TW202336607A - Information processing system and method of processing information - Google Patents

Abstract

An information processing system obtains a first data set for a plurality of first entities. A second data set is obtained for a plurality of second entities. A dimensionality reduction operation is performed on a group of the first feature vectors identified from the first data set and a group of the second feature vectors identified from the second data set. Thereby, a group of first low-dimensional feature vectors corresponding to the group of the first feature vectors and a group of second low-dimensional feature vectors corresponding to the group of the second feature vectors are generated. Each of the first entities is associated with at least one of the second entities based on the group of the first low-dimensional feature vectors and the group of the second low-dimensional feature vectors.

Description

Information processing systems and information processing methods

The present invention relates to an information processing system and an information processing method.

In the past, customers’ purchasing behavior was analyzed based on product sales data. It also analyzes customers’ contact behavior with mass media and online content. Collect various information about customers through questionnaires and/or face-to-face questions.

There is known a data fusion technology that combines a plurality of data collected using different means based on a common variable. In particular, a technology for data fusion between a first data set and a second data set has been disclosed, wherein the first data set has first characteristic data of each customer of a plurality of first customers, and the second data set has It is the second characteristic data of each customer including a plurality of second customers (for example, refer to Patent Document 1). Data fusion is to combine the first characteristic data and the second characteristic data of similar customers based on the shared variables between the first data set and the second data set, for example, based on the customer's demographic attributes. [Patent Document]

Patent Document 1 is Japanese Patent Application Laid-Open No. 2016-126609.

In the conventional data fusion technology, in order to use common variables to identify similar customers, common variables about customers are required between the first data set and the second data set that are the objects of fusion. Therefore, data that do not share variables cannot be combined with each other.

Therefore, according to a part of the present invention, it is expected to provide a technology that can achieve creation based on a first data set about a plurality of first entities and a second data set about a plurality of second entities without sharing variables. The correspondence between the first entity and the second entity.

According to a part of the present invention, an information processing system is provided. The information processing system includes a first acquisition part, a second acquisition part, a dimension reduction part and a correspondence creation part. The first acquisition part is configured to acquire a first data set regarding a plurality of first entities. The first data set may describe respective characteristics of a plurality of first entities.

The second acquisition part is configured to acquire a second data set regarding a plurality of second entities. The second data set may describe respective characteristics of a plurality of second entities.

The dimensionality reduction unit is configured to generate the same features as the first features by performing dimensionality reduction processing on a group of first feature vectors specified from the first data set and a group of second feature vectors specified from the second data set. The group of vectors corresponds to the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors corresponds to the group of second feature vectors. The group of second low-dimensional feature vectors may be a group of feature vectors having the same number of dimensions as the group of first low-dimensional feature vectors.

Each of the first feature vectors may represent a feature of a corresponding one of the plurality of first entities. Each of the second feature vectors may represent a feature of a corresponding one of the plurality of second entities.

The correspondence establishment unit is configured to associate each of the plurality of first entities with at least one of the plurality of second entities based on the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors.

In the case where the set of first entities and the set of second entities are subsets from a common parent or subsets from parents that are related to each other, even if there is no commonality between the first feature vector and the second feature vector Variables can also be used to express the characteristics of the first entity and the characteristics of the second entity by a combination of common components or a combination of components related to each other through dimensionality reduction.

That is, through dimensionality reduction, main feature components common to each other or main feature components related to each other can be extracted from the first feature vector and the second feature vector. Therefore, the degree of fitness between the first entity and the second entity can be appropriately discerned based on the comparison of the low-dimensional feature vectors.

Therefore, according to part of the present invention, the first entity and the second entity can be appropriately established based on the first data set regarding the plurality of first entities and the second data set regarding the plurality of second entities without sharing variables. Entity correspondence.

According to one part of the present invention, the correspondence establishing unit can be based on the similarity between the first entities specified from the group of the first low-dimensional feature vectors and the similarity between the second entities specified from the group of the second low-dimensional feature vectors. , establishing a correspondence between each of the plurality of first entities and at least one of the plurality of second entities, so that the mutual relationship between the first entities with respect to similarity is adapted to the mutual relationship between the second entities.

In the case where the set of the first entity and the set of the second entity are subsets from a common parent or are subsets from a parent that are related to each other, the mutual relationship regarding the similarity between the entities is the same as the parent, in The set of first entities and the set of second entities are generally common to or related to each other.

Therefore, each of the plurality of first entities is associated with at least one of the plurality of second entities so that the mutual relationship between the first entities with respect to the degree of similarity is adapted to the mutual relationship between the second entities, thereby enabling Each of the first entities is corresponding to an appropriate second entity with high identity or strong correlation.

According to one part of the present invention, a group of first low-dimensional feature vectors may be defined according to a first feature space. A group of second low-dimensional feature vectors may be defined according to the second feature space.

The correspondence establishment part may search for a mapping (Mapping) for mapping the plurality of first entities on the first feature space to the second feature space, so that the group of first low-dimensional feature vectors is specified in the first feature space. The distribution of the plurality of first entities is adapted to the distribution of the plurality of second entities in a second feature space specified from the group of second low-dimensional feature vectors.

The correspondence establishing unit may be configured to associate each of the plurality of first entities with at least one of the plurality of second entities based on the mapping.

According to a part of the present invention, the correspondence creation unit may be configured to search for the matrix Ω that maximizes the value Z(Ω) according to the following mathematical expression including the matrix K, the matrix L, and the matrix H as the matrix Ω ^* , and based on the matrix Ω ^* , establishing a correspondence between each of the plurality of first entities and at least one of the plurality of second entities. T is the transpose symbol. The trace system is the diagonal sum of the matrix X. [Mathematical formula 1]

Matrix K can be a matrix with N rows and N columns. The number of first entities may be N. The number of second entities may be the same as the number of first entities. The matrix K may be a first similarity matrix in which the value of the element in the i-th row and j-th column represents the similarity between the i-th entity and the j-th entity among the plurality of first entities.

The value system of the element in the i-th row and j-th column in the matrix K may be based on the first low-dimensional feature vector of the i-th entity among the plurality of first entities and the first low-dimensional feature vector of the j-th entity among the plurality of first entities. Calculated from low-dimensional feature vectors.

The matrix L can be a matrix with N rows and N columns. Matrix L system, the value of the element in the i-th row and j-th column represents a second similarity matrix representing the similarity between the i-th entity and the j-th entity among the plurality of second entities.

The value system of the element in the i-th row and j-th column in the matrix L may be based on the second low-dimensional feature vector of the i-th entity among the plurality of second entities and the second low-dimensional feature vector of the j-th entity among the plurality of second entities. Calculated from low-dimensional feature vectors.

The matrix H can be a matrix with N rows and N columns. The matrix H can be expressed as follows. When the value series of the element in the i-th row and j-th column is i=j, it means that the value is 1-1/N. When the value series of the element in the i-th row and j-th column is i≠j, it means that the value is 1-1/N. is a matrix of 0.

According to one aspect of the present invention, the correspondence creation unit can change the dimensionality reduction method in the dimensionality reduction process based on the matrix Ω ^* . For example, the correspondence establishment unit may change the dimensionality reduction method in the dimensionality reduction process so that the first low-dimensional feature vector and the third low-dimensional feature vector in the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors correspond to each other. The distance between the two low-dimensional feature vectors in the feature space is shortened.

According to a part of the present invention, the correspondence establishment unit may be configured to perform a re-search process on the matrix Ω ^* in a loop until a predetermined condition is satisfied, thereby improving the matrix Ω ^* , and based on the improved matrix Ω ^* , the plurality of th Each one of the entities is corresponding to at least one of the plurality of second entities.

The re-search process may include changing the dimensionality reduction method in the dimensionality reduction process based on the matrix Ω ^* . The re-search process may include causing the dimensionality reduction unit to perform the dimensionality reduction process in an updated dimensionality reduction manner, and re-searching based on the newly obtained group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors. Search matrix Ω ^* .

According to the information processing system including the correspondence establishment unit configured as described above, the process of establishing correspondence between the first entity and the second entity can be executed with high accuracy.

According to a part of the present invention, the first data set may include a plurality of first characteristic data. Each of the plurality of first characteristic data may represent a characteristic of a corresponding one of the plurality of first entities. The second data set may contain a plurality of second characteristic data. Each of the plurality of second characteristic data may represent a characteristic of a corresponding one of the plurality of second entities.

According to a part of the present invention, the information processing system may further include a data fusion unit. The data fusion unit may be configured to combine the plurality of second characteristic data with each of the plurality of first characteristic data based on the correspondence between the plurality of first entities and the plurality of second entities established by the correspondence creation unit. , from which an extended data set is generated. An extension set can contain multiple extensions. Each of the plurality of extended data may be data generated by combining a corresponding first characteristic data and a corresponding second characteristic data.

According to the above information processing system, it is possible to generate a data set with a large amount of information by combining a plurality of data sets.

According to part of the invention, the first entity may be a human being. The second entity can contact people. The first data set may be a data set describing the first characteristics of each of the plurality of individuals belonging to the first group. The second data set may be a data set describing the second characteristics of each of the plurality of individuals belonging to the second group.

It can be considered that characteristics such as people's behavior and attention will greatly affect demographic attributes, and even among different groups of people, the distribution of characteristics corresponding to demographic attributes will not change significantly. Therefore, according to the information processing system which is a part of the present invention, it is possible to appropriately establish correspondence between people in different groups without sharing variables.

According to a part of the present invention, the combination of the first characteristic and the second characteristic may be related to characteristics of purchasing behavior, characteristics of movement in at least one of online space and offline space, and/or characteristics of spatial perception. A combination of features for visiting multiple locations. The correspondence and even data fusion of entities established based on data sets related to the above characteristics help analyze human behavior.

According to a part of the present invention, the identification information of the information terminal corresponding to each of the plurality of second entities can be associated with the second data set.

According to a part of the present invention, the information processing system may include a selection unit that selects a set of second entities among the plurality of second entities that have been corresponding to any one of the plurality of first entities through the correspondence establishment unit. At least part of the information content is used as the transmission target.

According to a part of the present invention, the information processing system may include a transmission unit configured to transmit the information content to a set of information terminals corresponding to the transmission target of the information content based on the identification information.

The information processing system performs meaningful functions when the first entity and the second entity are human beings. According to the above transmission method, even when the relationship between the first entity and the information terminal is unclear, the identification information of the information terminal that has been associated with the second entity can be utilized to appropriately transmit the information content to the first entity. The information terminal of the second entity.

According to a part of the present invention, the selecting unit may be configured to select a first set and a second set as the transmission target of the information content, wherein the first set is associated with any one of the plurality of first entities through the corresponding establishing unit. A set of corresponding second entities has been established; the second set is a set among a plurality of second entities that has characteristics similar to those of the first set. According to the above selection of the transmission object, the transmission object can be expanded in an appropriate range based on the second data set, and the information content can be transmitted.

According to a part of the present invention, the second data set may be a data set describing characteristics of behavior of each entity of the plurality of second entities. In this case, the information processing system may include an inference part, which is about one or more entities of interest, and for each entity of interest, calculates an inference value about the behavior of the corresponding entity of interest. More than one entity of interest may be at least part of a plurality of first entities. The inferred value may be calculated based on the characteristics of the behavior of at least one entity of the plurality of second entities, wherein the plurality of second entities have established correspondence with the corresponding entity of interest. The first entity and the second entity may be related to people.

According to the information processing system including the above-mentioned inference unit, it is possible to infer the behavior of the first entity that cannot be identified based only on the first data set using the second data set. Presumptions can be predictions.

According to a part of the present invention, an information processing method corresponding to the method performed by the above-mentioned information processing system can be provided. According to a part of the present invention, an information processing method executed by a computer can be provided. The information processing method may include obtaining a first data set, wherein the first data set is a data set related to a plurality of first entities and describes respective characteristics of the plurality of first entities.

The information processing method may include obtaining a second data set, wherein the second data set is a data set related to a plurality of second entities and describes respective characteristics of the plurality of second entities.

The information processing method may include generating the first feature vector by performing dimensionality reduction on the group of first feature vectors specified from the first data set and the group of second feature vectors specified from the second data set. A group of first low-dimensional feature vectors corresponding to the group of and a group of second low-dimensional feature vectors corresponding to the group of second feature vectors. The group of second low-dimensional feature vectors may be a group of feature vectors having the same number of dimensions as the group of first low-dimensional feature vectors.

Each of the first feature vectors may represent a feature of a corresponding one of the plurality of first entities. Each of the second feature vectors may represent a feature of a corresponding one of the plurality of second entities.

The information processing method may include establishing a correspondence between each of the plurality of first entities and at least one of the plurality of second entities according to the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors.

According to a part of the present invention, establishing the correspondence may include based on the similarity between the first entities specified from the group of the first low-dimensional feature vector and the similarity between the second entities specified from the group of the second low-dimensional feature vector. degree, establishing a correspondence between each of the plurality of first entities and at least one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the degree of similarity is suitable for the mutual relationship between the second entities.

According to the above-mentioned information processing method, similarly to the above-mentioned information processing system, it is possible to realize the establishment of the second data set based on the first data set regarding the plurality of first entities and the second data set regarding the plurality of second entities without sharing variables. Correspondence between one entity and a second entity.

According to a part of the present invention, a computer program including commands for causing a computer to execute the above information processing method can be provided. According to a part of the present invention, a computer-readable recording medium storing a computer program can be provided.

According to a part of the present invention, a computer-readable recording medium storing a computer program can be provided, which is characterized in that after the computer loads and executes the computer program, the above information processing method can be completed.

According to a part of the present invention, a computer program product with a built-in computer program can be provided, which is characterized in that after the computer loads and executes the computer program, the above information processing method can be completed.

Hereinafter, illustrative embodiments of the present invention will be described with reference to the drawings. [First Embodiment] The information processing system 1 of this embodiment is configured by installing a dedicated computer program Pr in a general-purpose computer. As shown in Figure 1, the information processing system 1 includes a processor 11, a memory 13, a storage 15, a user interface 17 and a communication interface 19.

The processor 11 executes processing according to the computer program Pr stored in the memory 15 . The memory 13 is a first-level memory device including RAM, and is used as a work area when the processor 11 performs processing.

The storage 15 is a secondary memory device including, for example, a hard disk drive or a solid-state drive. In addition to the computer program Pr, it also stores various data for processing according to the computer program Pr.

The user interface 17 includes an input device and a display. The input device is configured to input operation signals from a user operating the information processing system 1 to the processor 11 . The display is configured to display various information to the user. Examples of input devices include keyboards and pointing devices.

The communication interface 19 includes a LAN (Local Area Network) interface and a USB (Universal Serial Bus) interface, and is used to communicate with external devices. The information processing system 1 sends and receives data with external devices through the communication interface 19 .

The processor 11 in the information processing system 1 generates an extended data set 15C by executing processing according to the computer program Pr. The extended data set 15C uses the second data set 15B to pair the first data set obtained from the external device through the communication interface 19. An extended data set generated by extending data set 15A.

The extended data set 15C is a data set in which the information included in the second data set 15B is added to the first data set 15A. Through expansion, the amount of information on each entity described in the first data set 15A increases. Entity systems such as people, especially individuals. The amount of information is increased for human behavior analysis and/or advertisement delivery based on the extended data set 15C.

Specifically, when an execution instruction from the user is input through the user interface 17, the processor 11 of the information processing system 1 executes the analysis process shown in FIG. 2 . When the analysis process shown in FIG. 2 is started, the processor 11 acquires the first data set 15A and the second data set 15B as data fusion targets (S110, S120).

In S110 and S120, the processor 11 may read out the first data set 15A and the second data set 15B pre-stored in the storage 15 from the storage 15. Thereby, the processor 11 can obtain the first data set 15A and the second data set 15B.

The user can specify the first data set 15A and the second data set 15B to be obtained. The user can collect the first data set 15A and the second data set 15B as data fusion objects in advance and store them in the storage 15 .

Alternatively, the processor 11 may obtain the first data set 15A from the first external device and the second data set 15B from the second external device through communication using the communication interface 19 .

The first data set 15A is a data set regarding a plurality of first entities, and is a data set describing first characteristics of each of the first entities. The first data set 15A is a set of first characteristic data, and each first characteristic data respectively represents the first characteristic of a corresponding entity among the plurality of first entities.

The second data set 15B is a data set regarding a plurality of second entities, and is a data set describing the second characteristics of each of the second entities. The second characteristic may be a different characteristic than the first characteristic. Specifically, the second data set 15B is a set of second characteristic data, and each second characteristic data respectively represents the second characteristic of a corresponding entity among the plurality of second entities.

The set of first entities and the set of second entities are, for example, different subsets of a common parent. The parent can be a collection of people or a collection of consumers. For example, the set of first entities may be a set of people corresponding to customers of the first enterprise. For example, the set of second entities may be a set of people that correspond to customers of a second business that is different from the first business.

Alternatively, the set of first entities may be a set of persons who are the collection objects of the first act. The set of second entities may be a set of persons who are the collection objects of the second act.

The first data set 15A shown in Figure 3A is data about the first set of entities and people, and includes characteristic data about each person's purchasing behavior. Each characteristic data is associated with the ID of the corresponding person, and uses two values of 1 or 0 to indicate whether the corresponding person purchased each of the plurality of products P1, P2, P3, ....

The second data set 15B shown in Figure 3B is data about the second set of entities and people, and includes characteristic data about each person's browsing behavior of web page content. Each characteristic data is associated with the ID of the corresponding person. For each website of the plurality of websites S1, S2, S3, ..., two values of 1 or 0 are used to indicate whether the corresponding person has visited the corresponding website and browsed the web page. content.

In S110, the processor 11 generates the M1-dimensional feature vector x=(x1, x2, x3, …). According to an example, the elements x1, x2, x3, ... of the feature vector x can respectively represent whether the corresponding person purchased the products P1, P2, P3, ....

Similarly, in S120, the processor 11 generates the M2-dimensional feature vector y=(y1, y2) of each second entity based on the acquired second feature data of each second entity included in the second data set 15B. ,y3,…). According to an example, the elements y1, y2, y3, ... of the feature vector y can respectively represent whether the corresponding person has browsed the web page content in the websites S1, S2, S3, ....

After that, the processor 11 transforms each feature vector x from an M1-dimensional feature vector into an M-dimensional feature vector smaller than the M1 dimension, that is, a low-dimensional feature, by performing dimensionality reduction processing (S130) on the group of feature vectors x. Vector Dx=(Dx1, Dx2,...). Thereby, the processor 11 generates a group of low-dimensional feature vectors Dx corresponding to the group of feature vectors x. The lower right area of Figure 3A shows an example of the low-dimensional feature vector Dx through a table.

The processor 11 further performs dimensionality reduction processing (S140) on the group of feature vectors y to transform each feature vector y from an M2-dimensional feature vector into an M-dimensional feature vector smaller than the M2 dimension, that is, a low-dimensional feature vector. Dy=(Dy1, Dy2,...). Thereby, the processor 11 generates a group of low-dimensional feature vectors Dy corresponding to the group of feature vectors y. The low-dimensional feature vector Dy is a feature vector having the same number of dimensions M as the low-dimensional feature vector Dx. The lower right area of Figure 3B shows an example of the low-dimensional feature vector Dy through a table.

As examples of algorithms for implementing mapping to low-dimensional spaces, Nonnegative Matrix Factorization, Latent Dirichlet Allocation, Singular value Decomposition, and Probabilistic Latent Semantics are known Analysis (probabilistic Latent Semantic Analysis). One of these algorithms may be used to perform the dimensionality reduction processing in S130 and S140.

According to the above algorithm, the feature vector can be low-dimensionalized to extract the main feature components that give strong characterization to each entity. Alternatively, the feature vectors may be reduced in dimensionality in a form such that less information is lost in distinguishing each entity.

After that, the processor 11 performs an alignment process (S150-S180), that is, an alignment process that calculates the relationship between the first entity and the second entity based on the group of low-dimensional feature vectors Dx and the group of low-dimensional feature vectors Dy. .

Alignment processing is performed using Kernelized Sorting technology. Details of the alignment process using kernel sorting are described below. However, alignment processing can also be achieved using adversarial learning, Gromov-wasserstein Alignment technology, or Unbalanced Optimal Transport technology.

In S150, the processor 11 generates a similarity matrix K regarding the set of first entities using the group of low-dimensional feature vectors Dx. The similarity matrix K is a square matrix with N rows and N columns. Here, N represents the number of low-dimensional feature vectors Dx. In other words, N represents the number of first entities.

The similarity matrix K is defined as a matrix in which the value Kij of the element in the i-th row and j-th column represents the similarity between the i-th entity and the j-th entity in the first entity set.

That is, the similarity matrix K is defined as a matrix describing the distribution of similarities between entities in the first entity set. In other words, the similarity matrix K is defined as a matrix that explains the distribution of the entities on the feature space using the scale of similarity between the entities with respect to the first entity set.

Specifically, it is calculated that the low-dimensional feature vector Dx of the i-th entity, that is, the low-dimensional feature vector Dx[i], and the low-dimensional feature vector Dx of the j-th entity, that is, the low-dimensional feature vector Dx[j], are substituted into the kernel function k The value k(Dx[i], Dx[j]) obtained from (a,b) is used as the similarity. That is, Kij=k(Dx[i], Dx[j]).

Examples of the kernel function k(a,b) include the Gaussian RBF (radial basis function) kernel function expressed by the following mathematical formula. The similarity calculated using this kernel function k(a,b) ranges from 0 to 1.

[Mathematical formula 2] According to the above kernel function k(a,b), the value Kij of the element of the similarity matrix K is 0<Kij≦1.

In S160, the processor 11 generates a similarity matrix L regarding the set of second entities using the group of low-dimensional feature vectors Dy. The similarity matrix L is a square matrix with N rows and N columns. Here, N represents the number of low-dimensional feature vectors Dy. In other words, N represents the number of second entities. That is, the number of first entities is the same as the number of second entities.

Similar to the similarity matrix K, the similarity matrix L is defined as, the value Lij of the element in the i-th row and j-th column represents the similarity between the i-th entity and the j-th entity in the second entity set. matrix. That is, the value of the element in the i-th row and j-th column is Lij=k(Dy[i], Dy[j]).

In the next S170, the processor 11 uses the similarity matrix K and the similarity matrix L, and searches for the matrix Ω that maximizes the value Z(Ω) as the matrix Ω ^* according to the following formula. [Mathematical formula 3] Here, the matrix H is a matrix with N rows and N columns, and when i=j, the value of the element in the i-th row and j-th column shows 1-1/N, and when i≠j, the value of the element in the i-th row and j-th column shows 1-1/N. The value of shows a diagonal matrix of 0. T is the transposition symbol. trace(X) is the diagonal sum of matrix X. The similarity matrices K and L are symmetric matrices. When the matrix Ω ^T L'Ω is found to be the ideal Ω of the transpose matrix of the matrix K', the value Z(Ω) is maximized.

The search matrix Ω ^* corresponds to the following. According to the similarity of the group-specific first entity from the low-dimensional feature vector Dx and the similarity of the group-specific second entity from the low-dimensional feature vector Dy, the plurality of first entities are Each of the entities is corresponding to at least one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the similarity is adapted to the mutual relationship between the second entities.

In other words, the search matrix Ω ^* corresponds to the following, searching for a mapping (Mapping) for mapping a plurality of first entities on the first M-dimensional feature space to the second M-dimensional feature space, so that the distribution of the first entity is suitable for the second M-dimensional feature space. The distribution of two entities, wherein the distribution of the first entity is the distribution of the first entity in the first M-dimensional feature space specified from the group of low-dimensional feature vectors Dx, and is the distribution defined by the similarity of the entities, and the The distribution of the two entities is the distribution of the second entity in the second M-dimensional feature space specified from the group of low-dimensional feature vectors Dy.

The left diagram in Figure 4A conceptually represents the distribution of the first entity, and the left diagram in Figure 4B conceptually represents the distribution of the second entity. The examples shown in Figure 4A and Figure 4B are only for illustrating the technology, and two-dimensional low-dimensional feature vectors Dx and Dy are defined. Each point marked with symbols E11, E12, E13, E14, E15, E16, and E17 represents the respective position of the first entity on the feature space. Each point marked with symbols E21, E22, E23, E24, E25, E26, and E27 represents the position of the second entity in the feature space.

As can be seen from Figure 4B, according to this example, the component Dy1 of the low-dimensional feature vector Dy corresponds to the component Dx2 of the low-dimensional feature vector Dx, and the component Dy2 of the low-dimensional feature vector Dy corresponds to the component Dx1 of the low-dimensional feature vector Dx.

That is, according to the example shown in FIG. 4A , the arrangement and arrangement of the respective entities of the first entity group and the second entity group are defined only in a different form between the similarity matrix K and the similarity matrix L. The order of dimensions essentially represents the similarity distribution of sets of identical entities.

When the group of the first entity and the group of the second entity have the properties of common or related groups based on reasons such as the same parent, through the low-dimensionalization of the feature vectors x and y, even as an information source There are no common variables between the first data set 15A and the second data set 15B, and it is possible to extract essentially common characteristic components of each entity.

However, even through the above-mentioned low-dimensionalization, the low-dimensional feature vectors Dx and Dy only have the same feature components, and the arrangement of the feature components cannot be aligned. In addition, the arrangement of entities between the first data set 15A and the second data set 15B is not aligned.

The search matrix Ω ^* corresponds to the following search operation, regarding the arrangement of entities and the arrangement of dimensions, using the identity of the similarity distribution as a clue to search for the correspondence between the unaligned feature vectors Dx and Dy.

In the next S180, the processor 11 establishes a correspondence between each of the first entities and at least one of the second entities according to the matrix Ω ^* . The element value of the i-th row and j-th column in the matrix Ω ^* represents, according to the distribution of similarity, the degree to which the i-th entity in the first entity set and the j-th entity in the second entity set correspond to each other or The magnitude of the possibility.

Each element of the matrix Ω ^* ideally takes a value of 0 or 1, and the sum of the element values in a row is 1 for each row, and the sum of the element values of a column is 1 for each column. When the matrix Ω ^* is the above-mentioned ideal matrix, the first entity corresponding to the row number of the element with a value of 1 and the second entity corresponding to the column number of the element correspond to each other.

That is to say, when the value of the element in the i-th row and j-th column in the matrix Ω ^* is 1, the i-th entity in the set of first entities and the j-th entity in the set of second entities correspond to each other.

However, in numerical calculations, the matrix Ω ^* rarely becomes the ideal matrix mentioned above. Therefore, in S180, each of the plurality of first entities is associated with at least one of the second entities according to any of the following methods.

(Method 1) In the i-th row of matrix Ω ^* , search for the element with the largest value. When the element with the largest value is in the c-th column, establish a correspondence between the i-th entity in the first entity set and the c-th entity in the second entity set. Do this for all rows.

In this method, one entity in the second entity may be corresponding to a plurality of first entities. To suppress this possibility, a proximity search can also be performed. As an example of proximity search, contextual dissimilarity measure is known.

(Method 2) In order to establish a strict one-to-one correspondence, by inputting the matrix Ω ^* and solving the optimal configuration problem, establish a correspondence between each of the plurality of first entities and one of the non-repeating second entities.

In S180, the processor 11 can further output the correspondence table shown in FIG. 5 as a table illustrating the correspondence relationship between the first entity and the second entity. That is, the processor 11 can output a correspondence table describing the ID of the second entity associated with each ID of the first entity and corresponding thereto, and store the correspondence table in the memory 15 .

The processor 11 further performs data fusion processing (S190). In the data fusion process, the processor 11 combines the first data set 15A and the second data set 15B based on the above-mentioned corresponding result or the above-mentioned correspondence table, and generates an extended data set 15C.

Extension set 15C contains multiple extensions. As shown in FIG. 6, each of the plurality of extended data is combined data generated by combining a first characteristic data and a second characteristic data corresponding to each other.

That is, the processor 11 combines each of the plurality of first characteristic data included in the first data set 15A with one of the plurality of second characteristic data included in the second data set 15B according to the correspondence table, This is used to generate extended data set 15C.

When the i-th entity in the first entity set has established a correspondence with the j-th entity in the second entity set, the processor 11 will explain the i-th entity in the first entity set according to the correspondence table. The first characteristic data of the characteristics and the second characteristic data describing the characteristics of the j-th entity in the set of second entities are combined to generate the above-mentioned extended data of the i-th entity.

The extended data set 15C generated in the above manner is stored in the storage 15 . For example, according to instructions input by the user through the user interface 17 , the extended data set 15C stored in the storage 15 is transmitted to other systems through the communication interface 19 .

Other systems may be, for example, advertising delivery systems. The advertisement delivery system can identify the entity of the advertisement delivery target according to the extended data set 15C and deliver the advertisement to the entity.

If the data fusion process ends in S190, the processor 11 ends the analysis process shown in Figure 2.

As explained above, according to the information processing system 1 of this embodiment, even if there are no common variables between the first data set 15A and the second data set 15B, the first entity and the second entity can be matched based on the distribution of similarity. establish correspondence.

In order to properly establish correspondence, it is preferable that the distributions of similarities between the set of first entities and the set of second entities are consistent with each other or similar to each other or correlated with each other.

When the set of first entities and the set of second entities are subsets from the same parent, the above preferred conditions are generally satisfied. Therefore, when the first entity and the second entity are persons, that is, when the data set representing the characteristics of the person is processed as the first data set 15A and the second data set 15B, the technology of this embodiment functions meaningfully. .

In particular, human behavior mostly shows tendencies that correspond especially to demographic attributes. Therefore, when the first data set 15A and the second data set 15B are data sets generated based on collected data from groups whose distributions of demographic attributes are presumed to be similar to each other, appropriate correspondence between entities can be achieved.

For example, even if the first data set 15A and the second data set 15B have no common variables and are data sets describing characteristics of people who belong to different groups, or data sets describing characteristics of different behaviors, the entity can be establish appropriate correspondence between them. Therefore, a data set useful for analyzing human psychology and behavior can be generated as the extended data set 15C.

According to the above example, the first data set 15A is a data set describing the characteristics of the purchasing behavior of a plurality of individuals belonging to the first group, and the second data set 15B is a data set describing the website access behaviors of a plurality of individuals belonging to the second group. and/or a collection of data that characterizes web content browsing behavior.

According to other examples, a data set describing characteristics of people's media exposure behavior, such as television viewing behavior, may also be used as one of the first data set 15A and the second data set 15B. As one of the first data set 15A and the second data set 15B, a data set describing characteristics of the usage status of a portable terminal such as a smartphone may be used.

A data set describing characteristics of people's movement in offline space (that is, real space) may also be used as one of the first data set 15A and the second data set 15B. The data set may, for example, describe features regarding visits to a plurality of places, movement paths and/or means of movement as features regarding a person's movement in an offline space.

A data set describing characteristics of human movement in an online space may be used as one of the first data set 15A and the second data set 15B. The data set may describe characteristics about human movement in a virtual reality (VR) space or net surfing as characteristics about human movement in an online space. A data set generated based on data collected through a questionnaire survey may also be used as one of the first data set 15A and the second data set 15B.

As a combination of the first data set 15A and the second data set 15B, a combination of a data set collected through a questionnaire survey and a data set on TV viewing behavior may be used, or a data set on movement history and a data set on purchases may be used. A combination of data sets.

In the above embodiment, ZCA whitening, normalization, standardization, etc. may also be performed on the group of low-dimensional feature vectors Dx and Dy.

In the above embodiment, the designer or user specifies the number of dimensions M of the low-dimensional feature vectors Dx and Dy. However, the information processing system 1 may also be configured to search for the most appropriate number of dimensions M. For example, the information processing system 1 can be configured to cyclically execute the analysis processing shown in FIG. 2 while changing the number of dimensions M for the same data sets 15A and 15B, and automatically select the most suitable one using the maximum value of Z (Ω) as an index. The number of dimensions M.

[Second Embodiment] The information processing system 1 of the second embodiment is configured such that the processor 11 executes the analysis process shown in FIG. 7 instead of the analysis process shown in FIG. 2 . The details of the analysis processing executed by the processor 11 will be selectively described below as an explanation of the second embodiment. It can be understood that the configuration of the information processing system 1 not mentioned in this embodiment is the same as that of the first embodiment.

When the processor 11 starts executing the analysis process shown in FIG. 7 , similarly to the first embodiment, the processor 11 acquires the first data set 15A and the second data set 15B that are data fusion targets (S310, S320).

Similar to the process in S110, the processor 11 generates the feature vector x of each first entity based on the first data set 15A (S310). Similar to the process in S120, the processor 11 generates the feature vector y of each second entity based on the second data set 15B (S320).

Similar to the processing in S130 and S140, the processor 11 further performs dimensionality reduction processing to generate a group of low-dimensional feature vectors Dx corresponding to the group of feature vectors x, and generates a group corresponding to the feature vector y. A group of low-dimensional feature vectors Dy (S330).

In the following S340, the processor 11 executes the same processing as the processing in S150, S160, and S170. That is, the processor 11 uses the group of low-dimensional feature vectors Dx to generate the similarity matrix K about the set of the first entity, and uses the group of low-dimensional feature vectors Dy to generate the similarity matrix L about the set of the second entity. .

Furthermore, the processor 11 uses the similarity matrix K and the similarity matrix L to search for the matrix Ω that maximizes the value Z(Ω) described in the first embodiment as the matrix Ω ^* (S340). Here, the searched matrix Ω ^* is expressed as a correspondence matrix Ω ^* .

After that, the processor 11 determines whether the loop end condition is satisfied (S350). When it is determined that the loop end condition is not satisfied (No in S350), the processor 11 executes the process of S360.

In S360, the processor 11 searches for a dimensionality reduction method that minimizes the cost of the Gromov-Wasserstein distance in the state of fixing the correspondence matrix Ω ^* that has been searched in S340.

The state of the fixed correspondence matrix Ω ^* corresponds to the state of the fixed correspondence between the first entity and the second entity. As described above, searching for the matrix Ω that maximizes the value Z(Ω) as the correspondence matrix Ω ^* corresponds to searching for a mapping for mapping a plurality of first entities on the first feature space to the second feature space ( Mapping), so that a plurality of first entities on the first feature space are suitable for the distribution of the second entity in the second feature space.

The cost system of Gromov-Wasserstein distance corresponds to the optimal transmission problem between the first entity and the second entity when mapping the set of the first entity to the second feature space. Transmission costs.

The cost of the Gromov-Wasserstein distance can be calculated using the similarity matrices K, L and the correspondence matrix Ω ^* . As mentioned above, the similarity matrix K is a matrix that contains the similarity between the first entities calculated based on the dimensionally reduced low-dimensional feature vector Dx in its elements. The similarity matrix L is a matrix that contains the similarity between the second entities calculated based on the dimensionally reduced low-dimensional feature vector Dy in its elements.

Searching for a dimensionality reduction method that minimizes the Gromov-Wasserstein distance corresponds to searching for a dimensionality reduction method for generating low-dimensional feature vectors Dx, Dy, where the low-dimensional feature vectors Dx, The Dy system optimally rationalizes the correspondence between the first entity and the second entity shown by the correspondence matrix Ω ^* .

The minimization of the cost corresponds to searching for a dimensionality reduction method according to the correspondence matrix Ω ^* so as to shorten the distance in the feature space between the first entity and the second entity that correspond to each other. In other words, to make the first entity have a lower The distance between the dimensional feature vector Dx and the low-dimensional feature vector Dy of the second entity is shortened in the feature space.

For example, when transforming an M1-dimensional feature vector x into an M-dimensional low-dimensional feature vector Dx, the transformation matrix Tx of M rows and M1 columns is applied to the feature vector x. When converting the M2-dimensional feature vector y into the M-dimensional low-dimensional feature vector Dy, the transformation matrix Ty of M rows and M2 columns is made to act on the feature vector y. At this time, the number of parameters m constituting the transformation matrices Tx and Ty is (M*M1+M*M2).

For example, the gradient method is used to search for the parameter m that minimizes the above cost as the parameter m of the transformation matrix Tx and Yy, thereby realizing the search for the dimensionality reduction method.

Afterwards, the processor 11 uses the searched dimensionality reduction method (for example, transformation matrices Tx, Ty) to reduce the dimensionality of the feature vectors x and y, and calculates new low-dimensional feature vectors Dx and Dy (S370).

The processor 11 uses the similarity matrix K based on the new low-dimensional feature vector Dx and the similarity matrix L based on the new low-dimensional feature vector Dy to search for the matrix Ω that maximizes the value Z(Ω) as a new correspondence matrix Ω ^* (S340).

By executing the processes of S360, S370, and S340 in this manner, the processor 11 searches again for a correspondence matrix Ω ^* with high matching accuracy together with searching for a better dimensionality reduction method.

When the loop end condition is satisfied (Yes in S350), the processor 11 executes the process of S380. For example, when the process of S340 has been executed a predetermined number of times, or when the change amount of the correspondence matrix Ω ^* found by re-searching does not reach the predetermined amount, the loop end condition is satisfied.

In S380, similarly to the process of S180 in the first embodiment, the processor 11 establishes each of the first entities and at least one of the second entities based on the correspondence matrix Ω ^* calculated at the end of the loop process. correspond. The processor 11 can further memorize and output a correspondence table describing the correspondence between the first entity and the second entity.

Thereafter, similarly to the processing in S190, the processor 11 performs data fusion processing to combine the first data set 15A and the second data set 15B, generate an extended data set 15C, and combine the generated extended data The set 15C is stored in the storage 15 (S390).

The information processing system 1 of the second embodiment described above can establish a more accurate correspondence between the first entity and the second entity through the above-mentioned loop processing. Therefore, an extended data set 15C with better accuracy can be generated.

[Third Embodiment] The information processing system 1 of the third embodiment is configured such that the processor 11 executes the evaluation process shown in FIG. 8 based on an execution instruction issued by the user through the user interface 17 . As an explanation of the third embodiment, the details of the evaluation process executed by the processor 11 will be described below. It can be understood that the configuration of the information processing system 1 not mentioned in this embodiment is the same as that of the first or second embodiment.

The evaluation process is performed to evaluate whether the data set to be evaluated is an excellent data set that can perform correspondence and data fusion with high accuracy in the analysis process shown in FIG. 2 or FIG. 7 . The data set to be evaluated corresponds to a data set that can be used as the first data set 15A or the second data set 15B in analysis processing.

When the execution of the evaluation process is started, the processor 11 obtains the execution instruction from the user and the data set of the designated evaluation target (S410). The processor 11 can obtain the data set of the designated evaluation object from the storage 15 .

Afterwards, the processor 11 generates a first feature vector x_1 and a second feature vector x_2 for each entity based on the data set of the evaluation object (S420). The data set of the evaluation object can include characteristic data representing the characteristics of the entity corresponding to each entity using (Q1 + Q2) elements.

The processor 11 can divide (Q1+Q2) elements into a first element group including Q1 elements and a second element group including Q2 elements. Each of the (Q1+Q2) elements can be randomly classified into either the first element group or the second element group.

The processor 11 can generate a first feature vector x_1 and a second feature vector x_2 for each entity based on the data set of the evaluation object, where the first feature vector x_1 describes the characteristics of the first element group of the corresponding entity, and the The two characteristic vectors x_2 describe the characteristics of the second element group of the corresponding entity.

For example, if the data set of the evaluation object contains each entity, the feature vector v=(v[1], v[2], v[3],..., v[Q] can be generated in S110, S120, S310 or S320 ]), the first feature vector x_1=(v[1], v[2],..., v[Q1]) containing Q1 elements can be generated and contains The second eigenvector x_2 of Q2 elements = (v[Q1+1], v[Q1+2],..., v[Q1+Q2]).

The first feature vector x_1 is the feature vector x corresponding to each entity in the set of first entities, and the second feature vector x_2 is corresponding to each entity in the set of second entities that is the same as the set of first entities. eigenvector y.

Thereafter, the processor 11 performs the same processing as the processing performed in S130 to S170 on the first feature vector x_1 and the second feature vector x_2 in S430 and S440.

In S430, similar to the processing in S130 and S140, the processor 11 performs dimensionality reduction processing on the first feature vector x_1 of each of the first entities and the second feature vector x_2 of each of the second entities. , and generate low-dimensional feature vector Dx_1 and low-dimensional feature vector Dx_2 with the same number of dimensions.

The processor 11 generates a similarity matrix corresponding to the similarity matrix K and representing the similarity of the low-dimensional feature vectors Dx_1 between the first entities based on the low-dimensional feature vector Dx_1 of each of the first entities. The processor 11 further generates a similarity matrix corresponding to the similarity matrix L and representing the similarity of the low-dimensional feature vectors Dx_2 between the second entities based on the low-dimensional feature vector Dx_2 of each second entity.

The processor 11 searches for the matrix Ω that maximizes the value Z(Ω) based on the similarity matrices as the correspondence matrix Ω ^* (S440).

Afterwards, regarding the set of first entities corresponding to the group of low-dimensional feature vectors Dx_1 and the set of second entities corresponding to the group of low-dimensional feature vectors Dx_2, the processor 11 calculates the correspondence matrix Ω ^* to correctly represent the first The degree of correspondence between one entity and the second entity is used as a score (S450).

Thereby, the processor 11 evaluates whether the evaluation target data set is an excellent data set that can perform correspondence and data fusion by analysis processing with high accuracy (S450).

When the processor 11 generates the feature vector x_1 of each entity of the first entity and the feature vector x_2 of each entity of the second entity in S420, the processor 11 can memorize the correct relationship between the first entity and the second entity in advance. corresponding relationship.

In an environment where the correct answer to the correspondence relationship has been memorized as described above, the processor 11 performs the same processing as the analysis process in S430 and S440 and calculates the correspondence matrix Ω ^* , and specifies the correspondence matrix Ω ^* The correspondence is compared with the correct answer.

For example, the processor 11 performs a process of correlating each of the first entities with one of the second entities in the same manner as the processes performed in S180 and S380 based on the correspondence matrix Ω ^* .

When the first entity and the second entity corresponding to each other are the same entity in the data set of the evaluation object, and the corresponding first entity and the second entity are established according to the correspondence matrix Ω ^* , the processor 11 determines that the correspondence has been successfully established, and in the case where they are not the same entity. , the processor 11 determines that the establishment of correspondence has failed.

The processor 11 can calculate the ratio of successfully establishing correspondence among all entities as the score of the data set to be evaluated (S450). After that, the processor 11 outputs the calculated score as the evaluation result (S460), and then ends the evaluation process.

When mapping and data fusion cannot be performed with high accuracy based on a data set, it can be inferred that the data set does not have sufficient information or data structure to achieve high-precision mapping and data fusion with respect to the set features.

Insufficient information also affects the accuracy of analysis and mapping and data fusion of two different data sets. Therefore, based on the above evaluation process, it can be estimated in advance whether the data set to be evaluated is a data set that can perform data fusion without common variables with high accuracy.

In S460, the processor 11 can convey to the user of the information processing system 1 whether the evaluation target data set is an excellent data set by outputting a score. Thereby, the user can use an appropriate combination of the first data set 15A and the second data set 15B in analysis processing, and obtain an extended data set 15C with high reliability.

In order to obtain the desired extended data set 15C, it may be considered that it is sufficient to use any one of a plurality of data sets that are similar to each other as the first data set 15A combined with the second data set 15B.

For example, consider a case where the first data set 15A on purchasing behavior and the second data set 15B on website access behavior/web content browsing behavior are combined to generate an extended data set 15C. In this case, it may be considered that it is sufficient to generate the extended data set 15C using a data set on the purchasing behavior of customers of any one of the plurality of distribution organizations as the first data set 15A.

Examples of a plurality of distribution organizations include a plurality of chain convenience stores. The data set on purchases at each convenience store chain may include information on similar purchasing behaviors at other convenience store chains as consumer purchasing behavior.

Therefore, it may be considered that it is sufficient to generate the extended data set 15C using a data set on the customer purchase behavior of any one of a plurality of chain convenience stores as the first data set 15A.

When there are a plurality of data sets as candidates for the first data set 15A (or the second data set 15B), the above-mentioned evaluation process can be used to select from the plurality of data sets from the viewpoint of establishing correspondence and data fusion accuracy. The most appropriate data set among the data sets is selected as the first data set 15A (or the second data set 15B).

For example, in any one of S110, S120, S310, and S320, the processor 11 may select one of the plurality of candidates for the data set to be the data fusion target by executing the selection process shown in FIG. 9 as necessary. Candidates to serve as data sets for data fusion objects. The data set as the data fusion target in S110 and S310 corresponds to the first data set 15A, and the data set as the data fusion target in S120 and S320 corresponds to the second data set 15B.

When the selection process shown in FIG. 9 is started, the processor 11 obtains a plurality of data sets as a plurality of candidates for the data set to be fused (S510). The processor 11 can obtain a plurality of data sets specified by the user from the storage 15 .

After that, the processor 11 sets one of the plurality of data sets as the data set to be evaluated (S520), and executes the evaluation process shown in FIG. 8 (S530). The processor 11 sets each data set as a data set to be evaluated (S520), and executes the evaluation processing (S530) in a loop until the evaluation processing on all the plurality of data sets is executed (in until Yes in S540). Thereby, the score calculated in S450 is obtained for each data set.

If the evaluation process has been performed on all the plurality of data sets and scores have been obtained (Yes in S540), the processor 11 uses the data set with the highest score among the plurality of data sets as the data set of the data fusion target (S550). After that, the selection process is ended. In S110, S120, S310, and S320, the processor 11 can generate a feature vector (x or y) based on the data set of the adopted data fusion object.

By performing the selection process as described above and selecting the most suitable data set from a plurality of candidates, the extended data set 15C with higher accuracy can be generated.

Supplementary explanation is as follows. In the example of purchasing behavior, the plurality of candidates for the data set as the data fusion target may include a plurality of data sets representing the consumer's purchasing behavior with different parameters. For example, the first candidate may be a data set that can generate a feature vector including the purchase number of each product in an element for each consumer as an entity. The second candidate may be a data set that can generate a feature vector including the purchase amount of each product in an element for each consumer as an entity.

Preparing multiple data sets using different parameters to describe the same characteristics mentioned above, and selecting a data set suitable for data fusion will help generate a better extended data set 15C.

[Fourth Embodiment] The transmission system 30 of the fourth embodiment shown in FIG. 10 uses the data fusion technology of the first embodiment or the second embodiment to combine the data set provided from outside the transmission system 30, that is, the external data set 35A, and the external data set 35A in the transmission system. A system that combines 30 internally maintained data sets, namely the internal data set 35B, and delivers advertisements based on the generated extended data set 35C.

As shown in FIG. 10 , the transmission system 30 includes a processor 31 , a memory 33 , a storage 35 and a communication interface 39 . The processor 31 executes processing according to the computer program Pr1 stored in the memory 35 . Storage 35 further contains an internal data set 35B.

As shown in Figure 11, the internal data set 35B includes characteristic data, which is associated with the advertising ID of the corresponding user for each user and describes the characteristics of the corresponding user's online behavior. As we all know, the advertisement ID is an identification code used for advertisements and is an ID inherent in the information terminal.

The characteristic data associated with the advertising ID describes the characteristics of the user's online behavior observed through the information terminal that has been assigned the corresponding advertising ID. Online behavior includes website access behavior and web content browsing behavior.

The delivery system 30 is connected to the local network through the communication interface 39 and provides advertisement delivery services through the local network. The company-side system that uses the advertisement delivery service uses the company-side system 40 to provide the delivery system 30 with advertising content and delivery designation information to be delivered. Advertising content is the information content used for advertising. The transmission designation information includes destination designation information that specifies the transmission destination and transmission quantity designation information that specifies the transmission quantity.

The enterprise-side system 40 further provides the delivery system 30 with a customer data set that describes the characteristics of the customer corresponding to the delivery target candidate as the external data set 35A.

The customer data set may be, for example, a data set describing characteristics of purchasing behavior of customers who use a store operated by the company. For example, the customer data set may include characteristic data describing the purchase quantity of each of the plurality of products by the corresponding customer as the characteristic data of each customer.

If the transmission request is input from the user-side system 40 to the processor 31 through the communication interface 39, the processor 31 executes the transmission control process shown in FIG. 12 according to the computer program Pr1.

When the delivery control process is started, the processor 31 acquires the advertising content to be delivered, the delivery designation information including the target designation information and the delivery quantity designation information, and the customer data as the external data set 35A from the user company's system 40 Set (S610).

After that, the processor 31 uses the external data set 35A as the first data set 15A and uses the internal data set 35B as the second data set 15B and performs the same processing as the processing of S110 to S190 in the analysis processing. Thereby, the processor 31 combines the external data set 35A and the internal data set 35B, and generates the extended data set 35C (S620).

Through the combination of the external data set 35A and the internal data set 35B, the characteristic data of each customer included in the external data set 35A is compared with the users included in the internal data set 35B whose customers are more likely to be the same person. Advertising ID is associated.

The extended data set 35C includes the following extended data, which is generated for each entity by combining the characteristic data of the corresponding customer's external data set 35A with the characteristic data of the corresponding user's internal data set 35B. information. Each extended data is associated with the corresponding user's advertising ID contained in the internal data set 35B.

The entity mentioned here refers to the combination of customers and users that have been corresponding to each other through data fusion. In data fusion, customers and users are corresponding one-to-one. For example, the extended data set 35C may be a data set having the following structure. That is, in the extended data set 15C shown in FIG. The entity's advertising ID.

Afterwards, the processor 31 calculates a score regarding the probability that each entity in the extended data set 35C is the transmission target (S630). For example, in the case where the external data set 35A is a data set about the customer's purchasing behavior, and the internal data set 35B is a data set about the user's online behavior, the processor 31 will expand the data set 35C about the purchases of each entity. The characteristic data of the behavior and the characteristic data about the online behavior are input into the predetermined function, and a score is calculated that digitizes the probability that the corresponding entity is the transmission target.

The transmission target is the consumer group that is the transmission target and is narrowed by parameters that characterize the consumer such as gender, age, purchasing tendency, online behavior tendency, interest and attention, and the transmission target is specified by the target specific information.

After calculating the score in S630, the processor 31 sorts the entities in the group of entities whose advertising IDs are associated with the number corresponding to the transmission quantity specified by the enterprise system 40 in descending order according to the calculated score. The sequence is determined as the content transmission target (S640). As described above, the processor 31 selects at least a portion of the plurality of users corresponding to the internal data set 35B, which has been established with any one of the plurality of customers corresponding to the external data set 35A, as the transmission target of the advertising content. correspond.

After that, the processor 31 sends the advertising content provided by the enterprise system 40 to the information terminal of the determined content transmission target via the local network (S650). The advertising content is delivered to the information terminal identified based on the advertising ID of the content delivery target. After that, the processor 31 ends the transfer control process.

According to the delivery system 30 of the fourth embodiment described above, the external data set 35A and the internal data set 35B are combined using data fusion technology that does not have common variables, thereby making it possible to combine the advertising ID and the characteristics of customers whose advertising ID is unknown. Data is associated. Therefore, the advertisement content can be delivered appropriately to the customers of the external data set 35A whose advertisement ID is unknown.

[Fifth Embodiment] The transmission system 30 of the fifth embodiment is configured such that the processor 31 executes the transmission control process shown in FIG. 13 instead of the transmission control process shown in FIG. 12 . As a description of the fifth embodiment, details of the transfer control processing executed by the processor 31 will be selectively described below. It can be understood that the structure of the transmission system 30 not mentioned in this embodiment is the same as that of the fourth embodiment.

In this embodiment, if a transmission request is input from the user-side system 40 to the processor 31 through the communication interface 39, the processor 31 executes the transmission control process shown in FIG. 13.

When the transmission control process is started, the processor 31 acquires the advertisement content to be transmitted, the transmission designation information, and the customer data set as the external data set 35A from the user company system 40 (S710).

The delivery designation information obtained in S710 does not include target designation information, but only contains delivery quantity designation information. The customer data set acquired as the external data set 35A is a specific customer data set describing the characteristics of the customer group corresponding to the transmission target that has been restricted by the utilization company.

Thereafter, similarly to the process in S620, the processor 31 combines the external data set 35A and the internal data set 35B and generates the extended data set 35C (S720). The extended data set 35C includes the following extended data. The extended data is for each entity by combining the characteristic data of the corresponding customer's external data set 35A with the characteristic data of the corresponding user's internal data set 35B. generated information.

In the process of S720 in this embodiment, the result that the customers of the external data set 35A are associated with all the users of the internal data set 35B is not generated. The extended data set 35C of this embodiment also includes the extended data of a user that is not associated with the customer of the using company as one entity. The extended data is the characteristic data of the user contained in the internal data set 35B that is not substantially expanded.

In this embodiment, among the entity groups corresponding to the extended data set 35C, the entity group that is associated with the customer group corresponding to the external data set 35A is marked as a seed, and the other entity groups are marked as seeds. Marked as non-seed.

After executing the process of S720, the processor 31 calculates the similarity of the features shown in the internal data set 35B between each non-seed entity and each seed entity based on the extended data set 35C (S730). The similarity can be calculated based on the distance in the feature space between each non-seed entity and each seed entity.

After calculating the similarity, the processor 31 determines entities whose number corresponds to the transmission quantity specified in the transmission designation information as transmission targets in order of similarity from high to low (S740). At this time, all entities corresponding to the seed are also determined as transmission targets.

As described above, the processor 31 combines a set of users, that is, a set of seeds, that has been established with a plurality of customers corresponding to the external data set 35A and a set of users with similar characteristics and seeds among the plurality of users corresponding to the internal data set 35B. Selected as the delivery target of advertising content.

Thereafter, similarly to the process in S650, the processor 31 sends the advertisement content provided by the enterprise system 40 to the information terminal of the content transmission target determined in S740 via the local network (S750). After that, the processor 31 ends the transfer control process.

According to the delivery system 30 of this embodiment described above, based on the data set of customer groups provided by the system 40 of the using company, advertising content can be delivered to a larger set of consumers showing characteristics similar to the customer groups. information terminal. Therefore, according to this embodiment, advertisements can be efficiently delivered to a large number of consumers.

[Sixth Embodiment] The delivery system 30 of the sixth embodiment is configured to provide a prediction service in addition to the advertisement delivery service like the delivery system 30 of the fourth or fifth embodiment.

In this embodiment, the processor 31 responds to an execution request from the user company's system 40 and executes the prediction process shown in FIG. 14 . As an explanation of the sixth embodiment, details of prediction processing executed by the processor 31 will be selectively described below. It can be understood that the configuration of the transmission system 30 not mentioned in this embodiment is the same as that of the fourth embodiment or the fifth embodiment.

If the processor 31 starts to execute prediction processing, the processor 31 obtains the data set of the analysis target and the analysis condition designation information from the user enterprise system 40 through the communication interface 39 (S810). The data set of the analysis object is a data set containing the characteristic data of each customer of the analysis object.

The analysis condition designation information may be information on a designated product that is used to evaluate a customer's purchase possibility. In the prediction process, the probability that each customer targeted for analysis will purchase the designated target product is predicted by calculating a predicted value of the purchase quantity of the target product. The prediction described here corresponds to the estimated customer behavior, and the predicted value corresponds to the estimated value regarding the behavior.

After executing the processing of S810, the processor 31 uses the data set of the analysis target as the first data set 15A, and uses the internal data set 35B as the second data set 15B, and by executing S110 to S170 or The processing of S310 to S370 is the same, and the correspondence matrix Ω ^* is calculated (S820). The correspondence matrix Ω ^* represents the relationship between each customer who is the analysis target and each user who has characteristic data in the internal data set 35B. corresponding relationship.

The processor 31 further extracts, for each customer of the analysis object, a predetermined number of users similar to the corresponding customer based on the calculated correspondence matrix Ω ^* , and uses the above-mentioned user data that can be specified from the internal data set 35B. The weighted average of the extracted quantity of the target product purchased by the user is used to calculate the predicted value of the corresponding quantity of the target product purchased by the customer (S830). As described above, the processor 31 infers the customer's purchasing behavior based on the purchasing behavior of the corresponding user. The internal data set 35B includes information that can specify the quantity of the target product purchased by each user.

Each element of the correspondence matrix Ω ^* represents the similarity between the customer and the user with a value from 0 to 1. The element in the i-th row and j-th column in the correspondence matrix Ω ^* represents, with a value of 0 to 1, the i-th user in the set of users corresponding to the internal data set 35B and the i-th user in the set of customers in the data set to be analyzed. The similarity between j customers.

For example, a weighted average is calculated using similarity as a weight. When it is assumed that the first user, the second user, and the third user have been extracted as three users similar to the customer, the weighted average can be calculated as follows.

That is, when the similarity between the customer and the first user is w1, the similarity between the customer and the second user is w2, the similarity between the customer and the third user is w3, the quantity of the target product purchased by the first user is p1, and the similarity between the customer and the third user is p1, the similarity between the customer and the third user is w3. When the quantity of the target product purchased by the user is p2 and the quantity of the target product purchased by the third user is p3, the predicted value pe of the customer's purchase quantity of the target product can be calculated by pe=(w1・p1＋w2・p2+w3・p3)/3 .

The similarity between each customer and all users can be specified based on the correspondence matrix Ω ^* (in other words, the size of the correspondence is established). Therefore, it is possible to calculate the predicted value of the quantity of the target product purchased by the customer based on the weighted average of the quantities of the target product purchased by all users without the process of extracting approximate users.

After executing the process of S830, the processor 31 outputs the forecast data describing the predicted value of the quantity of the corresponding product purchased by each customer to the execution requester of the forecast process (S840). After that, the processor 31 ends the prediction process shown in Fig. 14.

According to other examples, after performing the processing of S830, the processor 31 may also perform the following processing instead of outputting the prediction data or not only outputting the prediction data, that is, according to the predicted value of the quantity of the corresponding product purchased by each customer, the predicted value is In descending order, advertising content of recommended purchase target products is delivered to customers whose number corresponds to the delivery quantity designated by the using company (S840).

The delivery system 30 of the sixth embodiment has been described above. According to this embodiment, it is possible to provide meaningful advertisement delivery services by utilizing data fusion technology without shared variables, and further to provide meaningful marketing solutions.

[other] Example embodiments of the present invention are not limited to the above-described embodiments, and various forms may be adopted. The function of one of the constituent elements in the above-mentioned embodiment can also be distributed in a plurality of constituent elements. The functions of multiple constituent elements can also be synthesized into one constituent element. Part of the components of the above-described embodiment may be omitted. At least part of the configurations of the above-described embodiments may be added to or replaced with the configurations of other above-described embodiments. All aspects included in the technical ideas specified by the words described in the scope of the patent application are implementation forms of the present invention.

[Technical ideas disclosed in this specification] In this specification, it can be understood that the following technical ideas are disclosed. [Item 1] An information processing system, including: a first acquisition unit configured to acquire a first data set about a plurality of first entities, the first data set describing respective characteristics of the plurality of first entities; second The acquisition part is configured to acquire a second data set about a plurality of second entities, and the second data set describes the characteristics of each of the plurality of second entities; the dimensionality reduction part is configured to obtain a second data set from the first data performing a dimensionality reduction process on a specific group of first feature vectors and a specific group of second feature vectors from the second data set to generate a first low-dimensional feature vector corresponding to the group of first feature vectors and a group of second low-dimensional feature vectors corresponding to the group of second feature vectors, wherein each of the first feature vectors represents a corresponding one of the plurality of first entities. Features, each of the second feature vectors represents a feature of a corresponding one of the plurality of second entities, and the group of the second low-dimensional feature vectors has the same group as the first low-dimensional feature vector. The same number of dimensions; and the corresponding establishment part is configured to associate each of the plurality of first entities with the plurality of first entities according to the group of the first low-dimensional feature vectors and the group of the second low-dimensional feature vectors. At least one of the second entities establishes correspondence. [Item 2] The information processing system of Item 1, wherein the correspondence establishment unit is based on the similarity between the first entities specified from the group of the first low-dimensional feature vector and the second low-dimensional feature vector. The group-specific similarity between the second entities is associated with each of the plurality of first entities and at least one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the similarity is The relationship is suitable for the mutual relationship between the second entities. [Item 3] The information processing system of Item 1 or 2, wherein the group of the first low-dimensional feature vectors is defined according to the first feature space, and the group of the second low-dimensional feature vector is defined according to the second feature space, The correspondence establishment unit searches for a mapping (Mapping) for mapping the plurality of first entities on the first feature space to the second feature space, so that the group specific from the first low-dimensional feature vector The distribution of the plurality of first entities in the first feature space is adapted to the distribution of the plurality of second entities in the second feature space specified from the group of second low-dimensional feature vectors, and based on the mapping Each of the plurality of first entities is associated with at least one of the plurality of second entities. [Item 4] The information processing system of Item 1 or 2, wherein the correspondence establishment unit is configured to search for a matrix that maximizes the value Z(Ω) according to the following mathematical expression including the matrix K, the matrix L, and the matrix H. Ω serves as the matrix Ω ^* , and according to the matrix Ω ^* , each of the plurality of first entities is corresponding to at least one of the plurality of second entities, [Mathematical Formula 1] The quantity of the first entity is N, the quantity of the second entity is the same as the quantity of the first entity, the matrix K is a system, and the value of the element in the i-th row and j-th column represents the i-th of the plurality of first entities. The similarity between the entity and the j-th entity is based on the first low-dimensional feature vector of the i-th entity among the plurality of first entities and the j-th entity among the plurality of first entities. The first similarity matrix with N rows and N columns is calculated based on the first low-dimensional feature vector. In this matrix L, the value of the element in the i-th row and j-th column represents the i-th entity among the plurality of second entities. The similarity with the j-th entity is based on the second low-dimensional feature vector of the i-th entity among the plurality of second entities and the j-th entity among the plurality of second entities. The second similarity matrix of N rows and N columns calculated from the two low-dimensional feature vectors, the matrix H series, when the value of the element in the i-th row and j-th column is i=j, it means that the value is 1-1/N , when the value of the element in the i-th row and j-th column is i≠j, it represents a matrix of N rows and N columns with a value of 0. [Item 5] The information processing system of Item 4, wherein the correspondence establishment unit is configured to perform a re-search process on the matrix Ω ^* in a loop until a predetermined condition is met, thereby improving the matrix Ω ^* and based on The improved matrix Ω ^* establishes a correspondence between each of the plurality of first entities and at least one of the plurality of second entities. The re-search process includes: changing the dimensionality reduction process according to the matrix Ω ^* The dimensionality reduction method; causing the dimensionality reduction part to perform dimensionality reduction processing according to the modified dimensionality reduction method, and based on the newly obtained group of first low-dimensional feature vectors and the second low-dimensional feature vector The group searches the matrix Ω ^* again. [Item 6] The information processing system of Item 5, wherein the correspondence creation unit changes the dimensionality reduction method so that the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors correspond to each other The distance between the first low-dimensional feature vector and the second low-dimensional feature vector in the feature space is shortened. [Item 7] The information processing system of any one of Items 1 to 6, wherein the first data set includes a plurality of first characteristic data, and each of the plurality of first characteristic data represents the plurality of first characteristics. A characteristic of a corresponding one of the entities, the second data set includes a plurality of second characteristic data, each of the plurality of second characteristic data represents a characteristic of a corresponding one of the plurality of second entities, the The information processing system further includes a data fusion unit. The data fusion unit generates each of the plurality of first characteristic data based on the correspondence between the plurality of first entities and the plurality of second entities established by the correspondence creation unit. One is respectively combined with one of the plurality of second characteristic data, thereby generating an extended data set, the extended data set includes a plurality of extended data, each of the plurality of extended data is a corresponding first characteristic data and a corresponding Data generated by combining data with a second characteristic. [Item 8] The information processing system of any one of Items 1 to 7, wherein the first entity and the second entity are persons, and the first data set describes each of the plurality of individuals belonging to the first group. A data set of one characteristic, the second data set is a data set describing the second characteristics of each of the plurality of individuals belonging to the second group. [Item 9] The information processing system of Item 8, wherein the combination of the first feature and the second feature is a feature about purchasing behavior and a feature about movement in at least one of an online space and an offline space. , and/or a combination of features regarding visits to multiple locations in the space. [Item 10] The information processing system of any one of Items 1 to 9, wherein the first entity and the second entity are persons, and the identification information of the information terminal corresponding to each of the plurality of second entities To establish association with the second data set, the information processing system includes: a selection unit, the selection unit selects any one of the plurality of second entities that is associated with the plurality of first entities through the corresponding establishment unit. At least a part of the set of corresponding second entities has been established as the transmission target of the information content; and a transmission unit configured to, based on the identification information, transmit the information content to the transmission target of the information content. A collection of corresponding information terminals. [Item 11] The information processing system of Item 10, wherein the selection part selects a first set and a second set as transmission targets of the information content, and the first set is connected with the plurality of third sets by the correspondence establishing part. Any entity of an entity has established a corresponding set of the second entity; the second set is a set among the plurality of second entities that has similar characteristics to the first set. [Item 12] For example, any one of the information processing systems in Items 1 to 11, wherein the first entity and the second entity are persons, and the second data set describes each entity of the plurality of second entities. A data set of behavioral characteristics, the information processing system further includes an inference part, the inference part is about more than one attention entity, for each attention entity, calculates an inference value about the behavior of the corresponding attention entity, the one The above entity of concern is at least a part of the plurality of first entities, and the inferred value is calculated based on the characteristics of the behavior of at least one entity of the plurality of second entities, wherein the plurality of second entities have interacted with the plurality of second entities. Corresponding entities of interest establish correspondence. [Item 13] An information processing method, executed by a computer, including: obtaining a first data set, which is a data set about a plurality of first entities and describes each of the plurality of first entities. Characteristics; Obtain a second data set, which is a data set about a plurality of second entities and describes the characteristics of each of the plurality of second entities; by specifying a first data set specific from the first data set. The group of feature vectors and the group of second feature vectors specified from the second data set perform dimensionality reduction processing to generate a group of first low-dimensional feature vectors corresponding to the group of first feature vectors and a group of first low-dimensional feature vectors corresponding to the group of first feature vectors. The group of second feature vectors corresponds to a group of second low-dimensional feature vectors, wherein each of the first feature vectors represents a characteristic of a corresponding one of the plurality of first entities, and the second Each of the feature vectors represents a feature of a corresponding one of the plurality of second entities, and the group of the second low-dimensional feature vectors has the same number of dimensions as the group of the first low-dimensional feature vectors; And establishing a correspondence is to establish a correspondence between each of the plurality of first entities and at least one of the plurality of second entities according to the group of the first low-dimensional feature vectors and the group of the second low-dimensional feature vectors. . [Item 14] The information processing method of Item 13, wherein the establishing the correspondence is based on the similarity between the first entities specified from the group of the first low-dimensional feature vector and the similarity from the second low-dimensional feature vector. The group-specific similarity between the second entities corresponds to each of the plurality of first entities and one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the similarity is suitable. relationship between the second entity. [Item 15] A computer program product with a built-in computer program, which is characterized in that: after the computer loads and executes the computer program, it can complete the information processing method of Item 13 or Item 14. [Item 16] A computer-readable recording medium storing a computer program, which is characterized in that: after the computer loads and executes the computer program, the information processing method of Item 13 or Item 14 can be completed.

1:Information processing system 11,31: Processor 13,33:Memory 15,35:storage 15A: First data set 15B: Second data set 15C:Extended data set 17: User interface 19,39: Communication interface 30:Transmission system 35A: External data set 35B: Internal data set 35C:Extended data set 40: Utilize enterprise-side systems E11,E12,E13,E14,E15,E16,E17,E21,E22,E23,E24,E25,E26,E27: points Pr, Pr1: computer program P1, P2, P3: Commodity P1,S2,S3: website S110,S120,S130,S140,S150,S160,S170,S180,S190,S310,S320,S330,S340,S350,S360,S370,S380,S390,S410,S420,S430,S440,S450,S460,S5 10, S520,S530,S540,S550,S610,S620,S630,S640,S650,S710,S720,S730,S740,S750,S810,S820,S830,S840: Steps

[Picture 1] A block diagram showing the structure of an information processing system. [Figure 2] A flowchart showing the analysis processing performed by the processor. [Picture 3A] A diagram illustrating the composition of the first data set. [Picture 3B] A diagram illustrating the composition of the second data set. [Figure 4A] A diagram illustrating the search method for matrix Ω. [Picture 4B] A diagram illustrating the search method for matrix Ω. [Figure 5] This is an example of a diagram illustrating the composition of a correspondence table generated by a processor. [Figure 6] A diagram illustrating the composition of an extended data set generated by a processor. [Fig. 7] A flowchart showing analysis processing executed by the processor in the second embodiment. [Fig. 8] A flowchart showing evaluation processing executed by the processor in the third embodiment. [Fig. 9] A flowchart showing selection processing executed by the processor in the third embodiment. [Fig. 10] A block diagram showing the structure of the transmission system of the fourth embodiment. [Fig. 11] A diagram illustrating the structure of the internal data set in the fourth embodiment. [Fig. 12] A flowchart showing transfer control processing executed by the processor in the fourth embodiment. [Fig. 13] A flowchart showing transfer control processing executed by the processor in the fifth embodiment. [Fig. 14] A flowchart showing prediction processing executed by the processor in the sixth embodiment.

Claims (16)

An information processing system consisting of: The first acquisition unit is configured to acquire a first data set regarding a plurality of first entities, and the first data set describes the characteristics of each of the plurality of first entities; The second acquisition unit is configured to acquire a second data set regarding a plurality of second entities, and the second data set describes the characteristics of each of the plurality of second entities; The dimensionality reduction unit is configured to perform dimensionality reduction processing on a group of first eigenvectors specified from the first data set and a group of second eigenvectors specified from the second data set. A group of first low-dimensional feature vectors corresponding to the group of first feature vectors and a group of second low-dimensional feature vectors corresponding to the group of second feature vectors, wherein each of the first feature vectors represents a feature of a corresponding one of the plurality of first entities, each of the second feature vectors represents a feature of a corresponding one of the plurality of second entities, and the second low-dimensional feature vector The group has the same number of dimensions as the group of the first low-dimensional feature vector; and The correspondence establishment unit is configured to associate each of the plurality of first entities with at least one of the plurality of second entities based on the group of the first low-dimensional feature vectors and the group of the second low-dimensional feature vectors. Establish correspondence. The information processing system of claim 1, wherein the correspondence establishment unit is based on the similarity between the first entities specified from the group of the first low-dimensional feature vector and the group of the second low-dimensional feature vector. Specific similarity between the second entities, establishing correspondence between each of the plurality of first entities and at least one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the degree of similarity is suitable for The mutual relationship between the second entities. Such as requesting the information processing system of item 1, wherein, Define the group of the first low-dimensional feature vectors according to the first feature space, The group of second low-dimensional feature vectors is defined according to the second feature space, The correspondence establishment unit searches for a mapping for mapping the plurality of first entities on the first feature space to the second feature space, so that the first specific entity from the group of the first low-dimensional feature vector The distribution of the plurality of first entities in feature space is adapted to the distribution of the plurality of second entities in the second feature space specified from the group of second low-dimensional feature vectors, and Each of the plurality of first entities is corresponding to at least one of the plurality of second entities based on the mapping. The information processing system of Claim 1, wherein the correspondence establishment unit is configured to search for the matrix Ω that maximizes the value Z(Ω) according to the following mathematical expression including the matrix K, the matrix L, and the matrix H as the matrix Ω ^* , and establish a correspondence between each of the plurality of first entities and at least one of the plurality of second entities according to the matrix Ω ^* , [Mathematical Formula 1] The quantity of the first entity is N, the quantity of the second entity is the same as the quantity of the first entity, the matrix K is a system, and the value of the element in the i-th row and j-th column represents the i-th of the plurality of first entities. The similarity between the entity and the j-th entity is based on the first low-dimensional feature vector of the i-th entity among the plurality of first entities and the j-th entity among the plurality of first entities. The first similarity matrix with N rows and N columns is calculated based on the first low-dimensional feature vector. In this matrix L, the value of the element in the i-th row and j-th column represents the i-th entity among the plurality of second entities. The similarity with the j-th entity is based on the second low-dimensional feature vector of the i-th entity among the plurality of second entities and the j-th entity among the plurality of second entities. The second similarity matrix of N rows and N columns calculated from the two low-dimensional feature vectors, the matrix H series, when the value of the element in the i-th row and j-th column is i=j, it means that the value is 1-1/N , when the value of the element in the i-th row and j-th column is i≠j, it represents a matrix of N rows and N columns with a value of 0. The information processing system of claim 4, wherein the correspondence establishment unit is configured to perform a re-search process on the matrix Ω ^* in a loop until a predetermined condition is met, thereby improving the matrix Ω ^* and based on the improved The matrix Ω ^* establishes a correspondence between each of the plurality of first entities and at least one of the plurality of second entities. The re-search process includes: changing the dimensionality reduction in the dimensionality reduction process according to the matrix Ω ^* method; causing the dimensionality reduction part to perform dimensionality reduction processing according to the updated dimensionality reduction method, and based on the newly obtained group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors again Search for the matrix Ω ^* . The information processing system of claim 5, wherein the correspondence establishment unit changes the dimensionality reduction method so that the first group of the first low-dimensional feature vectors and the group of the second low-dimensional feature vectors correspond to each other. The distance between the low-dimensional feature vector and the second low-dimensional feature vector in the feature space is shortened. Such as requesting the information processing system of item 1, wherein, The first data set includes a plurality of first characteristic data, each of the plurality of first characteristic data represents a characteristic of a corresponding one of the plurality of first entities, The second data set includes a plurality of second characteristic data, each of the plurality of second characteristic data represents a characteristic of a corresponding one of the plurality of second entities, The information processing system further includes a data fusion unit. The data fusion unit generates the plurality of first characteristic data based on the correspondence between the plurality of first entities and the plurality of second entities established by the correspondence creation unit. Each of the plurality of second characteristic data is combined with one of the plurality of second characteristic data, thereby generating an extended data set. The extended data set includes a plurality of extended data. Each of the plurality of extended data is a corresponding first characteristic data and the corresponding Data generated by combining a second characteristic data. If the information processing system of claim 1 is requested, the first entity and the second entity are persons, The first data set is a data set describing the first characteristics of each of the plurality of individuals belonging to the first group, The second data set is a data set describing the second characteristics of each of the plurality of individuals belonging to the second group. The information processing system of claim 8, wherein the combination of the first feature and the second feature is a feature about purchasing behavior, a feature about movement in at least one of online space and offline space, and/ Or a combination of features regarding visits to multiple locations in the space. For example, the information processing system of any one of claims 1 to 9, wherein the first entity and the second entity are persons, The identification information of the information terminal corresponding to each of the plurality of second entities is associated with the second data set, The information processing system includes: A selection unit that selects, as information, at least a part of a set of second entities among the plurality of second entities that has been associated with any one of the plurality of first entities through the correspondence establishing unit. the subject of the transmission of the content; and The transmission unit is configured to transmit the information content to a set of information terminals corresponding to the transmission target of the information content based on the identification information. The information processing system of claim 10, wherein the selection unit selects a first set and a second set as transmission targets of the information content, and the first set is connected to the plurality of first entities through the correspondence establishment unit. Any entity has established a corresponding set of the second entity; the second set is a set among the plurality of second entities that has similar characteristics to the first set. For example, the information processing system of any one of claims 1 to 9, wherein the first entity and the second entity are persons, The second data set is a data set describing characteristics of the behavior of each of the plurality of second entities, The information processing system further includes an inference part, which is about one or more entities of interest, and for each entity of interest, calculates an inference value about the behavior of the corresponding entity of interest, The one or more entities of interest are at least part of the plurality of first entities, The inferred value is calculated based on the characteristics of the behavior of at least one entity of the plurality of second entities, wherein the plurality of second entities has established a correspondence with the corresponding entity of interest. An information processing method executed by a computer, including: Obtaining a first data set, the first data set is a data set related to a plurality of first entities and describes the respective characteristics of the plurality of first entities; Obtaining a second data set, the second data set is a data set related to a plurality of second entities and describes the respective characteristics of the plurality of second entities; Generating a group of first feature vectors specified from the first data set and a group of second feature vectors specified from the second data set by performing a dimensionality reduction process on the group of first feature vectors specified from the first data set a group of corresponding first low-dimensional feature vectors and a group of second low-dimensional feature vectors corresponding to the group of second feature vectors, wherein each of the first feature vectors represents the plurality of first A characteristic of a corresponding one of the entities, each of the second feature vectors represents a characteristic of a corresponding one of the plurality of second entities, and the group of the second low-dimensional feature vectors has the same characteristics as the third A group of low-dimensional feature vectors with the same number of dimensions; and Establishing a correspondence involves establishing a correspondence between each of the plurality of first entities and at least one of the plurality of second entities based on the group of first low-dimensional feature vectors and the group of second low-dimensional feature vectors. The information processing method of claim 13, wherein the establishing the correspondence is based on the similarity between the first entities specified from the group of the first low-dimensional feature vector and the group-specified from the second low-dimensional feature vector. The similarity between the second entities, each of the plurality of first entities is corresponding to one of the plurality of second entities, so that the mutual relationship between the first entities with respect to the similarity is suitable for the third entity The relationship between two entities. A computer program product with a built-in computer program, which is characterized by: After the computer program is loaded and executed, the information processing method of claim 13 or claim 14 can be completed. A computer-readable recording medium storing computer programs, which is characterized by: After the computer program is loaded and executed, the information processing method of claim 13 or claim 14 can be completed.

Images