JP5133294B2 - Spatio-temporal search device, method and program - Google Patents

Description

  The present invention relates to a spatio-temporal search apparatus, method, and program, and in particular, the date, time, and place where an individual's experience is described in natural language and experienced (or the experience described therein is actually realized). The present invention relates to a spatio-temporal search device, method, and program for analyzing the relationship between human emotions and spatio-temporal regions for data having a structure with numerical information that can be uniquely identified as a target.

  In recent years, media called so-called CGM (Consumer Generated Media) such as blogs, SNSs (social network services), and Internet bulletin boards have been widely used. Unlike advertising information and objective data issued by companies, these media are characterized by including many human experiences, that is, individual behavior histories and subjective descriptions. Therefore, the necessity of using these media is increasing in various fields such as marketing, corporate management, and consumption behavior. In addition, with the spread of mobile terminals equipped with GPS such as mobile phones, opportunities to transmit CGM and web text with location information (latitude, longitude) and time information (date) are increasing.

  As a first conventional technique, there is known an experience mining technique for discovering behaviors and feelings of people peculiar to time or space for such CGM, in particular, a blog. This technology derives the relationship between time, space, action, and emotion using the co-occurrence relationship between the time information given to the blog and the place name, action word, and emotion word existing in the blog (for example, Non-Patent Document 1).

  As a conventional second technique, a technique for automatically extracting the polarity (positive / negative) of a word is known. This technology uses the number of hits of a Web search engine and determines the polarity of a word based on information about whether a word is likely to co-occur with a positive word (good) or a negative word (bad). Determine automatically. According to this technique, it can be seen that the word “accident” has a negative meaning stronger than positive (for example, see Non-Patent Document 2).

  The conventional third technique is an extension of the first technique that evaluates the evaluation polarity of a word in one axis, and the polarity of the word is composed of four axes (<happy, sad>, <surprise, anticipation>, <Acceptance, disgust>, <anger, fear>) (see, for example, Non-Patent Document 3).

  Further, according to the second and third techniques of the prior art, most simply, when a positive word appears more than a negative word in a document, the content is positive as a whole. It can be applied to document classification as well.

Experience mining from large-scale texts: Ken Kurashima, Ko Fujimura, Hidenori Okuda, The 19th Data Engineering Workshop / The 6th Annual Conference of the Database Society of Japan (DEWS2008), A1-4, 2008. Measuring Praise and Criticism: Inference of Semantic Orientation from Association, P. Turney and M.L.Littman, ACM transaction on Information Systems, Vol. 21, No. 4, 2003. Proposal of Impression Mining from News Articles, T. Kumamoto and K. Tanaka, Proc. Of the 10th International Conference on Knowledge-Based & Intelligent Information & Engineering Systems (KES 2005), LNAI 3681, pp. 901-910, 2005.

  In the first conventional technique, when time attributes (dates) and spatial attributes are given as numerical data, these numerical attributes are reduced to category attributes and solved, so the obtained solution has not been optimal. . Here, attributes that take two or more values such as blood type and gender are called category attributes, and attributes that take general numerical values such as time, latitude / longitude, weight, and height are called numerical attributes. Reducing to a category attribute means ignoring the continuity of numeric attributes. For example, at the moment when the value “November 06, 2008” of the numerical attribute “date” is treated as the value “November 6, 2008” of the simple category attribute “date”, November 5, November 6 As a result, it becomes impossible to obtain a tendency to occur across a plurality of category attribute values such as a tendency that occurred in one week from November 1 to November 7. The same applies to the spatial attribute. The first conventional technique finds a trend as one unit of the area represented by the place name, and therefore spans the area represented by the place name or the area represented by the place name. I couldn't find a trend that happened in some areas.

  In addition, the conventional second and third techniques are techniques for discovering a relationship between a fixed word and polarity that is not influenced by time, and it is possible to discover a relationship that changes depending on temporal / spatial factors. Can not.

  The present invention has been made in view of the above points, and the personal experience is described in natural language, and the date, time, and place where the experience was experienced (or the experience described therein was actually made) are unique. A spatio-temporal search device, method, and program capable of obtaining an optimal spatio-temporal region that best characterizes people who have a certain emotion are targeted for data having a structure in which numerical information that can be identified is assigned thereto For the purpose.

  FIG. 1 is a principle configuration diagram of the present invention.

The present invention (Claim 1) is based on a document in which an individual's experience is described in a natural language, and analysis target data having a structure in which position information and time information on which the experience has been given are given as numerical data. A spatio-temporal search device for finding an optimal spatio-temporal region that best characterizes a specific person,
Input means 10 for inputting data to be analyzed, constraints on the spatiotemporal area to be obtained, conditions for characterizing people, and storing them in the input information storage means 11;
A target value deriving unit 30 that sets a condition characterizing people given by the input unit 10 as a target value for obtaining an optimum spatiotemporal region;
Document polarity determination means 21 for determining whether the content of the text of the analysis target data is written by people corresponding to the conditions characterizing people;
Based on the determination result of the document polarity determination unit 21 and the position information and time information given to the analysis target data, a transaction having the position information, time information, and determination result as elements is generated and stored in the transaction storage unit 15. Transaction generation means 22;
A space-time region condition that maximizes the appearance probability of the transaction corresponding to the target value is derived from the set of transactions stored in the transaction storage unit 15 by extracting a numerical attribute correlation rule based on the constraints on the space-time region to be obtained. Numerical attribute association rule extraction means 40;
Have

Further, the present invention (Claim 2) is the spatio-temporal search device according to Claim 1,
The input means 10
As a condition that characterizes people, by specifying emotional polarity consisting of two emotions with opposite meanings, including means to characterize people in terms of whether they have one emotion or the opposite,
The target value deriving means 30
Means for setting the two emotions given by the input means 10 as target values for obtaining an optimal space-time region,
The document polarity determination means 21
Means for determining which content of the text of the analysis target data belongs to which of the two emotions which are the polarity values of the emotion polarity.

Further, the present invention (Claim 3) is the spatio-temporal search device according to Claim 1,
The input means 10
As a condition to characterize people, it includes means for characterizing people from the perspective of accepting multiple emotional polarities that consist of two emotions with specified opposite meanings,
The target value deriving means 30
For each emotional polarity given by the input means, including means for setting the objective value to a Cartesian product set of sets based on two opposing emotions constituting the emotion polarity,
The document polarity determination unit 21 includes a unit that determines, for each emotion polarity, the content of the text of the analysis target data belongs to two emotions that are the polarity values of the emotion polarity.

  The present invention (Claim 4) further comprises output means for outputting an optimum spatio-temporal region that best characterizes a specific person in the spatio-temporal search device according to any one of Claims 1 to 3.

  FIG. 2 is a diagram for explaining the principle of the present invention.

The present invention (Claim 5) includes a document in which an individual's experience is described in a natural language, and analysis target data having a structure in which the positional information and time information on which the experience is given are given as numerical data, A spatio-temporal search method that finds the optimal spatio-temporal region that best characterizes a particular person,
An input step (step 1) for inputting analysis target data, constraints on the space-time region to be obtained, conditions for characterizing people, and storing them in the input information storage means;
A target value deriving step (step 2) in which the conditions characterizing the people given in the input step (step 1) are set as target values for obtaining an optimal spatiotemporal region;
A document polarity determination step (step 3) for determining whether the text content of the analysis target data is written by people corresponding to the conditions characterizing people;
Based on the determination result of the document polarity determination step (step 3) and the position information and time information given to the analysis target data, a transaction having the position information, time information, and determination result as elements is generated and stored in the transaction storage means. A transaction generation step (step 4) to store;
A numerical value derived by extracting a numerical attribute correlation rule from a set of transactions in the transaction storage means, based on the constraints on the spatiotemporal domain to obtain the spatiotemporal domain condition that maximizes the appearance probability of the transaction corresponding to the target value An attribute correlation rule extraction step (step 5) is performed.

Further, according to the present invention (Claim 6), in the input step (Step 1) of the spatio-temporal search method according to Claim 5, an emotion polarity consisting of two emotions having opposite meanings is specified as a condition characterizing people. To characterize people in terms of whether they have a feeling or vice versa,
In the target value derivation step (step 2), two emotions having opposite meanings given in the input step are set as target values for obtaining the optimum spatiotemporal region,
In the document polarity determination step (step 3), it is determined which of the two emotions, which is the polarity value of the emotion polarity, the content of the text of the analysis target data.

Further, according to the present invention (Claim 7), in the input step (Step 1) of the spatio-temporal search method according to Claim 5, as a condition characterizing people, a plurality of emotions having two designated opposite meanings are provided. Characterize people in terms of accepting emotional polarity input and having multiple emotions at the same time,
In the objective value deriving step (step 2), for each emotion polarity given in the input step, the objective value is a Cartesian product set of a set based on two opposing emotions constituting the emotion polarity,
In the document polarity determination step (step 3), for each emotion polarity, it is determined which of the two emotions, which are the polarity values of the emotion polarity, the content of the text of the analysis target data belongs to.

  Further, according to the present invention (Claim 8), in the spatiotemporal search method according to any one of Claims 5 to 7, an output step of outputting an optimal spatiotemporal region that best characterizes a specific person to the display means. Is further performed.

  The present invention (Claim 9) is a spatiotemporal search program for causing a computer to function as each means constituting the spatiotemporal search apparatus according to any one of Claims 1 to 4.

  According to the present invention, a person who has a certain emotion is best characterized from text in which personal experience is described in natural language, and positional information and time information experienced in the form of numerical data. It is possible to obtain an optimal spatiotemporal region.

It is a principle block diagram of this invention. It is a figure for demonstrating the principle of this invention. It is a block diagram of the spatiotemporal search apparatus in one embodiment of this invention. It is a flowchart of outline | summary operation | movement of the spatio-temporal search apparatus in one embodiment of this invention. It is an example of the area | region group on the two-dimensional plane in one embodiment of this invention. It is an example of the transaction produced | generated by the transaction production | generation part in one embodiment of this invention. It is a flowchart showing the process of the target value derivation | leading-out part in one embodiment of this invention. It is an example which provided the objective attribute to the transaction of FIG. 6 in one embodiment of this invention. It is the example which mapped the transaction in one embodiment of this invention on the two-dimensional plane. It is the example which calculated | required the optimal square convex area | region from FIG. 9 in one embodiment of this invention.

  Hereinafter, embodiments of the present invention will be described with reference to the drawings.

  FIG. 3 shows the configuration of the spatio-temporal search device in one embodiment of the present invention.

  The spatiotemporal search device includes an input unit 10, an analysis target data storage unit 11, a threshold storage unit 12, an emotion polarity storage unit 13, a region group storage unit 14, a transaction storage unit 15, an optimal solution storage unit 16, and a transaction generation function unit 20. The target value deriving unit 30, the numerical attribute correlation rule extracting unit 40, and the output unit 50 are used to obtain an optimal spatiotemporal region that best characterizes a specific person, typically a person who has a certain emotion. In the present invention, a spatiotemporal region is defined as a concept including time, a spatial region, or a spatiotemporal region.

  FIG. 4 is a flowchart of the schematic operation of the spatio-temporal search device in one embodiment of the present invention.

  Step 100) The input unit 10 includes analysis target data input from the user, threshold value (minimum support), emotion polarity (two emotions having opposite meanings such as <positive, negative>), region group (two The plane with numerical attributes and the type of space with three numerical attributes) are stored in the analysis target data storage unit 11, the threshold storage unit 12, the emotion polarity storage unit 13, and the region group storage unit 14, respectively.

  Step 200) The objective value deriving unit 30 reads the emotion polarity from the emotion polarity storage unit 13, and obtains an objective value that is a target for obtaining the optimum time, space region, and spatiotemporal region. Alternatively, for each emotion polarity, a Cartesian product set of a set based on two ranges of emotions constituting the emotion polarity is obtained and set as a target value.

  Step 300) The transaction generation function unit 20 reads out one or more emotion polarities from the emotion polarity storage unit 13 and analysis target data from the analysis target data storage unit 11, and analyzes each emotion polarity. It is determined whether the data belongs to two emotions that are polar values.

  Step 400) The transaction generation function unit 20 generates a transaction having the emotion polarity of the analysis target data determined in Step 300 and the date, latitude, longitude, and emotion polarity value of the analysis target data as elements, and the transaction storage unit 15 To store.

  Step 500) The numerical attribute correlation rule extraction unit 40 stores the transaction stored in the transaction storage unit 15, the target value obtained by the target value deriving unit 30, the region group stored in the region group storage unit 14, and the threshold storage. The minimum support degree stored in the unit 12 is acquired, the optimal space region and the spatio-temporal region that characterize the specific people are obtained, and stored in the optimal solution storage unit 16 as the optimal solution.

  Step 600) The output unit 50 outputs the optimum solution stored in the optimum solution storage unit 16.

  Below, detailed operation | movement is demonstrated for every component.

  The input unit 10 receives analysis target data, a threshold value used when extracting an association rule, an emotion polarity, and a region group from a user. The input unit 10 includes, for example, a keyboard, an OCR, a pen input, a voice recognition device, a terminal equipped with a GPS, a unit that reads a text file placed on a network, and the like.

  The analysis target data is a personal experience history, impressions and evaluations based on the experience are described in natural language, and the time and position at which the experience is described or actually performed are described. It is assumed that the numerical information that can be identified is the data attached thereto. In the present embodiment, it is assumed that the date is specified as time information and the latitude and longitude are specified as position information.

  As the threshold value, a minimum support level is given. The minimum support level is a threshold value for the support level, which is one of the measures indicating the usefulness of the association rule described later.

The above emotional polarities are intended to show the polarities to narrow down people with certain characteristics by specifying two concepts with opposite meanings, such as <positive, negative>. Here, the emotional axis consisting of two concepts with affirmatives such as <positive, negative> is called "emotional polarity" (or "polarity" for short), and "positive" and "negative" are called "polarity values", respectively. I will do it. Typical emotional polarities other than <positive, negative>
(1) <happy and sad>
(2) <Surprise, anticipation>
(3) <Anger, fear>
(4) <Acceptance, disgust>
Can be mentioned. Moreover, although it is known that these four axes are orthogonal concepts, the user can inquire about mixed emotions by specifying a plurality of orthogonal emotion polarities. In the present embodiment, it is assumed that the user designates m emotion polarities Q ₁ = {p ₁ , n ₁ },..., Q _m = {P _m , N _m }. In addition to the emotional polarity shown here, antonym pairs of adjectives such as <good, bad> and <bright, dark> may be input. A certain keyword or concept may be <included, not included>.

The above-mentioned region group is a type of plane on which two numerical attributes are stretched and a space on which three numerical attributes are stretched. The user designates this region group when obtaining the optimum space region (two-dimensional) and space-time region (three-dimensional). When calculating the time, it is not necessary to specify a single numeric attribute. For example, as an example of a region family in the plane where the binary attribute extends,
(1) rectangular area,
(2) x monotonic region,
(3) A rectangular convex region is mentioned.

  As shown in FIG. 5A, the rectangular area (1) is an area that is expressed by the direct product of two numerical attribute sections and is parallel to the axis on the plane formed by the two numerical attributes.

  The x monotonic region in (2) is a connected region in which the intersection with a straight line perpendicular to the x axis (or y axis) is one section or empty, as shown in FIG.

  As shown in FIG. 5C, the rectangular convex area (3) is an x monotone and y monotone connected area.

  The analysis target data storage unit 11 stores the analysis target data input from the input unit 10. The analysis target data storage unit 11 may be anything as long as the structure of the input data is stored and can be restored. For example, it is stored in a specific area of a database or a general-purpose storage device (such as a memory or a hard disk device) provided in advance.

  The threshold storage unit 12 stores the threshold value input by the input unit 10. Similar to the analysis target data storage unit 11, the threshold storage unit 12 may be anything as long as it can store and restore the minimum support level. For example, it is stored in a specific area of a database or a general-purpose storage device (such as a memory or a hard disk device) provided in advance.

  The emotion polarity storage unit 13 stores the emotion polarity input by the input unit 10. The emotion polarity storage unit 13 may be anything as long as it can store and restore the emotion polarity. For example, it is stored in a specific area of a database or a general-purpose storage device (a memory, a hard disk device, etc.) provided in advance.

  The region family storage unit 14 stores the region family input by the input unit 10. Anything can be used as long as it can save and restore the input region family. For example, it is stored in a specific area of a database or a general-purpose storage device (a memory, a hard disk device, etc.) provided in advance.

  The transaction generation function unit 20 includes a document polarity determination unit 21 and a transaction generation unit 22. The document polarity determination unit 21 determines the polarity value of the content of the text portion of the analysis target data stored in the analysis target data storage unit 11 based on the polarity stored in the emotion polarity storage unit 13. For example, if the text is evaluated on the axis of <positive, negative>, the text "The clerk's attitude was bad" was classified as "negative" and "The food was very delicious" was classified as "positive" To do. Also, for m polarities, one text is determined m times with different polarities.

  The document polarity determination unit 21 can be realized by a document classification technique using machine learning, the above-described Non-Patent Document 2, Non-Patent Document 3, or a method using a thesaurus such as an evaluation expression dictionary or an emotion dictionary.

  The transaction generation unit 22 generates a transaction of the following format from the result of the document polarity determination unit 21 and the date, latitude, and longitude of the analysis target data stored in the analysis target data storage unit 11.

R = {ID, date, latitude, longitude, emotion polarity ₁ ,..., Emotion polarity _m }
The ID is a transaction identifier. The emotion polarity has a polarity value obtained by the document polarity determination unit 21 as a value. FIG. 6 shows an example of the result when two emotion polarities <joyful, sad> and <surprise, anticipation> are input.

  The transaction storage unit 15 stores the transaction generated by the transaction generation unit 22. The transaction storage unit 15 may be anything as long as it stores the transaction structure. For example, it is stored in a specific area of a database or a general-purpose storage device (a memory, a hard disk device, etc.) provided in advance.

  The target value deriving unit 30 derives a set P of target values for which an optimum time, space region, and spatio-temporal region are obtained from m emotion polarities input by the user stored in the evaluation polarity storage unit 13. .

  The target value set P is a Cartesian product set of m sets Q1,..., Qm as follows.

例えば、３つの感情極性Ｑ_１＝｛嬉しい，悲しい｝，Ｑ_２＝｛驚き，予期｝，Ｑ_３＝｛怒り，恐れ｝が入力された場合には、Ｐ＝｛（嬉しい，驚き，怒り），（嬉しい，驚き，恐れ），（嬉しい，予期，怒り），（嬉しい，予期，恐れ），（悲しい，驚き，怒り），(悲しい，驚き，恐れ)，（悲しい，予期，怒り），（悲しい，予期，恐れ）｝となる。つまり、ｍ個の感情極性から２^ｍ個の要素を持つ目的値集合Ｐが生成される。但し、ｍ＝１のときは、単一の感情極性集合Ｑを集合Ｐにセットする。目的値導出部３０のフローチャートを図７に示す。

For example, when three emotional polarities Q ₁ = {happy, sad}, Q ₂ = {surprise, expectation}, Q ₃ = {anger, fear} are entered, P = {(happy, surprise, anger) , (Happy, surprise, fear), (happy, expectation, anger), (happy, expectation, fear), (sad, surprise, anger), (sad, surprise, fear), (sad, expectation, anger), ( Sad, anticipation, fear)}. That is, a target value set P having 2 ^m elements is generated from ^m emotion polarities. However, when m = 1, a single emotion polarity set Q is set to the set P. A flowchart of the target value deriving unit 30 is shown in FIG.

Step 201) The target value deriving unit 30 reads the emotion polarities O ₁ ,..., O _m from the evaluation polarity storage unit 13.

  Step 202) If m ≧ 2, proceed to Step 103, otherwise proceed to Step 208.

  Step 203) The primary variable i is initialized (i ← 1).

  Step 204) If i ≦ m, go to Step 205, otherwise go to Step 207.

Step 205) generates a set _{Q i} to the original polarity value of semantic orientation _{O i.}

  Step 206) Set i = i + 1 and return to Step 204.

Step 207) If i> m in Step 204, set the Cartesian product set of Q ₁ ,..., Q _m to the target value set P, and go to Step 209.

Step 208) In step 202, if m <2, a target value set P based on the polarity value of the emotion polarity O ₁ is generated.

  Step 209) The target value set P is stored in a memory (not shown) in the target value deriving unit 30.

  The numerical attribute correlation rule extraction unit 40 includes a transaction stored in the transaction storage unit 15, a target value set P derived by the target value deriving unit 30, a region family stored in the region group storage unit 14, and a threshold storage unit. The optimal certainty factor attribute correlation rule is extracted based on the minimum support degree stored in FIG.

  The numerical attribute correlation rule extraction unit 40 sequentially obtains the optimal time, space region, and spatio-temporal region for all elements of the target value set P obtained by the target value deriving unit 30.

Element _{p = [q 1, ...,} q m] of the target value set P │q ₁ ∈Q _1, ..., optimum time for _{q m} ∈Q _m, consider the case of obtaining the spatial domain, the space-time region. First, the numerical attribute correlation rule derivation unit 40 applies the transaction polarity attribute part p ′ = [q ′ ₁ ,..., Q ′ _m ] | q ′ ₁ εQ to the transaction stored in the transaction storage unit 15. ₁ ,..., Q ′ _m εQ _m is assigned a purpose attribute E for determining whether or not p is equal to p. In the present embodiment, the objective attribute E is 1 when p = p ′, and 0 otherwise. FIG. 8 shows an example in which a purpose attribute is added to the example of FIG.

  The numerical attribute correlation rule extraction unit 40 extracts a one-dimensional numerical attribute correlation rule when obtaining the optimum time. Further, when obtaining an optimal space region, a two-dimensional numerical attribute correlation rule is extracted. Further, when obtaining the optimal space-time region, a three-dimensional numerical attribute correlation rule is extracted. Numeric attribute association rules are represented in the following format:

(A∈ (v ₁ , v ₂ )) → (E = 1)
A is a numerical attribute, v ₁ ≦ v ₂ is a value in the domain of A, and E is a target attribute. The term on the left of the arrow is called the condition part, and the term on the right is called the conclusion part. Support and confidence are used as a measure of the usefulness of the association rule. Number of transactions where N is the total number of transactions, s is the number of transactions whose attribute A is included in v ₁ ≦ v ₂ , the number of transactions whose attribute A is included in v ₁ ≦ v ₂ , and 1 is the value of attribute E If h is h, the support can be calculated by h / N and the certainty can be calculated by h / s. Further, in consideration of ease of setting, the degree of support may be calculated as h. Further, rules having 1, 2, and 3 numeric attributes in the condition part are referred to as one-dimensional, two-dimensional, and three-dimensional numeric attribute correlation rules, respectively.

  In order to obtain the optimum time, a one-dimensional numerical attribute correlation rule of the following format is extracted.

(Tε [t1, t2]) → (E = 1)
Here, T is a time attribute, and t1 ≦ t2 is a value in the T domain. For example, the fact that “a person has a high probability of being happy from January 1, 2008 to January 2, 2008” is expressed by the following association rule.

(T∈ [2008-01-01,2008-01-02]) → (E = 1)
As described above, the target attribute E is an attribute that takes 1 when the polarity value is “happy”. From the transaction set stored in the transaction storage unit 15, select a rule (optimum certainty correlation rule) having a certain degree of support (a minimum support) or more and the highest certainty among them. . If there are a plurality of rules that maximize the certainty factor, the rule that maximizes the support factor is preferentially selected. The time region indicated by the condition part of the rule is the optimum region that best characterizes the value of the target attribute.

  In order to obtain the optimum spatial region, a two-dimensional numerical attribute correlation rule having the following format is extracted.

(<L, A> ∈R) → (E = 1)
Here, L is a latitude attribute, and A is a longitude attribute. R is the numerical attributes L and A and the planar area that they stretch. The format of the region R is a region group stored in the region group storage unit 14. As described above, typical region groups in two-dimensional numerical attributes are:
1) rectangular area,
2) x monotonic region,
3) A rectangular convex region.

  The rectangular area of 1) is expressed by the direct product of the sections of two numerical attributes, and is an area parallel to the axis on the plane created by the two numerical attributes. The association rule of the format corresponds to the rectangular area.

(L∈ [35.0000,36.0000]) ∧ (A∈ [140.0000,141.0000]) → (E = 1)
From a transaction set stored in the transaction storage unit 15, a certain level or more (minimum support level) is selected. If there are a plurality of rules that maximize the certainty factor, the rule that maximizes the support factor is preferentially selected. The spatial region <L, A> indicated by the condition part of the rule is set as the optimum region that best characterizes the value of the objective attribute. This method has a possibility of deriving different optimum solutions depending on whether the assumed region group is (1) rectangular region, (2) x monotone region, or (3) rectangular convex region.

  To find the optimal spatio-temporal region, consider a three-dimensional numerical attribute correlation rule of the form

(<T, L, A> ∈R) → (E = 1)
Here, R is a numerical attribute T, L, A and a three-dimensional space region spanned by them. The format of the region R is a region group stored in the region group storage unit 14. The shape of the region includes a region parallel to the spatial axis created by the three numerical attributes, a connected region in which the intersection with a straight line perpendicular to one axis is empty, or two A connection region where the intersection with a straight line perpendicular to the axis is empty in one section or a connection region where the intersection with a straight line perpendicular to all axes is empty in one section is conceivable.

  From the transaction set stored in the transaction storage unit 15, a rule (optimum certainty rule) that has a certain degree of support (greater than the minimum support) and the highest certainty among them is selected. If there are a plurality of rules that maximize the certainty factor, the rule that maximizes the support factor is preferentially selected. The spatiotemporal region <T, L, A> indicated by the condition part of the rule is the optimum region that best characterizes the value of the target attribute. This method may lead to different optimal solutions depending on the assumed region family.

  There is no limitation on a specific method for obtaining each numerical attribute correlation rule, such as a simple method of enumerating all the sections and areas and selecting a section having the highest certainty factor, or a more efficient existing technique.

  For example, a one-dimensional numerical attribute correlation rule divides a numerical attribute into M sections without bias, and a sequence of points on a two-dimensional plane for k = 1,.

を考える。ここで、ｕ_ｉは区画ｉに含まれるトランザクション数で、ｖ_ｉは区画ｉに含まれるトランザクションの中で、目的属性の値がユーザからの要求を満たすトランザクション数である。ｘ方向に最小支持度以上離れた２点で、それらを結ぶ直線の傾きが最大のものを発見するというように、幾何学の問題に置き換えて効率的に求める手法が知られているのでこれを使ってもよい（例えば、Mining optimized association rules for numeric attributes, T. Fukuda, Y. Morimoto, S. Morishita, and T. Tokuyama, ACM SIGACT-SIGART Symposium on Principles of Database Systems, pp.182-191, 1996）。矩形領域は、二次元、三次元数値属性相関ルールも、一次元数値属性相関ルールの問題に還元することで効率的に解くことができる。また、二次元数値属性相関ルールのｘ単調領域や直方凸領域を効率的に求める手法も知られている（例えば、Data mining using two-dimensional optimized association rules: Scheme, algorithms, and visualization, T. Fukuda, Y. Morimoto, S. Morishita, and T. Tokuyama, ACM SIGMOD Conference on Management of Data, pp.13-23, 1996）。

think of. Here, u _i is the number of transactions included in the partition i, and v _i is the number of transactions whose purpose attribute value satisfies the request from the user among the transactions included in the partition i. There is a known method for efficiently finding this by replacing it with a geometrical problem, such as finding the one with the largest slope of the straight line connecting them at two points that are more than the minimum support in the x direction. (E.g. Mining optimized association rules for numeric attributes, T. Fukuda, Y. Morimoto, S. Morishita, and T. Tokuyama, ACM SIGACT-SIGART Symposium on Principles of Database Systems, pp.182-191, 1996 ). The rectangular area can be efficiently solved by reducing the two-dimensional and three-dimensional numerical attribute correlation rules to the problem of the one-dimensional numerical attribute correlation rules. There are also known methods for efficiently obtaining x-monotone regions and rectangular convex regions of two-dimensional numerical attribute association rules (for example, Data mining using two-dimensional optimized association rules: Scheme, algorithms, and visualization, T. Fukuda , Y. Morimoto, S. Morishita, and T. Tokuyama, ACM SIGMOD Conference on Management of Data, pp. 13-23, 1996).

  The optimal solution storage unit 16 stores combinations of objective values, numerical attribute correlation rules, and indices (support level, confidence factor) indicating superiority of various rules obtained by the numerical attribute correlation rule extraction unit 40. Only the optimal solution (the rule having the highest certainty factor) may be stored, or all the numerical attribute correlation rules having the support level equal to or higher than the minimum support level may be stored. That is, all the information obtained by the numerical attribute correlation rule extraction unit 40 can be stored. For example, it is stored in a specific area of a database or a general-purpose storage device (a memory, a hard disk device, etc.) provided in advance.

  The output unit 50 indicates the superiority of the objective value, the numerical attribute correlation rule (or the time domain, the spatial domain, the spatio-temporal domain of the condition part of the correlation rule), and the correlation rule stored in the optimal solution storage unit 16. Output indicators (support level, confidence level). Here, the output is a concept including display on a display, printing on a printer, sound output, transmission to an external device, accumulation in a recording medium, and the like. The output unit 50 may or may not include an output device such as a display or a speaker. The output unit 50 can be realized by output device driver software, output device driver software, and an output device.

  In the above embodiment, the operation in which the numerical attribute correlation rule extraction unit 40 generates a numerical attribute correlation rule will be described using an example in which an optimal space region is obtained. Here, it is assumed that two numerical attributes, a latitude attribute L and a longitude attribute A, are assigned as transaction position information. Further, it is assumed that the region group designated by the user is a rectangular convex region. Further, it is assumed that the frequency “6” is designated as the minimum support level (for the sake of simplicity, the support level is defined on a frequency basis).

  First, in order to obtain an optimized region, a domain defined by a plane formed by two numerical attributes L and A is divided into grids of appropriate granularity, and pixels on the grid are connected to create a region. Next, each transaction stored in the transaction storage unit 15 is mapped on a plane spanned by L and A based on position (latitude, longitude) information. At this time, the transaction having the value of the target attribute “1” and the transaction “0” are mapped in a discriminable format. FIG. 9 is an example of a region in which a transaction having a target attribute value “1” is represented by a black circle and a transaction “0” is represented by a white circle and mapped based on latitude and longitude.

  When a rectangular convex area having a support level equal to or higher than the minimum support level is obtained from the area shown in FIG. 9 using the numerical attribute correlation rule algorithm, as shown in FIG. A rectangular convex area B and a rectangular convex area C) are obtained. Of these, the rectangular convex region with the highest certainty, that is, the rectangular convex region A is the optimal space region.

  The operations of the components of the spatio-temporal search device described above can be constructed as a program, installed in a computer used as the spatio-temporal search device, executed, or distributed via a network.

  Further, the constructed program can be stored in a portable storage medium such as a hard disk, a flexible disk, or a CD-ROM, and can be installed or distributed in a computer.

  The present invention is not limited to the above-described embodiment, and various modifications and applications can be made within the scope of the claims.

  The present invention can be applied to a technique for classifying personally transmitted information such as a blog in space-time.

DESCRIPTION OF SYMBOLS 10 Input means, Input part 11 Input information storage means, Analysis object data storage part 12 Threshold value storage part 13 Emotion polarity storage part 14 Area group storage part 15 Transaction storage means, transaction storage part 16 Optimal solution storage part 20 Transaction generation function part 21 Document polarity determination unit 22 Transaction generation unit, transaction generation unit 30 Objective value calculation unit, objective value derivation unit 40 Numeric attribute correlation rule extraction unit, Numeric attribute correlation rule extraction unit 50 Output unit

Claims (9)

The most suitable for characterizing a specific person from a document in which personal experience is described in a natural language and the analysis target data of the structure in which the experienced location information and time information are given as numerical data. A spatio-temporal search device for obtaining a spatio-temporal region,
Input means for inputting the analysis target data, constraints on the space-time region to be obtained, conditions for characterizing people, and storing them in storage means;
A target value deriving unit that sets a condition characterizing the people given by the input unit as a target value for obtaining an optimal space-time region;
Document polarity determination means for determining whether the content of the text of the analysis target data is written by people who meet the conditions characterizing the people;
Based on the determination result of the document polarity determination means and the position information and time information given to the analysis target data, a transaction having the position information, time information, and determination result as elements is generated and stored in the transaction storage means. Transaction generation means;
Extracting a numerical attribute correlation rule from a set of transactions in the transaction storage means based on a constraint on the spatio-temporal region to determine a spatio-temporal region condition that maximizes the appearance probability of the transaction corresponding to the target value. Numeric attribute association rule extraction means for deriving;
A spatio-temporal search device characterized by comprising: The input means includes
The means for characterizing the people includes means for characterizing people in terms of whether they have a certain emotion, or vice versa, by specifying an emotion polarity consisting of two emotions having opposite meanings,
The target value deriving means includes
Each of the two emotions given by the input means includes means for obtaining a target value for obtaining an optimal spatiotemporal region,
The document polarity determination means includes
The spatio-temporal search device according to claim 1, further comprising means for determining which of the two emotions, which are polar values of emotion polarity, the content of the text of the analysis target data belongs to. The input means includes
Means for characterizing people in terms of accepting multiple emotional polarities consisting of two emotions with opposite specified meanings as conditions for characterizing the people,
The target value deriving means includes
For each emotion polarity given by the input means, including means for setting a direct product set of a set based on two opposing emotions constituting the emotion polarity as a target value,
The document polarity determination means includes
The spatio-temporal search device according to claim 1, further comprising means for determining, for each emotional polarity, the content of the text of the analysis target data belongs to two emotions which are polarity values of the emotional polarity. 4. The spatio-temporal search device according to claim 1, further comprising output means for outputting an optimal spatiotemporal region that best characterizes a specific person. The most suitable for characterizing a specific person from a document in which personal experience is described in a natural language and the analysis target data of the structure in which the experienced location information and time information are given as numerical data. A spatiotemporal search method for obtaining a spatiotemporal region,
An input step for inputting the analysis target data, restrictions on the space-time region to be obtained, conditions for characterizing people, and storing them in storage means;
A target value deriving step in which the conditions characterizing the people given in the input step are set as target values for obtaining an optimal spatiotemporal region;
A document polarity determination step for determining whether the text content of the analysis target data is written by people corresponding to a condition characterizing the people;
Based on the determination result of the document polarity determination step and the position information and time information given to the analysis target data, a transaction having the position information, time information, and determination result as elements is generated and stored in the transaction storage means. A transaction generation step;
Extracting a numerical attribute correlation rule from a set of transactions in the transaction storage means based on a constraint on the spatio-temporal region to determine a spatio-temporal region condition that maximizes the appearance probability of the transaction corresponding to the target value. A numerical attribute association rule extraction step to be derived;
A spatio-temporal search method characterized by: In the input step,
Characterize people in terms of whether they have a certain emotion, or vice versa, by specifying an emotion polarity consisting of two emotions with opposite meanings as a condition that characterizes the people,
In the target value derivation step,
Each of the two emotions given in the input step is set as a target value for obtaining an optimal spatiotemporal region,
In the document polarity determination step,
The spatio-temporal search method according to claim 5, wherein the text content of the analysis target data is determined to belong to one of two emotions, which is a polarity value of emotion polarity. In the input step,
As a condition for characterizing the people, it accepts input of a plurality of emotion polarities consisting of two emotions having opposite meanings specified, and characterizes people in terms of whether to hold a plurality of emotions simultaneously,
In the target value derivation step,
For each emotion polarity given in the input step, the objective value is a Cartesian product set of a set based on two opposing emotions constituting the emotion polarity,
In the document polarity determination step,
6. The spatiotemporal search method according to claim 5, wherein, for each emotion polarity, the content of the text of the analysis target data is determined to belong to two emotions which are polarity values of the emotion polarity. 8. The spatio-temporal search method according to claim 5, further comprising an output step of outputting an optimal spatio-temporal region that best characterizes a specific person.   A spatio-temporal search program for causing a computer to function as each means constituting the spatio-temporal search device according to any one of claims 1 to 4.

Images