JP2013257747A - Free time estimation device, method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To appropriately estimate free time of a user from a document posted on the Internet.SOLUTION: A preprocessing unit performs a morphological analysis of each blog document included in an input blog document collection, and extracts data of the posting date and time added to each blog document. A word n-gram extraction unit 141 extracts a word n-gram from the blog document collection by using the result of the morphological analysis. An average value calculation unit 142 acquires the average value of the frequency of the tweets for each time zone in a day by using the data of the posting date and time. A frequency for each time zone extraction part 143 extracts the average value of the frequency of the tweets for each time zone as identity. A standard deviation calculation unit 144 calculates standard deviation showing the dispersion of the frequency of the user's tweets in a day. An identity extraction unit 14 estimates the free time of the user by using the identity which puts together the word n-gram, the frequency of tweets for each time zone and the standard deviation of the frequency of the tweets; and a free time estimation model which is generated from the blog document collection for learning in advance.

Description

  The present invention relates to a free time estimation apparatus, method, and program, and more particularly, to a free time estimation apparatus, method, and program for estimating a user's free time based on a document such as a blog posted on the Internet.

  Currently, a user's attributes (gender, residence, etc.) are determined from a document (text data) posted by the user on the Internet, such as a microblog such as Twitter (registered trademark) (for example, Non-Patent Document 1). For example, a model in which n-grams of related words for each attribute are learned as a feature is created, and from this word, the user's word is included in a document posted by a user whose attribute is unknown. Estimating attributes. With such a method, when the correlation between the content written in the document and the attribute is strong, the attribute can be determined with high performance.

Tsutomu Okura, Nobuyuki Shimizu, Hiroshi Nakagawa, “Scalable and Versatile Blog Author Attribute Determination Method”, Information Processing Society of Japan Research Report, 2007-NL-181 (1)

  However, unlike attributes such as gender and place of residence, there is often no strong correlation between the user's free time and the content of the document posted by the user. Here, the free time is an attribute of the user indicating leisure time other than work, housework, sleep, meal, etc. per certain period (for example, one day, one week, one month, etc.). For example, when there are a plurality of users who frequently post political topics on microblogs, it is difficult to think that the free times of these users are the same. Thus, it is difficult to estimate the user's free time based only on the content of the posted document.

  The present invention has been made in consideration of the above facts, and an object thereof is to provide a free time estimation device, method, and program capable of appropriately estimating a user's free time from a document posted on the Internet. And

  In order to achieve the above object, the free time estimation device of the present invention is based on a document set including a plurality of document data posted by the same user to which data indicating posting date is added. A feature extraction means for extracting a first feature based on the appearance frequency of the included word and a second feature indicating a distribution of the number of posted document data for each divided period obtained by dividing a certain period into a plurality of periods; The first feature and the second feature extracted from each of a plurality of learning document sets to which a correct answer label of free time in a certain period is given, and a correct label given to each of the learning document set The estimation model that learned the association, and the first feature and the second feature extracted by the feature extraction means from the document set of the estimation target user whose free time in the predetermined period is unknown Zui and is configured to include a an estimation means for estimating the free time in the predetermined period of the estimated target user.

  According to the free time estimation device of the present invention, the feature extraction means is included in the document set based on a document set including a plurality of document data posted by the same user to which data indicating the posting date and time is added. A first feature based on the appearance frequency of words and a second feature indicating the distribution of the number of posts of document data for each divided period obtained by dividing a certain period into a plurality of periods are extracted. Then, the estimation means assigns each of the first feature and the second feature extracted from each of the plurality of learning document sets to which each user's free time correct answer label is given, to each of the learning document sets. Based on the estimation model that learned the association with the correct label and the first feature and the second feature extracted by the feature extraction means from the document set of the estimation target user whose free time is unknown for a certain period Estimate the free time of the target user in a certain period.

  In this way, in addition to the first feature based on the appearance frequency of words, by using the second feature indicating the distribution of the number of posted document data for each divided period in a certain period, the document posted on the Internet can be Free time can be estimated appropriately.

  Further, the feature extraction means can extract the average of the number of posts for each of the divided periods and the variation in the number of posts for each of the divided periods in the certain period as the second feature. By extracting such second features, it is possible to capture the features of the user's free time.

  In addition, the free time estimation apparatus of the present invention can be configured to include learning means for learning the estimation model using the plurality of learning document sets. Thereby, it can have a learning function.

  In the free time estimation method of the present invention, the feature extraction means is included in the document set based on a document set including a plurality of document data posted by the same user to which data indicating the posting date is added. A first feature based on the appearance frequency of the word to be extracted, and a second feature indicating a distribution of the number of postings of document data for each divided period obtained by dividing a certain period into a plurality of periods, and an estimation unit is configured to determine the predetermined feature of each user Correspondence between the first feature and the second feature extracted from each of the plurality of learning document sets to which the correct answer label of the free time in the period is given, and the correct label given to each of the learning document set On the basis of the first feature and the second feature extracted by the feature extraction means from the document set of the estimation target user whose unknown free time in the fixed period is unknown A method of estimating the free time in the predetermined period of the estimated target user.

  The free time estimation program of the present invention is a program for causing a computer to function as each means constituting the above free time estimation device.

  According to the free time estimation device, method, and program of the present invention, in addition to the first feature based on the appearance frequency of words, the second feature indicating the distribution of the number of posted document data for each divided period in a certain period is used. Thus, it is possible to appropriately estimate the user's free time from a document posted on the Internet.

It is a block diagram which shows the functional structure of the free time estimation apparatus which concerns on this Embodiment. It is a block diagram which shows the functional structure of a feature extraction part. It is a figure which shows distribution of the tweet count for every user. It is a flowchart which shows the learning process routine in this Embodiment. It is a flowchart which shows the estimation process routine in this Embodiment.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. In the present embodiment, a case where the user's free time per day is estimated will be described as an example.

  A free time estimation device 10 according to the present embodiment includes a CPU, a RAM, and a ROM that stores a program and various data for executing a free time estimation processing routine including a learning process and an estimation process described later. Consists of a computer. Further, an HDD may be provided as a storage means.

  As shown in FIG. 1, this computer can be functionally represented by a configuration including a learning unit 11 and an estimation unit 12. The learning unit 11 further includes a preprocessing unit 13 and a feature extraction unit 14. And an estimation model learning unit 15. The estimation unit 12 can be expressed by a configuration including a preprocessing unit 13, a feature extraction unit 14, and a free time estimation unit 16. it can. Note that the preprocessing unit 13 and the feature extraction unit 14 are functional units common to the learning unit 11 and the estimation unit 12.

  The free time estimation device 10 accepts a blog document posted on a microblog as an input document (text data). Hereinafter, posting a blog document on a microblog, or a posted document itself, is also referred to as “tweet”. “Mutter” consists of writing a small amount of text data of up to about 140 characters, and its contents generally represent the current situation, location, idea, etc. of the user. Further, each blog document handled in the present embodiment is added with data of a posting date (date and time when the user tweeted) when the user posted the blog document.

  The input document input to the free time estimation device 10 is performed in units of a blog document set including a plurality of blog documents posted by a certain user during a predetermined period (for example, one year). The set of blog documents input to the learning unit 11 is a set of blog documents for each of a plurality of users, and each blog document set is manually assigned a correct label indicating the user's free time. The set of blog documents input to the estimation unit 12 is a set of blog documents posted in a predetermined period by the estimation target user who wants to estimate free time.

  Hereinafter, each part of the free time estimation apparatus 10 will be described in detail.

  The preprocessing unit 13 divides each blog document included in the input blog document set into words by morphological analysis, which is an existing technique, and outputs a morpheme analysis result in which part-of-speech information is added to each word. For example, “I arrived in Yokohama. When a blog document (text data) is entered, “Yokohama (noun) / ni (case particle: continuous use) / tsu (verb stem: K) / i (verb inflection ending) / ta (verb suffix: end ) /. The morphological analysis result “(punctuation)” is output.

  In addition, the preprocessing unit 13 extracts and outputs the posting date data added to each blog document.

  As shown in FIG. 2, the feature extraction unit 14 is represented by a configuration including a word n-gram extraction unit 141, an average value calculation unit 142, a time-period number extraction unit 143, and a standard deviation calculation unit 144. be able to.

  The word n-gram extraction unit 141 extracts the word n-gram as a feature from the input blog document set using the morphological analysis result output from the preprocessing unit 13. The word n-gram is a feature represented by a morpheme notation and an appearance frequency of the notation in the blog document set. This feature is derived from the content of the blog document posted by the user, and is information that reflects the lifestyle and occupation related to the user's free time.

For example, if a blog document included in the blog document set is
When we arrived in Yokohama, the weather is nice.
In this case, the following word n-gram (here, n = 1) is extracted from the morphological analysis result.
Yokohama: 2 to 1: Arrive: 1 is: 1 Good: 1 Weather: 1

  Note that the word n-gram may be extracted from only the content words representing the lexical meaning among the words included in the blog document.

  The average value calculation unit 142 uses the posting date data output from the preprocessing unit 13 to determine the average value of the number of tweets for each time period in a certain period (here 1 day). For example, if a user's tweet count from midnight to 1 am is 3 times on March 21, 5 times on March 22, and 4 times on March 23, this user's midnight to 1 am The average value of the number of tweets is 4.

  The number-of-times extraction unit 143 extracts the average value of the number of tweets for each time period calculated by the average value calculation unit 142 as a feature. For example, if the average value of the number of tweets from midnight to 1 am for a certain user is 4, the feature representing the number of tweets from 0:00 to 1 am for this user is 4.

  Based on the average value of the number of tweets for each time period calculated by the average value calculation unit 142, the standard deviation calculation unit 144 calculates a standard deviation indicating the variation in the number of tweets per day of the user and outputs it as a feature. .

  Here, as shown in FIG. 3, the distribution of the number of tweets for each time zone is different for each user. Further, it is considered that there is a correlation between the distribution of the number of tweets and the user's free time. For example, a user who tweetes at a certain frequency in any time zone is considered to have a lot of free time, and a user who tweetes only in a specific time zone is a time for work or housework, and the free time It is thought that there are few. Thus, by using the tweet count distribution such as the tweet count for each time zone and the standard deviation of the tweet count as a feature, it is possible to capture the characteristics of the user's free time as described above.

  The feature extraction unit 14 collects the above word n-gram, the number of tweets for each time zone, and the standard deviation of the number of tweets, and outputs them as features extracted from one blog document set. As mentioned above, the word n-gram is information reflecting the lifestyle and occupation related to the user's free time, and the distribution of the number of tweets is information that captures the characteristics of the user's free time. Are used as features, so that the learning of the free time estimation model 20 and the estimation of the free time can be performed with high accuracy.

  When the learning unit 11 functions as the feature extraction unit 14, the feature extracted for each blog document set and the correct answer label assigned to the blog document set are paired, and the estimated model learning unit 15 in the subsequent stage. Hand over to. When functioning as the feature extraction unit 14 in the estimation unit 12, the extracted feature is transferred to the subsequent free time estimation unit 16.

  The estimation model learning unit 15 learns the association between the feature and correct label output from the feature extraction unit 14 using an existing technique, and estimates the free time estimation model 20 for estimating the user's free time. Is generated. For example, the free time estimation model 20 can be generated by regression analysis.

  The free time estimation unit 16 estimates and outputs the user's free time using the free time estimation model 20 generated by the estimation model learning unit 15 and the features extracted from the blog document set of the estimation target user. .

  Next, the operation of the free time estimation apparatus 10 according to the present embodiment will be described. When a set of blog documents for each of a plurality of users is input to the free time estimation apparatus 10 in the learning stage, the learning processing routine shown in FIG. In the estimation stage, when a set of blog documents of an estimation target user whose free time is to be estimated is input to the free time estimation apparatus 10, the estimation unit 12 executes an estimation processing routine shown in FIG. Hereinafter, each process is explained in full detail.

  First, in the learning processing routine, in step 100, the preprocessing unit 13 acquires a set of input blog documents for each of a plurality of users. Next, in step 102, the preprocessing unit 13 selects a blog document set of one user from a plurality of blog document sets. Next, in step 104, the preprocessing unit 13 divides each blog document included in the selected blog document set into words by morphological analysis, which is an existing technology, and further adds part-of-speech information to each word. Is output. In addition, the preprocessing unit 13 extracts and outputs the posting date data added to each blog document.

  Next, in step 106, the word n-gram extraction unit 141 extracts the word n-gram from the blog document set selected in step 102 using the morpheme analysis result output in step 104.

  Next, in step 108, the average value calculation unit 142 uses the posting date data output in step 104 to obtain the average value of the number of tweets for each time period in a certain period. Next, in step 110, the number of times extraction unit 143 for each time period extracts the average value of the number of tweets for each time period calculated in step 108 as a feature. Next, in step 112, the standard deviation calculation unit 144 calculates a standard deviation indicating the variation in the number of tweets per day based on the average value of the number of tweets for each time period calculated in step 108. , Output as a feature.

  Next, in step 114, the feature extraction unit 14 determines the word n-gram extracted in step 106, the number of tweets for each time period extracted in step 110, and the number of tweets calculated in step 112. The standard deviations are collected and used as the features of the blog document set selected in step 102, and a pair with the correct answer label for the free time given to the blog document set is created.

  Next, in step 116, the learning unit 11 determines whether or not the process of creating the correct label / feature pair has been completed for all users who have received the blog document set, and there are unprocessed users. If so, the process returns to step 102, the next user's blog document set is selected, and the processing of steps 104 to 114 is repeated. When the processing has been completed for all users, the process proceeds to step 118, where the estimation model learning unit 15 learns the pairs of features and correct labels created in step 114, and determines the free time estimation model. 20 is generated. The generated free time estimation model 20 is stored in a predetermined storage area, and the learning processing routine is terminated.

  Next, in the estimation processing routine, in step 120, the preprocessing unit 13 acquires the input blog document set of the estimation target user. Next, in steps 122 to 130, the preprocessing unit 13 and the feature extraction unit 14 extract the features of the input blog document set by the same processing as the learning processing steps 104 to 112.

  Next, in step 132, the free time estimation unit 16 estimates the user's free time using the free time estimation model 20 generated by the learning process and the features extracted from the blog document set of the estimation target user. Then, the estimation result is output and the estimation processing routine is terminated.

  As described above, according to the free time estimation device 10 according to the present embodiment, the user's free time is appropriately set from a document posted on the Internet by using the feature using the distribution of the number of postings of the document. Can be estimated.

  Note that the present invention is not limited to the above-described embodiment, and various modifications and applications are possible without departing from the gist of the present invention.

  For example, although the case where the learning unit and the estimation unit are configured by one computer has been described in the above embodiment, the learning unit and the estimation unit may be configured by separate computers.

  In the above embodiment, the case where the free time per day is estimated has been described as an example. However, the free time per longer period, for example, one week or one month may be estimated. You may make it estimate the free time from a short fixed period, for example, from 8:00 to 22:00.

  In the above embodiment, as the feature indicating the distribution of the number of tweets, the average value of the number of tweets for each time period and the standard deviation of the number of tweets for each time period in one day have been described. It is not limited to. For example, the frequency of tweets for each time zone, the cumulative frequency, the variance of the number of tweets per day, the deviation, etc. may be used.

  Moreover, although the above-mentioned free time estimation apparatus has a computer system inside, if a "computer system" is using the WWW system, it shall also include a homepage provision environment (or display environment). .

  In the present specification, the embodiment has been described in which the program is installed in advance. However, the program can be provided by being stored in a computer-readable recording medium. The present invention can also be realized by installing a program on a known computer via a medium or a communication line.

DESCRIPTION OF SYMBOLS 10 Free time estimation apparatus 11 Learning part 12 Estimation part 13 Preprocessing part 14 Feature extraction part 15 Estimation model learning part 16 Free time estimation part 20 Free time estimation model 141 Word n-gram extraction part 142 Average value calculation part 143 Every time zone Number extraction unit 144 Standard deviation calculation unit

Claims (5)

Based on a document set including a plurality of document data posted by the same user to which data indicating the posting date and time is added, a first feature based on the appearance frequency of words included in the document set and a plurality of fixed periods Feature extraction means for extracting a second feature indicating the distribution of the number of postings of document data for each divided period divided into the periods;
The first feature and the second feature extracted from each of the plurality of learning document sets to which the correct answer label for the free time in the predetermined period of each user is assigned, and each of the learning document sets is assigned to each of the learning document sets Based on the estimation model that learned the correspondence with the correct answer label, and the first feature and the second feature extracted by the feature extraction unit from the document set of the estimation target user whose free time in the fixed period is unknown Estimating means for estimating a free time in the fixed period of the estimation target user;
A free time estimation device including:   The free time estimation apparatus according to claim 1, wherein the feature extraction unit extracts, as the second feature, an average of the number of posts for each of the divided periods and a variation in the number of posts for each of the divided periods in the certain period.   The free time estimation apparatus according to claim 1, further comprising learning means for learning the estimation model using the plurality of learning document sets. A first feature based on an appearance frequency of words included in the document set based on a document set including a plurality of document data posted by the same user to which data indicating the posting date and time is added to each And extracting a second feature indicating the distribution of the number of submissions of document data for each divided period obtained by dividing a certain period into a plurality of periods
The estimation means includes the first feature and the second feature extracted from each of the plurality of learning document sets to which the correct answer label of the free time in the certain period of each user is assigned, and each of the learning document sets The first feature and the second feature extracted by the feature extraction means from the document set of the estimation target user whose unknown free time in the fixed period is unknown, and learning the association with the correct label given to Based on the above, a free time estimation method for estimating a free time in the certain period of the estimation target user.   The free time estimation program for functioning a computer as each means which comprises the free time estimation apparatus of any one of Claims 1-3.

Images