature word appearance frequency is simply referred to as the appearance frequency. For example, assume that the feature word history illustrated in FIG. 4 is generated. In this case, the appearance frequency of “security” is three times, and the appearance frequency of “ubiquitous” is two times. FIG. 7 shows the appearance frequency derived based on the feature word history exemplified in FIG.

  The appearance interval calculation unit 600 refers to the feature word history generated by the feature word history generation unit 400 to obtain the appearance time interval of each feature word. The appearance time interval is an interval between use times of each content whose features are represented by feature words. The appearance interval calculation unit 600 may calculate the difference between the usage times associated with the feature words for each feature word with reference to the feature word history. Further, the appearance interval calculation means 600 counts the number of times each appearance time interval has occurred. For example, assume that the feature word history illustrated in FIG. 4 is generated. “Security” appears twice in “2007/09/01” and once in “2007/09/02” (see FIG. 4). In this case, the appearance interval calculation means 600 calculates the appearance time interval of “0 day interval” as the difference between “2007/09/01”, and counts the number of times “0 day” has occurred “1”. To do. In addition, as a difference between “2007/09/01” and “2007/09/02”, an appearance time interval of “1 day interval” is calculated, and the number of times “1 day interval” occurs is counted as “1 time”. To do. Here, the appearance time interval of “security” is illustrated, but the appearance time interval is similarly obtained for other feature words. Further, when only one use time is associated with the feature word, the appearance interval calculation unit 600 sets the count number to 0 for all the appearance time intervals. For example, in the case of “server” illustrated in FIG. 4, the appearance time intervals such as “0 day interval” and “1 day interval” are set to “0”. FIG. 8 shows the appearance time intervals derived based on the feature word history exemplified in FIG.

  The feature word evaluation unit 700 refers to the feature word history generated by the feature word history generation unit 400 and the appearance time interval of the feature word obtained by the appearance interval calculation unit 600, and calculates an evaluation value for each feature word. This evaluation value is a value representing the degree of steady interest of the set user or group. The feature word evaluation means 700 calculates the difference between the distribution of appearance time intervals and the probability distribution as a model, and calculates the evaluation value of the feature word according to the difference. Here, the model probability distribution is the probability distribution of the appearance time intervals when it is assumed that the feature words appear uniformly (that is, the document whose features are represented by the feature words are used randomly). is there. The feature word evaluation unit 700 sequentially selects each feature word and calculates an evaluation value for the selected feature word.

  It is known that when an event occurs at random, the time from when an event occurs to the next event follows an exponential distribution. Therefore, the appearance time interval when it is assumed that the feature words appear uniformly follows the exponential distribution. Therefore, in the present embodiment, an exponential distribution is used as a probability distribution serving as a model.

  The probability that a feature word will appear at a certain appearance time interval (that is, the probability that a document whose feature is represented by the feature word will be used at a certain appearance time interval) is the number of times the feature word has actually appeared at that appearance time interval. , The value divided by the sum of the number of appearances of the feature word at each appearance time interval. That is, when the appearance time interval is t, the probability that a feature word appears at the appearance time interval t (denoted as P ′ (t)) is expressed by the following equation (1).

  F (t) shown in Equation (1) is the number of times a feature word has appeared at the appearance time interval t, and the denominator on the right side of Equation (1) is the sum of the number of appearances of the feature word at each appearance time interval. is there.

  Further, in the probability distribution (exponential distribution) serving as a model, if the probability that a feature word appears at an appearance time interval t is P (t), P (t) is expressed by the following equation (2).

P (t) = (K / T) e− ^{(K / T) t} formula (2)

  Here, K is the appearance frequency of the selected feature word. T is a feature word history derivation target period (in other words, a period during which an access history that is a basis for generating a feature word history is collected).

  The feature word evaluation unit 700 calculates P ′ (t) according to the expression (1) and P (t) according to the expression (2) for each appearance time interval of the selected feature word, and the difference between them. The absolute value | P (t) −P ′ (t) | The feature word evaluation unit 700 obtains the sum of absolute values | P (t) −P ′ (t) | of the differences calculated at each appearance time interval, and calculates an evaluation value corresponding to the sum. In the present embodiment, the feature word evaluation unit 700 calculates the evaluation value of the selected feature word by calculating the following equation (3).

  V on the left side of Equation (3) is the evaluation value of the selected feature word. In addition, the exponent part on the right side of Equation (3) is a value obtained by multiplying the sum of the absolute values | P (t) −P ′ (t) | When the evaluation value is calculated in this way, the evaluation value V increases as the sum of | P (t) −P ′ (t) | decreases. That is, the smaller the difference between the appearance time interval distribution and the model probability distribution, the larger the evaluation value V.

  In addition, an evaluation value is set to 0 for a feature word whose appearance frequency is once and whose appearance time interval cannot be obtained. That is, when the number of occurrences for any t is 0, V = 0 is determined.

  A specific example of processing of the feature word evaluation unit 700 will be shown. Assume that the feature word history generation unit 400 generates a feature word history illustrated in FIG. 9A as a feature word history between September 1 and September 14 for a certain feature word A. In this case, the appearance frequency calculation means 500 obtains the appearance frequency “9” of the feature word A as shown in FIG. Further, the appearance interval calculation means 600 calculates an appearance time interval and counts the number of times that the appearance time interval has occurred. For example, in the feature word history shown in FIG. 9A, the appearance time interval 0 day occurs three times, and the appearance time interval 1 day occurs twice. In this example, the appearance time interval shown in FIG. 9C is obtained.

  The feature word evaluation unit 700 obtains the absolute value of the difference between the actual probability of each appearance time interval and the probability in the model probability distribution, and calculates the evaluation value using the sum. FIG. 10 is an explanatory diagram showing this calculation process. In this example, the appearance frequency K of the feature word A is 9. Further, since the feature word history is obtained for the period from September 1 to September 14, the period T = 14 in the equation (2). Therefore, K / T = 0.642857.

  Taking the appearance time interval 0 days as an example, the probability that a feature word will actually appear at the appearance time interval is obtained. That is, assuming that t = 0, P ′ (0) is obtained from Equation (1). As shown in FIG. 9C, the number of occurrences f (0) = 3 that occurred on the appearance time interval 0 days. Further, the sum of the number of appearances of feature words at each appearance time interval is 3 + 2 + 1 + 1 + 1 = 8 (see FIG. 9C). Therefore, P ′ (0) = 3/8 = 0.375.

In the probability distribution (exponential distribution) serving as a model, the probability P (0) of occurrence of an appearance time interval of 0 days is 0.642857 × e− ^{0.642857 × 0} = 0. 642857. Therefore, the absolute value of the difference between the actual probability P ′ (0) at the appearance time interval 0 days and the probability P (0) in the model probability distribution is | 0.642857−0.375 | = 0. 268. Similarly, when | P (t) −P ′ (t) | is calculated for other appearance time intervals t, the result is as shown in FIG. When this sum is calculated, it is 0.570. FIG. 10 also shows P ′ (t) and P (t) calculated in the process of obtaining the absolute value of this difference.

The feature word evaluation means 700 calculates the evaluation value V of the feature word A by calculating Equation (3) using this sum. In this example, V = e− ^0.570 = 0.565. Although the feature word A has been described as an example here, evaluation values are similarly calculated for other feature words.

  Here, the example shown in FIG. 9 is used for explanation. However, when the appearance frequency and the appearance time interval shown in FIGS. 7 and 8 are obtained from the feature word history shown in FIG. 4 and the evaluation value of each feature word is obtained, FIG. 11 as shown. Since “server” and “storage” have an appearance frequency of 1 and an appearance time interval cannot be obtained, the evaluation value is set to 0.

  In this embodiment, an exponential distribution is used as a model probability distribution. However, the model probability distribution is not limited to the exponential distribution, and represents a distribution of appearance time intervals when feature words appear randomly. If so, a probability distribution other than the exponential distribution may be used as a model.

  The feature word selection unit 800 specifies a feature word based on the evaluation value of each feature word obtained by the feature word evaluation unit 700. In this embodiment, feature words having an evaluation value equal to or greater than a threshold value are selected. The feature word specified in this way is interest information indicating the steady interest of the designated user or group, and the feature word selection unit 800 selects the feature word based on the evaluation value obtained as described above. By specifying, the interest information indicating stationary interest is specified. For example, if the threshold value is 0.3 and the evaluation value of each feature word is obtained as shown in FIG. 11, the feature word selection unit 800 displays “security” as a feature word representing stationary interest. And select Network.

  Here, the case where the feature word is specified using the threshold value has been described, but the feature word selection unit 800 may specify the feature word by another method. For example, the feature word selection unit 800 may sort the feature words so that the evaluation values are in descending order, and select a predetermined number of higher-order feature words having a large evaluation value.

  The interest information presentation unit 900 presents the feature word specified by the feature word selection unit 800 to the user. For example, the interest information presentation unit 900 may display the feature word on the portal site. That is, on the top screen of the portal site, the feature word selected by the feature word selection unit 800 may be presented as a “recommended search keyword”. Moreover, the interest information identification system of this invention may be provided with the search means 950 which searches a content using a feature word in addition to said each means 100-900, as shown in FIG. Then, the search unit 950 may search the content using the feature word specified by the feature word selection unit 800 as a search word, and the interest information presentation unit 900 may display the search result together.

  FIG. 13 is an explanatory diagram illustrating an example of a screen output by the interest information presenting unit 900. As shown in FIG. 13A, on the portal site, the feature word selected by the feature word selection unit 800 is displayed as “recommended search keyword”, and the search result searched by the search unit 950 (shown in FIG. 13A). In the example, “Recommended News”) may also be displayed. The portal site illustrated in FIG. 13A is used, for example, when presenting a regular interest of a person who logs in at the time of login or the like. Moreover, you may use this invention for the use which investigates the regular interest of others and other groups (for example, other departments) instead of the regular interest of the principal. For example, a certain person can input “USER1”, which is another person, into the extraction target setting unit 100 and examine the steady interest of “USER1”. FIG. 13B shows an example of the feature word output screen in this case. Other information may also be displayed on each screen shown in FIG.

  12 may search for content stored in the content management unit 300, or may be a content database outside the interest information identification system. Alternatively, various Web pages may be searched.

  When the extraction target setting unit 100 sets a group (for example, a department) as a target for specifying a feature word as interest information, the interest information specifying system sets each feature word for each user belonging to the group. An evaluation value may be calculated, and the evaluation value of each user may be added for each feature word. Then, the result may be an evaluation value of the feature word in the group. Alternatively, the feature word history of the group may be collectively generated from the access histories of all users belonging to the group, and the evaluation value of the feature word in the group may be calculated.

  Extraction target setting means 100, access history storage means 200, content management means 300, feature word history generation means 400, appearance frequency calculation means 500, appearance interval calculation means 600, feature word evaluation means 700, feature word selection means 800, interest information The presentation unit 900 and the search unit 950 may be realized by a CPU that operates according to a program (interest information specifying program), for example. And each said means may be implement | achieved by the same CPU. For example, the program is stored in a storage device included in the interest information identification system, and the CPU reads the program, and in accordance with the program, the extraction target setting unit 100, the access history storage unit 200, the content management unit 300, and the feature word history generation unit 400. , Appearance frequency calculation means 500, appearance interval calculation means 600, feature word evaluation means 700, feature word selection means 800, interest information presentation means 900, and search means 950. The extraction mode setting unit 100 may be realized by the CPU and an input device such as a keyboard. The access history storage unit 200 and the content management unit 300 are realized by a CPU and a storage device. The interest information presentation unit 900 is realized by a CPU and a display device.

  Here, the case where each unit is realized by the same computer is illustrated, but the configuration of the interest information specifying system of the present invention is not limited to the case where it is realized by one computer. An example is shown below.

  The extraction target setting unit 100 and the interest information presentation unit 900 may be realized by an information processing device such as a PDA (Personal Data Assistants), a personal computer, and a mobile phone, which includes a display device and an input device and operates according to a program. .

  The access history storage unit 200 includes a storage device that stores an access history, and may be realized by a personal computer or a server type computer that operates according to a database program. The content management means 300 may be realized by a personal computer or a server type computer that includes a storage device that stores a document and meta information in association with each other and operates according to a database program.

  The feature word history generation unit 400, the appearance frequency calculation unit 500, the appearance interval calculation unit 600, the feature word evaluation unit 700, the feature word selection unit 800, and the search unit 950 may be realized by the same computer. Alternatively, it may be realized by a different computer, and processing may be performed by performing communication using a communication protocol such as TCP / IP (Transmission Control Protocol / Internet Protocol).

  Next, the operation will be described. FIG. 14 is a flowchart illustrating an example of processing progress of the interest information identification system according to the first embodiment.

  The extraction target setting unit 100 sets a user or a group that is a target for specifying stationary interest information (step S1). Then, the feature word history generation unit 400 uses the access history stored in the access history storage unit 200 and the meta information stored in the content management unit 300 to relate to the user or group set in step S1. A feature word history to be generated is generated (step S2). The appearance frequency calculation unit 500 refers to the feature word history to obtain the appearance frequency of each feature word (step S3), and the appearance interval calculation unit 600 refers to the feature word history and the appearance of each feature word. A time interval is obtained (step S4).

  Subsequently, the feature word evaluation unit 700 refers to the appearance time interval obtained in step S4, and calculates the actual probability that the feature word appears at each appearance time interval (step S5). That is, for each appearance time interval, the equation (1) is calculated, and the probability P ′ (t) for each appearance time interval is calculated.

  The feature word evaluation unit 700 refers to the appearance frequency obtained in step S3 and the appearance time interval obtained in step S4, and calculates the probability that the feature word appears at each appearance time interval in the model probability distribution (exponential distribution). Calculate (step S6). For example, K / T is calculated from the appearance frequency K and the derivation target period T of the feature word history, the equation (2) is calculated for each appearance time interval, and the probability P ( t) is calculated.

  Subsequently, the feature word evaluation unit 700 calculates, for each appearance time interval, a deviation amount (absolute value of the difference) between the actual appearance probability obtained in step S5 and the appearance probability on the exponential distribution obtained in step S6. Calculate (step S7). Further, the feature word evaluation unit 700 calculates the sum of the deviation amounts calculated for each appearance time interval in step S7 (step S8), and calculates the evaluation value of the feature word using the sum (step S9). In step S9, the evaluation value V may be calculated by calculating equation (3).

  The feature word evaluation unit 700 performs the processing of steps S5 to S9 for each feature word. In each of steps S5 to S9, feature words may be sequentially selected, and processing may be performed for each selected feature word. Further, before executing the process of step S5, the feature word is selected, the process of steps S5 to S9 is performed on the selected feature word, the next feature word is further selected, and the process of steps S5 to S9 is performed in the same manner. You may go.

  After calculating the evaluation values for all the feature words, the feature word selecting unit 800 specifies the feature words based on the evaluation values, thereby indicating the interest information indicating the steady interest of the user or group set in step S1. Is specified (step S10). In step S10, for example, a feature word whose evaluation value is greater than or equal to a predetermined threshold value may be selected.

  The interest information presentation unit 900 displays the feature word specified in step S10 (step S11). As a result, the user of the interest information specifying system can know the steady interest of the set user or group.

  A specific example of the above process will be shown below. In the following example, it is assumed that the interest information identification system is realized by a personal computer including an input device such as a mouse and a keyboard, a user interface such as a button, and a display device that displays characters. Further, it is assumed that the access history storage unit 200 and the content management unit 300 are operated by a database program. Further, it is assumed that the access history storage unit 200 stores the access history shown in FIG. In the access history shown in FIG. 2, the date (usage time) when the user browsed or downloaded the document, the document ID, the user name, and the department name (group name) are associated with each other. It is assumed that the person who stores the meta information shown in FIG. In the meta information shown in FIG. 3, the document ID, the document name, and the feature word are associated with each other.

  In step S1 (see FIG. 14), the extraction target setting unit 100 displays the input screen illustrated in FIG. 6 and prompts the user name or department name to be input. In this example, it is assumed that “USER1” is input. The extraction target setting unit 100 determines the input “USER1” as an interest information identification target.

  In step S2, the feature word history generation unit 400 acquires the date and document ID corresponding to “USER1” from the access history storage unit 200, acquires the feature word corresponding to the document ID from the document management unit 300, and A feature word history is generated by associating a feature word with a date. For example, there are “2007/09/01” and “ID001” as the date and document ID corresponding to “USER1” (see FIG. 2). The feature word history generation unit 400 acquires the feature words “security”, “ubiquitous”, and “network” corresponding to the document ID “ID001” from the document management unit 300, and sets “2007/09 / 01 ”is associated. The feature word history generation unit 400 performs this process for all document IDs associated with “USER1” in the access history. As a result, the feature word history shown in FIG. 4 is generated.

  In the next step S3, the appearance frequency calculation means 500 obtains the appearance frequency of each feature word using the generated feature word history. For example, since “security” appears in “2007/09/01”, “2007/09/01”, and “2007/09/02”, the appearance frequency is set to 3 times. Appearance frequencies are also found for other feature words. As a result, the appearance frequency shown in FIG. 7 is obtained.

  In step S4, the appearance interval calculation means 600 obtains the appearance time interval of each feature word using the generated feature word history, and counts the number of times each appearance time interval has occurred. For example, regarding “security”, since it appears in “2007/09/01”, “2007/09/01”, and “2007/09/02”, the appearance time interval 0 day is once, the appearance time interval One day is once. Similar processing is performed for other feature words. As a result, the result shown in FIG. 8 is obtained. If there is only one date when the document was used, the number of occurrences of any appearance time interval is set to zero.

  In step S5 to S9, the feature word evaluation unit 700 calculates the evaluation value of the feature word in the feature word history shown in FIG. 4 using the appearance frequency obtained in step S3 and the appearance time interval obtained in step S4. . In the following, a specific example of the evaluation value calculation process will be described by taking as an example the case of obtaining the evaluation value of the feature word “security”. FIG. 15 is an explanatory diagram showing the process of calculating the evaluation value.

  In step S5, the feature word evaluation unit 700 uses the appearance time intervals shown in FIG. 8 to calculate the actual probability P ′ (t) that the feature word appears at the appearance time interval t for each appearance time interval. . In addition, the number of occurrences of the appearance time interval t being 0 is the number of times the document has been used at the appearance time interval t, and in this case, P ′ (t) = 0. With regard to “security” shown in FIG. 8, the 0-day interval (t = 0) is 1 time and the 1-day interval (t = 1) is 1 time, so P ′ (0) and P ′ (1) Are 0.5 respectively.

In step S6, the feature word evaluation means 700 uses the target period T of the feature word history and the appearance frequency K, and the probability P that the feature word appears at the appearance time interval t in the model probability distribution (exponential distribution). (T) is calculated for each appearance time interval. In this example, since the feature word history is generated from the access history for the period from September 1, 2007 to September 7, 2007, T = 7. The period T is obtained by, for example, recording the start and end of creation of the access history in the access history storage unit 200, and calculating the period from the start to the end of the feature word evaluation unit 700. May be. Alternatively, the period T may be input from the outside. Since the appearance frequency K of “security” is 3 and T = 7, the feature word evaluation unit 700 calculates K / T = 0.4286. Further, the feature word evaluation unit 700 calculates the equation (2) using this value to obtain P (t). For example, in the case of an interval of 0 days (t = 0), P (0) = 0.4286 × e− ^{0.4286 × 0} = 0.4286.

  In step S7, the feature word evaluation unit 700 calculates the deviation amount | P (t) −P ′ (t) | for each appearance time interval. In the subsequent step S8, the feature word evaluation unit 700 calculates for each appearance time interval. The sum of P (t) −P ′ (t) | is obtained. In this example, as shown in FIG. 15, the total sum of the deviation amounts is 0.670.

In step S9, the feature word evaluation unit 700 calculates the evaluation value V based on the sum of the deviation amounts obtained in step S8. When the calculation of Expression (3) is performed to obtain the evaluation value V, V = e ^−0.670 = 0.512. Here, the case of obtaining the evaluation value of “security” is illustrated, but the evaluation value is calculated in the same manner for other feature words. As a result, the evaluation value of each feature word is obtained as shown in FIG.

  In step S10, the feature word selection unit 800 selects a feature word whose evaluation value is equal to or greater than a threshold value. In this example, it is assumed that the threshold is set to 0.3 in advance. The feature word selection unit 800 selects “security” and “network” having an evaluation value of 0.3 or more (see FIG. 11).

  In step S11, the interest information presenting means 900 displays the selected feature words “security” and “network” as words indicating the steady interest of the user or group. For example, it may be displayed as “recommended search keyword”.

  As shown in FIG. 12, the interest information identification system includes a search unit 950, and the search unit 950 uses the feature word selected in step S10 as a search word, a document that matches the search word, a Web page, and a news article. And the interest information presenting means 900 may display the search results together with the feature words.

  Alternatively, when the interest information presentation unit 900 displays the feature word in step S11, the search unit 950 does not perform the search, and after the feature word is displayed, any one of the displayed feature words is displayed to the user of the interest information specifying system. When specified, the search unit 950 may search various contents (documents, Web pages, etc.) using the specified feature word as a search word. The search result may be displayed by, for example, the interest information presentation unit 900. Even in this case, the content stored in the content management unit 300 may be a search target, or a content database outside the interest information identification system or various Web pages may be the search target.

  FIG. 16 is an explanatory diagram showing an example of a screen when performing a search using feature words. In step S11, the interest information presentation unit 900 displays a screen 1401 illustrated in FIG. The screen 1401 includes an identified feature word 1402, a search word input field 1403, and a search button. When a feature word is specified by clicking on the displayed feature word or by an operation such as inputting to the input field 1403 and clicking the search button, the search unit 950 searches for the specified feature word. And the interest information presenting means 900 displays the search result. A screen 1411 illustrated in FIG. 16B shows an example of a search result display screen. The screen 1411 includes, for example, the feature word 1412 and the search result 1414 specified in step S10. FIG. 16B shows a case where the designated feature word (in this example, “security”) is displayed in the input field 1413.

  The search result display screen 1411 may display a different feature word from the screen 1401 that first displays the feature word. For example, among the feature words specified in step S10, feature words that are not displayed on the first screen 1401 may be displayed. Alternatively, in step S10, the feature word selection unit 800 sorts the feature words in descending order of evaluation value, and when the interest information presentation unit 900 switches the screen as in the screens 1401 and 1411, the feature word is higher in the previous screen. May be displayed, and the feature word to be displayed may be switched.

  In addition, the interest information identification system may obtain and enumerate characteristic words that represent steady interest for each user and each department.

  Moreover, the interest information identification system may identify and display feature words about the user and feature words about the group to which the user belongs. Furthermore, the process of steps S2 to S10 may be performed for each of a plurality of periods instead of just one period, and feature words may be presented for each period. For example, the screen illustrated in FIG. 17 may be displayed. In the screen illustrated in FIG. 17, a feature word 1502 indicating a one-month stationary interest, a feature word 1503 indicating a one-year stationary interest, and a stationary interest of a group to which the set user belongs. A feature word 1504 is displayed. If any one of these feature words is specified, the search unit 950 may perform a search using the feature word as a search word, and display the search result 1505 as shown in FIG.

  In addition, in the example shown in FIG. 17, the case of showing a constant interest for one month and one year is illustrated, but a period such as the first half, the second half, or January, February, March, etc. A feature word may be specified for each display.

  When specifying a feature word indicating a constant interest for each period, for example, the feature word history generation unit 400 selects one period at a time, extracts only the access history in the selected period, and uses the access history Thus, the processing after step S2 may be performed. Further, the selected period may be used as the period T used in step S7. Then, each period is selected one by one, and a feature word for each period may be specified.

  Further, when the feature word selection unit 800 specifies a feature word in step S10, after specifying the feature word based on the evaluation value, the search unit 950 searches the content using each selected feature word as a search word. The feature words whose search results are 0 may be excluded from the selected results.

  In addition, when the content set to be searched by the search unit 950 and the content set stored by the content management unit 300 are the same, the search unit 950 searches for the content with the feature word specified by the user. Thereafter, the search result may be used to perform the process again from step S1, identify a feature word indicating a steady interest, and display the feature word.

  In the present embodiment, a feature word history is generated based on the accumulated access history, and a probability P ′ (t) that a feature word appears at a certain appearance time interval is calculated for each appearance time interval. Further, the probability P (t) that a feature word appears at a certain appearance time interval in the model probability distribution is calculated for each appearance time interval. Then, for each appearance time interval, the absolute value | P (t) −P ′ (t) | of the difference between the two is calculated, and the evaluation value of the feature word is calculated according to the sum. Here, since the probability distribution as a model is a probability distribution when feature words appear at random, the evaluation value is a degree of how much the feature words appear at random. In the present embodiment, feature words are specified based on such evaluation values, so that feature words that appear randomly and uniformly within a certain period are specified rather than feature words that temporarily appear within a certain period. can do. Therefore, according to the present embodiment, feature words that appear randomly within a certain period can be specified as interest information.

  In the first embodiment, the method for calculating the evaluation value based on the total sum of the deviation amounts is not limited to the calculation of Expression (3). For example, the sum of the deviation amounts | P (t) −P ′ (t) | at each appearance time interval may be used as the evaluation value. In this case, since the evaluation value becomes smaller as the feature word appears at random, the feature word selection unit 800 may select, for example, a feature word whose evaluation value is equal to or less than a threshold value.

  Further, in the processing example of the feature word evaluation unit 700 described above, the deviation amount | P (t) −P ′ (t) | for each appearance time interval is calculated and the total sum thereof is obtained. The method may be any method as long as the evaluation value is calculated according to the difference between the distribution of the appearance time intervals and the model probability distribution, and is not limited to the above calculation method.

  For example, the evaluation value V of the feature word may be calculated using the ratio of P (t) and P ′ (t). An example of calculating the evaluation value V using the ratio of P (t) and P ′ (t) will be described. In this example, the feature word evaluation unit 700 calculates P (t) / P ′ (t) as a deviation amount at each appearance time interval in step S7. In subsequent step S8, the feature word evaluation unit 700 multiplies P (t) / P ′ (t) calculated for each appearance time interval. That is, the feature word evaluation unit 700 calculates the following formula (4).

  The absolute value of the difference between the calculation result of Equation (4) and 1 indicates the magnitude of the deviation between the distribution of the appearance time intervals and the probability distribution as a model, and the difference between the calculation result of Equation (4) and 1 The larger the absolute value, the greater the deviation. The feature word evaluation means 700 may calculate the following expression (5) using the calculation result of expression (4) to obtain the evaluation value V of the feature word.

The exponent part on the right side of Equation (5) is a value obtained by multiplying the absolute value of the difference between the product of P (t) / P ′ (t) calculated for each appearance time interval by −1. The method of calculating the evaluation value V using the ratio of P (t) and P ′ (t) is not limited to the calculation method of Equation (5). For example, {log (P (t) / P ′ (t))} ² is calculated for each appearance time interval, and the product of the values is used to calculate Equation (6) below, The evaluation value V may be obtained.

The exponent part on the right side of Equation (6) is a value obtained by multiplying the product of {log (P (t) / P ′ (t))} ² calculated for each appearance time interval by −1.

  Further, in the expressions (5) and (6), when the ratio of P (t) and P ′ (t) is used, the product of the values obtained at every appearance time interval is calculated. The evaluation value may be obtained by adding the values obtained for each interval. For example, | 1- (P (t) / P ′ (t)) | is calculated for each appearance time interval, and the following expression (7) is calculated using the sum, and the evaluation value of the feature word is calculated. V may be obtained.

  The exponent part on the right side of Equation (7) is a value obtained by multiplying the sum of | 1- (P (t) / P '(t)) | calculated for each appearance time interval by -1.

  When the evaluation value V is obtained by the calculation exemplified in the equations (5) to (7), the evaluation value V increases as the deviation between the appearance time interval distribution and the model probability distribution decreases.

| P (t) −P ′ (t) | in Expression (3) is an example of a value indicating the degree of deviation between the distribution of the appearance time intervals and the model probability distribution. Similarly, P (t) / P ′ (t) in equation (5), {log (P (t) / P ′ (t))} ² in equation (6), and | 1− in equation (7). (P (t) / P ′ (t)) | is also an example of a value indicating the degree of deviation.

Embodiment 2. FIG.
Next, a second embodiment of the present invention will be described. Similar to the first embodiment, the interest information specifying system of the second embodiment also includes an extraction target setting unit 100, an access history storage unit 200, a content management unit 300, a feature word history generation unit 400, and an appearance. Frequency calculation means 500, appearance interval calculation means 600, feature word evaluation means 700, feature word selection means 800, and interest information presentation means 900 are provided. Further, a search unit 950 may be provided. Hereinafter, a second embodiment will be described with reference to FIG.

  In the second embodiment, the method for calculating the evaluation value by the feature word evaluation unit 700 is different from that in the first embodiment. The operations of the components other than the feature word evaluation unit 700 are the same as those in the first embodiment, and a description thereof will be omitted.

  In the second embodiment, the feature word evaluation unit 700 calculates the evaluation value of the feature word so that the value increases as the tendency of the feature word to appear regularly increases. For example, an evaluation value is calculated so that the value of a feature word that appears seven times in a concentrated manner on one day and a feature word that appears once a day for seven days are larger. It can be said that the feature word that repeats regularly appears more representatively than the feature words that occur intensively at a certain time. In the present embodiment, calculation is performed so that a high evaluation value is assigned to such a feature word. Specifically, the feature word evaluation unit 700 calculates the evaluation value so that the value increases as the appearance period is longer and the standard deviation and average value of the appearance time interval are smaller. The appearance period is a period from when the feature word of interest first appears to when it last appears. That is, it is a period from the time when the content whose feature is expressed by the feature word of interest is first used until the time when it is used last.

When the appearance interval calculation unit 600 obtains the appearance time interval of each feature word, the feature word evaluation unit 700 calculates the standard deviation and the average value of the appearance time interval for each feature word. The standard deviation may be calculated using sample variance or may be calculated using unbiased variance. This standard deviation is denoted as STDEV, and the average value is denoted as AVE. In addition, the first appearance time for each feature word is _denoted as T _0, and the last appearance time is denoted as T _last . Similarly to the first embodiment, T is a feature word history derivation target period (a period during which an access history that is a basis for generating a feature word history is collected). The feature word evaluation unit 700 obtains the evaluation value V of the feature word by calculating the following expression (8) using the parameter β.

V = {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} Equation (8)

  The parameter β is a parameter for adjusting whether or not a feature word having a short appearance time interval is particularly easily extracted. When the value of β is large, a larger evaluation value can be given when the appearance time interval is short than when the appearance time interval is long, and the difference between the evaluation values when the appearance time interval is short and long Can be increased. The value of β may be determined in advance. Alternatively, for example, it may be input by the user of the interest information specifying system via the extraction target setting unit 100. The timing at which the value of β is input may be before the calculation of Expression (8). For example, it may be input together with a user name or a group name.

  In addition, the feature word evaluation unit 700 sets the evaluation value to a predetermined value without calculating the formula (8) for the feature word whose appearance frequency is equal to or less than the predetermined number of times. This predetermined value may be a value indicating that the evaluation value is the lowest. In the following description, it is assumed that the predetermined value is zero.

  In addition, the “predetermined number of times” related to the appearance frequency is, for example, twice, but may not be twice. That the appearance frequency is 2 times or less requires only one appearance time interval at the maximum. If no appearance time interval is obtained, the standard deviation of the appearance time interval cannot be obtained. Further, even if the appearance frequency is 2 and one appearance time interval is obtained, the standard deviation due to unbiased variance is not obtained, and the standard deviation due to sample variance can be obtained, but the standard deviation is 0. Therefore, it cannot be distinguished from the case where feature words appear at uniform intervals. Therefore, when the appearance frequency is, for example, twice or less, the evaluation value is set to 0. Even if the appearance frequency of the feature word is more than twice, the standard deviation may be zero if the appearance frequency is low and the number of appearance time intervals obtained from the appearance frequency is small. Therefore, the “predetermined number of times” may be more than two times. The “predetermined number of times” may be a specified value, or a value indicating the “predetermined number of times” may be input from the user of the interest information specifying system. In the following description, the case where the evaluation value of a feature word whose appearance frequency is twice or less is set to 0 will be described as an example.

  The feature word evaluation unit 700 sequentially selects each feature word in the feature word history and calculates an evaluation value of the selected feature word.

  A specific example of the processing of the feature word evaluation unit 700 in the second embodiment will be shown. Assume that the feature word history generation unit 400 generates a feature word history illustrated in FIG. 9A as a feature word history between September 1 and September 14 for a certain feature word A. In this case, the appearance frequency calculation means 500 obtains the appearance frequency “9” of the feature word A as shown in FIG. The appearance interval calculation means 600 obtains the appearance time interval shown in FIG. Since the appearance frequency is more than twice, the feature word evaluation unit 700 calculates the formula (8) to obtain the evaluation value. Here, explanation will be made assuming that β = 0.1.

Since the feature word history is generated from the access history from September 1st to September 14th, T = 14. Further, as shown in FIG. 9A, since the last appearance time of the feature word A is September 1 and the last appearance time is September 12, T _last −T ₀ = 11. . Further, the feature word evaluation unit 700 calculates the standard deviation STDEV and the average value of each appearance time interval shown in FIG. In this example, AVE = 1.375 and STDEV = 1.506. Therefore, the feature word evaluation unit 700 calculates the evaluation value V of the feature word A by calculating (11/14) · e ^{(−0.1 × 1.375 × 1.506)} . In this example, V = 0.639.

  Here, the example shown in FIG. 9 is used for explanation. However, when the appearance frequency and the appearance time interval shown in FIGS. 7 and 8 are obtained from the feature word history shown in FIG. 4 and the evaluation value of each feature word is obtained, FIG. As shown in FIG. Characteristic words "Ubiquitous", "PC", "Server", and "Storage" have an appearance frequency of 2 or 1 and cannot be determined for the appearance time interval or only one, so evaluation The value is set to 0.0.

  In this example, since a high evaluation value is given to a feature word that appears regularly and expresses a constant interest, the feature word selection unit 800, for example, has an evaluation value equal to or greater than a threshold value. What is necessary is just to select the feature word. If the threshold is 0.1, the feature word selection unit 800 selects “security” and “network” from the feature words shown in FIG.

  Next, the operation will be described. FIG. 19 is a flowchart illustrating an example of processing progress of the interest information identification system according to the second embodiment. Steps S1 to S4 and steps S10 and S11 shown in FIG. 19 are the same processes as those in the first embodiment, and a description thereof will be omitted.

When the appearance frequency and appearance time interval of each feature word are obtained in the processing up to step S4, the feature word evaluation unit 700 calculates the average value AVE and standard deviation STDEV of each appearance time interval (step S4a). Subsequently, the feature word evaluation unit 700 calculates Expression (8) using AVE and STDEV obtained in Step S4a, and calculates the evaluation value V of the feature word (Step S4b). In step S4b, the feature word evaluation unit 700 may refer to the feature word history and set the first appearance time of the feature word of interest as T ₀ and the last appearance time as T _last .

  The feature word evaluation unit 700 performs the processing of steps S4a and S4b for each feature word. In each of steps S4a and S4b, feature words may be sequentially selected, and processing may be performed for each selected feature word. Further, before executing the process of step S4a, a feature word is selected, the process of steps S4a and S4b is performed for the selected feature word, the next feature word is further selected, and the process of steps S4a and S4b is performed in the same manner. You may go.

  The subsequent steps S10 and S11 are the same as those in the first embodiment. Various modifications described in the first embodiment may be applied to the second embodiment.

  A specific example of the above process is shown below. Note that the processing up to step S4 is the same as in the first embodiment, and a description thereof will be omitted. Assume that the feature word history shown in FIG. 4 is generated in step S2, the appearance frequency shown in FIG. 7 is calculated in step S3, and the appearance time interval shown in FIG. 8 is obtained in step S4. Further, in the following description, the calculation process of the “security” evaluation value is exemplified. FIG. 20 is an explanatory diagram showing this calculation process.

  In step S4a, the feature word evaluation unit 700 calculates the average value AVE and the standard deviation STDEV of the appearance time intervals for each feature word using the appearance time intervals. For example, in the case of the feature word “security”, the appearance time interval “0 day” is once and “1 day” is once. Therefore, the average value AVE = 0.5, and the standard deviation STDEV = 0.707.

Next, in step S4b, the feature word evaluation unit 700 calculates the expression (8) to obtain the evaluation value V of the feature word. In the case of the feature word “security”, T _last is “2007/09/02”, and T ₀ is “2007/09/01”. Therefore, T _last −T ₀ = 1. Further, since the derivation target period T of the feature word history is 7, V = (1/7) · e ^{(−0.1 × 0.5 × 0.707)} = 0.138.

  Similarly, when evaluation values are obtained for other feature words, they are as shown in FIG. For "Ubiquitous", "PC", "Server", and "Storage", the appearance frequency is twice or once, and the appearance time interval cannot be obtained or only one can be obtained. Is defined as 0.0.

  After calculating the evaluation value of each feature word, the processes of steps S10 and S11 are performed. This process is the same as in the first embodiment.

  In the present embodiment, since the evaluation value of each feature word is calculated by the equation (8), a higher evaluation value is given to a feature word having a small appearance interval and no bias for a long period of time. Since feature words are selected based on the evaluation value, it is possible to extract feature words that regularly appear in a longer period than feature words that frequently appear in a short period of time. It can be used as a feature word representing interest.

  The parameter β is a parameter for adjusting whether or not a feature word having a short appearance time interval is easily extracted with priority. By adjusting the value of β, the appearance time interval of the feature word is set. Precisely extract short feature words, or assign a relatively high evaluation value to feature words that appear regularly even if the appearance time interval is not short so that such feature words can be easily extracted. And can be adjusted. Therefore, on the premise that the evaluation value becomes larger as the appearance period is longer and the standard deviation and average value of the appearance time interval are smaller, as the value of β is larger, the feature word and the longer feature whose appearance time interval is shorter The difference between the evaluation values of the words can be increased to facilitate the extraction of the former.

  FIG. 21 is an explanatory diagram showing a comparison between an evaluation value when β = 0.1 and an evaluation value when β = 1.0. The horizontal axis shown in FIG. 21 is the appearance time interval, and the vertical axis is the evaluation value. Even when β = 0.1 and β = 1.0, the evaluation value decreases as the appearance time interval increases, but the value decreases more rapidly when β = 1.0. That is, when β = 1.0, the difference between the evaluation values increases when the appearance time interval is short and when it is long. For example, when β = 1.0, when the appearance time interval becomes about 5, the evaluation value approaches 0. When the appearance time interval is short, a larger evaluation value is given than 0, and the appearance time interval is short. The difference in evaluation value increases between the case and the case where the case is long. As a result, an evaluation value with a short appearance time interval is easily extracted.

  Next, the outline of the present invention will be described. FIG. 22 is a block diagram showing an outline of the present invention. The interest information specifying system of the present invention includes appearance frequency calculating means 971, appearance interval calculating means 972, feature word evaluating means 973, and feature word specifying means 974.

  The appearance frequency calculation means 971 (for example, the appearance frequency calculation means 500 shown in FIG. 1) includes a feature word representing the characteristics of the content used by the person or group, and each content in which the characteristics of the person or group are represented by the feature word. The feature word appearance frequency, which is the use frequency of the content represented by the feature word, is obtained for each feature word with reference to the feature word history including the use time using the.

  The appearance interval calculation unit 972 (for example, the appearance interval calculation unit 600 shown in FIG. 1) refers to the feature word history and obtains the appearance time interval that is the use time interval of the content represented by the feature word for each feature word.

  The feature word evaluation unit 973 (for example, the feature word evaluation unit 700 shown in FIG. 1) refers to the feature word appearance frequency and the appearance time interval, and for each feature word, the appearance time interval distribution and the model probability distribution The evaluation value of the feature word is obtained according to the divergence.

  The feature word specifying unit 974 (for example, the feature word selecting unit 800 shown in FIG. 1) specifies the feature word based on the evaluation value.

  With such a configuration, the evaluation value of the feature word is calculated according to the difference between the actual probability distribution and the model probability distribution, and the feature word is specified based on the evaluation value. Appearing feature words can be identified. Therefore, it is possible to specify a regular interest of a person or group.

  In the first embodiment, the feature word evaluation unit selects each feature word, and for each appearance time interval in the selected feature word, the probability that the feature word appears at the appearance time interval and the probability distribution that serves as a model Is used to calculate a value indicating the degree of deviation from the probability that a feature word will appear at the appearance time interval, and based on the value indicating the degree of deviation calculated for each appearance time interval, the evaluation value of the selected feature word is calculated. The required configuration is described.

  In the first embodiment, the feature word evaluation unit selects each feature word, and the probability word appearance at the appearance time interval for each appearance time interval in the selected feature word is used as a model. Obtain the absolute value of the difference between the probability of occurrence of a feature word at the appearance time interval in the probability distribution and obtain the evaluation value of the selected feature word based on the sum of the absolute values of the differences calculated at each appearance time interval The configuration is described.

In the first embodiment, the feature word evaluation unit calculates the probability that the feature word appears at the appearance time interval t when the appearance time interval is t, and the number of times the feature word appears at the appearance time interval t. Probability distribution that is a model when the feature word history derivation period is T and the feature word appearance frequency is K, calculated by dividing by the sum of the number of appearances of the feature word at each appearance time interval Describes a configuration in which the probability that a feature word appears at an appearance time interval t is calculated by calculating (K / T) e- ^{(K / T) t} .

The feature word evaluation unit 973 calculates the standard deviation and the average value of the appearance time intervals for each feature word, sets the standard deviation to STDEV, sets the average value to AVE, and the first use time of the content represented by the feature word and When the last use time is T _last and T ₀ and the feature word history derivation target period is T, using the parameter β, {(T _last −T ₀ ) / T} · e ^{(−β ·} By calculating ( ^{STDEV · AVE)} , an evaluation value of the feature word may be obtained, and an evaluation value of the feature word whose feature word appearance frequency is equal to or less than a predetermined number of times may be set to a predetermined value.

  In this case, a large evaluation value is assigned to a feature word having a long appearance period and a small standard deviation and average value of the appearance time interval, and the feature word can be specified. Thus, by identifying feature words that appear regularly, it is possible to identify the regular interest of a person or group. In addition, by adjusting the value of the parameter β, feature words having a short appearance time interval can be preferentially extracted, or the feature words that appear regularly even if the appearance time interval is not short can be relatively By assigning a high evaluation value, such feature words can be easily extracted or adjusted.

  In the second embodiment, a configuration including parameter input means (for example, realized by the extraction target setting means 100) for inputting the value of the parameter β is disclosed.

  In each embodiment, a usage history storage unit (for example, an access history storage unit 200) that stores usage history including content identification information, content usage time, and a person or a group that uses the content, and content identification A meta information storage unit (for example, content management unit 300) that stores meta information including information and a feature word representing a feature of the content, a usage history and the meta information, and a person who is a target of interest identification or A feature word history generating unit (for example, a feature word history generating unit 400) that generates a feature word history by specifying a feature word representing a feature of content used by a group and associating the use time of the content with the feature word. The structure provided with these is disclosed.

  In each embodiment, a specific target setting unit (for example, an extraction target setting unit 100) that sets a person or a group that is a target of interest information is provided, and the feature word history generating unit includes the use history and the meta information. A configuration for generating a feature word history by specifying a feature word representing a feature of content used by a person or group set in the target setting unit and referring to the use time of the content with the feature word Is disclosed.

  In each embodiment, a configuration is disclosed that includes search means (for example, search means 950 shown in FIG. 12) for searching for content using the feature words specified by the feature word specifying means.

  In each embodiment, a configuration is disclosed that includes display means (for example, interest information presenting means 900) that displays the feature words specified by the feature word specifying means.

  In each embodiment, a configuration including a search unit (for example, a search unit 950 shown in FIG. 12) that searches for content using a specified feature word among the feature words displayed by the display unit is disclosed. .

  The present invention is suitably applied to, for example, an interest information specifying system used for an information search system for searching data stored in a database and an information recommendation system for recommending suitable information from data stored in a database. Is done. It can also be applied to interest information identification systems used for directory services that display information on the portal site according to the user's interest on the Internet or intranet, and search for the affiliation, activity, and interest of users and employees. It is.

It is a block diagram which shows the example of the interest information specific system of the 1st Embodiment of this invention. It is explanatory drawing which shows the example of an access history. It is explanatory drawing which shows the example of meta information. It is explanatory drawing which shows the example of a characteristic word log | history. It is explanatory drawing which shows the example of the evaluation value of each feature word. It is explanatory drawing which shows the example of the input screen of a user name or a group name. It is explanatory drawing which shows the example of appearance frequency. It is explanatory drawing which shows the example of an appearance time interval. It is explanatory drawing which shows the example of the characteristic word log | history used for evaluation value calculation, appearance frequency, and appearance time interval. It is explanatory drawing which shows the calculation process of an evaluation value. It is explanatory drawing which shows the example of the calculated evaluation value. It is a block diagram which shows the example of an interest information specific system provided with a search means. It is explanatory drawing which shows the example of the screen which an interest information presentation means outputs. It is a flowchart which shows the example of the process progress of the interest information identification system of 1st Embodiment. It is explanatory drawing which shows the calculation process of an evaluation value. It is explanatory drawing which shows the example of a screen in the case of performing the search by a feature word. It is explanatory drawing which shows the example of the screen which an interest information presentation means outputs. It is explanatory drawing which shows the example of the calculated evaluation value. It is a flowchart which shows the example of the process progress of the interest information identification system of 2nd Embodiment. It is explanatory drawing which shows the calculation process of the evaluation value in 2nd Embodiment. It is explanatory drawing which shows the comparison with the evaluation value in the case of (beta) = 0.1, and the evaluation value in the case of (beta) = 1.0. It is a block diagram which shows the outline | summary of this invention.

Explanation of symbols

100 extraction target setting means 200 access history storage means 300 content management means 400 feature word history generation means 500 appearance frequency calculation means 600 appearance interval calculation means 700 feature word evaluation means 800 feature word selection means 900 interest information presentation means 950 search means 971 appearance Frequency calculation means 972 Appearance interval calculation means 973 Feature word evaluation means 974 Feature word identification means

Claims (21)

An interest information identification system that identifies interest information representing the interest of a person or group,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. And an appearance frequency calculating means for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
With reference to the feature word history, for each feature word, an appearance interval calculating means for obtaining an appearance time interval that is a use time interval of the content represented by the feature word;
A feature word evaluation unit that refers to a feature word appearance frequency and an appearance time interval, and obtains an evaluation value of the feature word according to a deviation between a distribution of the appearance time interval and a probability distribution as a model for each feature word;
An interest information specifying system comprising: feature word specifying means for specifying a feature word based on an evaluation value. The feature word evaluation means is
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest information specifying system according to claim 1, wherein a value indicating a degree of deviation from the selected feature word is calculated based on a value indicating the degree of deviation calculated for each appearance time interval. . The feature word evaluation means is
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest information specification according to claim 1, wherein an absolute value of a difference between the selected feature word and an evaluation value of the selected feature word is obtained based on a sum of absolute values of the differences calculated at each appearance time interval. system. The feature word evaluation means is
When the appearance time interval is t, the probability that the feature word appears at the appearance time interval t is the sum of the number of times the feature word appears at the appearance time interval t. Calculated by dividing by
When the characteristic word history derivation target period is T and the appearance frequency of the feature word is K, the probability that the feature word appears at the appearance time interval t in the model probability distribution is (K / T) The interest information identification system according to claim 3, which is obtained by calculating e ^{− (K / T) t} . An interest information identification system that identifies interest information representing the interest of a person or group,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. And an appearance frequency calculating means for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
With reference to the feature word history, for each feature word, an appearance interval calculating means for obtaining an appearance time interval that is a use time interval of the content represented by the feature word;
For each feature word, the standard deviation and average value of the appearance time intervals are calculated, the standard deviation is STDEV, the average value is AVE, and the first use time and the last use time of the content represented by the feature word are T {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} is calculated by using the parameter β, where _last and T ₀ are T and the characteristic word history derivation target period is T. Thus, an evaluation value of the feature word is obtained, and the feature word evaluation means for setting the evaluation value of the feature word that is equal to or less than the predetermined number of times that the feature word appearance frequency is determined,
An interest information specifying system comprising: feature word specifying means for specifying a feature word based on an evaluation value. The interest information specifying system according to claim 5, further comprising parameter input means for inputting a value of the parameter β. Use history storage means for storing a use history including content identification information, use time of the content, and identification information of a person or group using the content;
Meta-information storage means for storing meta-information including content identification information and feature words representing the characteristics of the content;
By referring to the usage history and the meta information, a feature word representing a feature of the content used by a person or group of interest is specified, and the use time of the content is associated with the feature word The interest information identification system according to any one of claims 1 to 6, further comprising: feature word history generation means for generating a feature word history. With specific target setting means for setting a person or group as a target of specific interest information,
The feature word history generation means refers to the use history and the meta information, specifies a feature word representing the feature of the content used by the person or group set in the specification target setting means, and sets the content to the feature word The interest information identification system according to claim 7, wherein a feature word history is generated by associating the use times of. The interest information specifying system according to any one of claims 1 to 8, further comprising search means for searching for content using the feature word specified by the feature word specifying means. The interest information specifying system according to any one of claims 1 to 9, further comprising display means for displaying the feature words specified by the feature word specifying means. The interest information identification system according to claim 10, further comprising search means for searching for content using a specified feature word among the feature words displayed by the display means. An interest information identification method for identifying interest information representing an interest of a person or group,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. In addition, an appearance frequency calculating step for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
With reference to the feature word history, for each feature word, an appearance interval calculation step for obtaining an appearance time interval that is a use time interval of content represented by the feature word;
A feature word evaluation step for obtaining an evaluation value of a feature word according to a difference between a distribution of appearance time intervals and a probability distribution as a model for each feature word with reference to the feature word appearance frequency and the appearance time interval;
And a feature word specifying step of specifying a feature word based on the evaluation value. In the feature word evaluation step,
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest information specifying method according to claim 12, wherein a value indicating a degree of divergence is calculated, and an evaluation value of the selected feature word is obtained based on a value indicating the degree of divergence calculated for each appearance time interval. . In the feature word evaluation step,
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest information specification according to claim 12 or 13, wherein an absolute value of a difference between the selected feature word and an evaluation value of the selected feature word is obtained based on a sum of absolute values of the differences calculated at each appearance time interval. Method. An interest information identification method for identifying interest information representing an interest of a person or group,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. In addition, an appearance frequency calculating step for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
With reference to the feature word history, for each feature word, an appearance interval calculation step for obtaining an appearance time interval that is a use time interval of content represented by the feature word;
For each feature word, the standard deviation and average value of the appearance time intervals are calculated, the standard deviation is STDEV, the average value is AVE, and the first use time and the last use time of the content represented by the feature word are T {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} is calculated by using the parameter β, where _last and T ₀ are T and the characteristic word history derivation target period is T. A feature word evaluation step for determining a feature word evaluation value and setting a feature word evaluation value equal to or less than a predetermined number of times the feature word appearance frequency is determined;
And a feature word specifying step of specifying a feature word based on the evaluation value. The interest information specifying method according to claim 15, further comprising a parameter input step in which a value of the parameter β is input. An interest information identification program installed in a computer for identifying interest information representing the interest of a person or group,
In the computer,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. In addition, an appearance frequency calculation process for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
An appearance interval calculation process for obtaining an appearance time interval that is a use time interval of the content represented by the feature word for each feature word with reference to the feature word history,
With reference to the feature word appearance frequency and the appearance time interval, for each feature word, a feature word evaluation process for obtaining an evaluation value of the feature word according to a deviation between the distribution of the appearance time interval and the probability distribution as a model, and
An interest information specifying program for executing a feature word specifying process for specifying a feature word based on an evaluation value. On the computer,
In the feature word evaluation process,
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest information specification according to claim 17, wherein a value indicating a degree of deviation from the selected feature word is calculated based on a value indicating the degree of deviation calculated for each appearance time interval. Program. On the computer,
In the feature word evaluation process,
Each feature word is selected, and for each appearance time interval in the selected feature word, the probability that the feature word will appear at the appearance time interval, and the probability that the feature word will appear at the appearance time interval in the model probability distribution The interest according to claim 17 or 18, wherein an absolute value of a difference between the selected feature word and an evaluation value of the selected feature word is obtained based on a sum of absolute values of the differences calculated at each appearance time interval. Information identification program. An interest information identification program installed in a computer for identifying interest information representing the interest of a person or group,
In the computer,
Each feature word is referred to by referring to a feature word history including a feature word representing a feature of the content used by the person or group and a use time when each person or group uses the content whose feature is represented by the feature word. In addition, an appearance frequency calculation process for obtaining a feature word appearance frequency that is a use frequency of the content represented by the feature word,
An appearance interval calculation process for obtaining an appearance time interval that is a use time interval of the content represented by the feature word for each feature word with reference to the feature word history,
For each feature word, the standard deviation and average value of the appearance time intervals are calculated, the standard deviation is STDEV, the average value is AVE, and the first use time and the last use time of the content represented by the feature word are T {(T _last −T ₀ ) / T} · e ^{(−β · STDEV · AVE)} is calculated by using the parameter β, where _last and T ₀ are T and the characteristic word history derivation target period is T. Thus, an evaluation value of the feature word is obtained, a feature word evaluation process in which the evaluation value of the feature word that is equal to or less than the predetermined number of times the feature word appearance frequency is determined, and
An interest information specifying program for executing a feature word specifying process for specifying a feature word based on an evaluation value. On the computer,
The interest information specifying program according to claim 20, wherein a parameter input process in which a value of parameter β is input is executed.

Images