JP5220582B2 - Apparatus, method and program for supporting evaluation of new customer candidates - Google Patents

Description

The present invention relates to a technology for supporting evaluation work for new customer candidates, and more particularly to a technology for evaluating each target of a new customer candidate based on known customer information.

  For companies, finding new customers is the most important management issue. In order to develop new customers, various sales activities such as direct mail, media advertising and promotion, and events such as seminars and exhibitions are held. And in order to discover a customer more efficiently, collecting and analyzing the reaction of the other party with respect to such a sales activity and narrowing down a prospective customer are performed. In recent years, since many companies have their own homepages, a list of prospective customers is also created by analyzing their access.

  Further, there is a technique for improving the access analysis and extracting target information related to sales activities by analyzing SFA (Sales Force Automation) data and Web access log data in a cross-sectional and comprehensive manner.

For example, Patent Document 1 discloses a data integration processing unit that integrates a total result of SFA data and a total result of Web access log data including common customer names, product categories, and contact degree information of different customers, and this data integration processing. Based on the gap between the customer contact level information included in the SFA data relating to a specific product category or a specific customer name from the integrated data integrated by the means and the contact level information included in the Web access log data, the target of the specific product category Disclosed is a customer information analysis system provided with organization gap detection means for extracting a target product that is a target customer name or a target customer name.
JP 2004-348682

  However, conventional narrowing down of prospective customers was based on some kind of reaction from the other party, such as inquiries and consultations on sales activities, actual business negotiations and sales history, and access to the company's homepage. Therefore, a partner who cannot obtain such information has not been judged as a prospective customer even if there is a sufficient possibility of becoming a customer.

Further, in the conventional new customer development, it has not been performed to extract and evaluate customer candidates such as individuals, companies, and organizations that can become new customers by using customer information of current, past, or both.

This invention has been made to solve the above-mentioned problems, and whether it can be a new customer even if it is an individual, a company, an organization, etc. that has not been contacted in the past and has not received any response. The purpose is to provide a technology that can evaluate this. Moreover, it aims at providing the technique which can evaluate the appropriateness as a new customer based on the customer information of the present, the past, or both.

  The present invention that achieves the above-described object is realized by an apparatus that supports the following new customer candidate evaluation work. Such a support device is a plurality of preselected multiples determined for each of a plurality of first objects having either a customer or non-customer label and a plurality of second objects listed in the customer candidate list. A feature vector storage unit that stores a feature vector whose element is attribute information about the attribute of the attribute, and a discrimination that calculates a discriminant plane in the support vector machine using a plurality of feature vectors of the first target read from the feature vector storage unit as training data For each of the surface calculation means, the distance calculation means for calculating the distance between the second object and the first object as the length obtained by projecting the distance between the feature vectors onto the normal vector of the discrimination surface, and the second object The first object that is located in the vicinity of the second object and has a customer label is extracted according to the calculated distances, and is used as evaluation information for evaluating the second object. And a determination means for determining.

  Here, the non-customer information means information related to objects such as individuals, companies, organizations, etc. that are not customers at the present or present and in the past. In other words, non-customer information may be the target of individuals, companies, organizations, etc. who are not customers at present or at present and in the past, and it has been found that it is not always possible to become a clear customer in the future by conducting sales activities. It doesn't have to be a subject. As an example, the non-customer information may be information obtained by removing customer information of a service user from customer information for a service provider that provides a service user with a support service for evaluating a new customer candidate. This is based on the assumption that there is a high possibility that the target has already been introduced to the service user if there is a target such as an individual, a company, or an organization that can be a customer.

  Preferably, the discrimination surface is a discrimination surface obtained by applying a biased support vector machine.

  Preferably, the discriminant plane is a discriminant plane obtained by applying a soft margin method in a biased support vector machine.

  Preferably, the attribute information regarding a plurality of attributes that are elements of the feature vector of each target includes attribute information regarding third-party evaluation of the target, attribute information regarding the scale and sales of the target, and attribute information regarding the target industry and region. Includes any attribute information.

  Preferably, the feature vector for each object is a normalized feature vector. The normalization of the feature vector may be performed so that the average of each element of the feature vector becomes zero. Further, the normalization of the feature vector may be further corrected so that the variance of each element of the feature vector becomes the same value.

  Preferably, the vicinity of the second object is a distance that reaches the top Kth when the first object is arranged in order of increasing distance from the second object. Here, K is a positive integer.

  Preferably, the support device further includes an evaluation unit that evaluates each of the second objects based on the number of the first objects having a customer label determined as the evaluation information of the second object.

  Preferably, the support device includes an evaluation unit that evaluates each of the second objects based on a sum of attribute values of specific attributes of the first object having a customer label determined as information for evaluation of the second object. In addition.

  Preferably, the specific attribute is sales information of a first target.

  Preferably, the support device sets each of the second objects from the top in order of increasing distance from the second object among the first objects having the customer label determined as the evaluation information of the second object. An evaluation unit that evaluates based on one or more first objects included up to the Mth is further included. Here, M is a positive integer.

  The present invention has been described above as an apparatus that supports the evaluation work of new customer candidates. However, the present invention can also be understood as a method or program for supporting evaluation work for new customer candidates, which is executed in such a support device.

  According to the present invention, it is possible to evaluate whether an individual, a company, an organization, or the like who has not been contacted in the past and has not received any reaction can become a new customer. In addition, it is possible to evaluate the appropriateness of a new candidate as a new customer based on current or past customer information or both. Other effects of the present invention will be understood from the description of each embodiment.

  BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, the best mode for carrying out the present invention will be described in detail with reference to the drawings. However, the following embodiments do not limit the invention according to the claims, and are described in the embodiments. Not all combinations of features that are present are essential to the solution of the invention. Note that the same numbers are assigned to the same elements throughout the description of the embodiment.

  FIG. 1 is a diagram showing an example of a hardware configuration of an information processing apparatus suitable for realizing a support apparatus 200 that supports an evaluation work for a new customer candidate according to an embodiment of the present invention. The information processing apparatus includes a CPU (central processing unit) 1 and a main memory 4 connected to the bus 2. The hard disk devices 13 and 30 and the CD-ROM devices 26 and 29, the flexible disk device 20, the MO device 28 and the DVD device 31 are removable storages (external storage systems in which recording media can be exchanged). Are connected to the bus 2 via the IDE controller 25, the SCSI controller 27, and the like.

  A storage medium such as a flexible disk, MO, CD-ROM, or DVD-ROM is inserted into the removable storage. In these storage media, the hard disk devices 13 and 30, and the ROM 14, instructions of a computer program for carrying out the present invention can be recorded by giving instructions to the CPU or the like in cooperation with the operating system.

  That is, the above-described numerous storage devices of the information processing device as the support device 200 include a program for supporting the evaluation work of a new customer candidate, customer information, non-customer information, and customer used by the program. Customer candidate information including a list of candidates and feature vectors can be stored. The computer program is executed by being loaded into the main memory 4. The computer program can be compressed or divided into a plurality of pieces and recorded on a plurality of media.

  The information processing apparatus receives input from an input device such as a keyboard 6 or a mouse 7 via the keyboard / mouse controller 5. The information processing apparatus receives input from the microphone 24 via the audio controller 21 and outputs sound from the speaker 23. The information processing apparatus is connected via a graphics controller 10 to a display device 11 for presenting visual data to the user. The information processing apparatus can connect to a network via a network adapter 18 (Ethernet (R) card or token ring card) or the like, and can communicate with other computers.

  As described above, the information processing apparatus suitable for realizing the support apparatus 200 according to the embodiment of the present invention is realized by an information processing apparatus such as a normal personal computer, workstation, mainframe, or a combination thereof. It will be readily understood that In addition, the component demonstrated above is an illustration, All the components are not necessarily an essential component of this invention.

  FIG. 2 is a diagram showing an example of a functional configuration of the support apparatus 200 that supports the evaluation work of new customer candidates according to an embodiment of the present invention. The support apparatus 200 according to the embodiment of the present invention includes a non-customer information storage unit 205, a customer information storage unit 210, a customer candidate information storage unit 215, a feature vector creation unit 220, a feature vector storage unit 225, and a determination. A surface calculation unit 230, a determination surface information storage unit 235, a distance calculation unit 240, a determination unit 245, an evaluation unit 250, and a score storage unit 255 are provided. In the following description, it is assumed that the service providing company A provides the service user B with a support service for evaluating new customer candidates.

  The non-customer information storage unit 205 stores a non-customer list for the service user B and information on each target in the list. As described above, the non-customer in the present invention means information related to an object such as an individual, a company, or an organization that is not a customer for the service user B at present or at present and in the past. In the present embodiment, customer information of a customer excluding the customer for the service user B from the customer for the service provider A is used as the non-customer information. This is based on the assumption that there is a high possibility that the target has already been introduced to the service user if there is a target such as an individual, a company, or an organization that can be a customer.

  The customer information storage unit 210 stores a list of customers for the service user B and information on each target in the list. In addition, the customer candidate information storage unit 215 stores a list of customer candidates for the service user B and information on each target in the list. FIG. 3 shows the relationship between the non-customer, customer, and customer candidate information used in this embodiment. As shown in FIG. 3, the customer candidate information C315 in this embodiment is information about individuals, companies, organizations, and the like who are not customers to the service provider A and the service user B.

Information of each target included in each of the storage units 205 to 215 may relate to a plurality of attributes listed below.
(A) Non-customer ratings by third parties (b) Non-customer capital, number of employees, number of factories, number of establishments, sales, profits, dividends, reported income (c) industry, region The information of each target is not limited thereto, and may further include other information that can indicate the characteristics of each target of customers, non-customers, and customer candidates. Further, each target included in the non-customer information storage unit 205 has a label of a non-customer, and each target included in the customer information storage unit 210 has a label of a customer.

  The feature vector creation unit 220 creates a feature vector whose elements are attribute information related to a plurality of attributes selected in advance for each target of a non-customer, a customer, and a customer candidate. Here, attribute information related to a plurality of attributes that are elements of the feature vector of each target includes attribute information related to third party evaluation of the target, attribute information related to the scale and sales of the target, and attribute information related to the target industry and region. Any plurality of attribute information may be included. The feature vector creation unit 220 reads out these pieces of information from the storage units 205 to 215, and creates feature vectors for the targets of the non-customer, the customer, and the customer candidate. The feature vector creation unit 220 stores the created feature vector in the feature vector storage unit 225.

  Here, a specific method of creating a feature vector by the feature vector creating unit 220 will be described. In this embodiment, the feature vector x is basically defined as x = (score, capital, number of employees,...). However, the attribute information such as capital and the number of employees classified in (b) above is logarithmic. That is, the feature vector is defined as x = (score, log (capital), log (number of employees),...). In addition, the attribute information of the business category and region classified as (c) is expressed by 0 or 1 vector. That is, for example, when defining the feature vector x = (..., Agriculture, forestry, fishery,...), The feature vector x of a company specializing in fishery is expressed as (..., 0, 0, 1...). The regional attribute information can also be expressed in vectors in the same way as the type of business in each prefecture.

  Preferably, the feature vector creation unit 220 normalizes the feature vector x of each target. As an example, the feature vector creation unit 220 corrects so that the average of each element of the feature vector x becomes zero. That is, if feature vector x = (score, capital, number of employees, ...), normalized feature vector x '= (score-average score, capital-average capital, number of employees-employee Average number of members, ...). Furthermore, the feature vector creation unit 220 may correct the variance of each element of the feature vector x so as to have the same value. That is, if feature vector x = (score, capital, number of employees, ...), normalized feature vector x “= ((score-average score) / standard deviation of evaluation, (capital-capital) Average value) / standard deviation of capital, (number of employees-average value of number of employees) / standard deviation of number of employees, ...).

  The discriminant plane calculation unit 230 uses, as training data, feature vectors of a plurality of targets (hereinafter referred to as first targets) having either customer or non-customer labels read from the feature vector storage unit 225 as support data machines (Support). The discriminant plane in Vector Machines (hereinafter referred to as SVM) is calculated. Although details will be described later, in the present invention, the distance between feature vectors is defined using a discrimination surface in SVM. The discriminant plane calculation unit 230 stores the calculated discriminant plane information (a parameter w described later) in the discriminant plane information storage unit 235.

  Here, SVM will be described. SVM is a classification method for determining from which training data belonging to one of two classes to which class unknown data belongs. Pattern recognition fields such as speech recognition, character recognition, figure recognition, and medical diagnosis fields It is applied to various fields such as. The SVM uniquely determines a discriminant plane for class classification based on a clear criterion of “margin maximization”.

For example, assume that a set of training data is given as follows. (x ₁ , y ₁ ), (x ₂ , y ₂ ), ..., (x _n , y _n ). Here, x _i is an input vector, and y _i is a class label that satisfies y _i ε {1, -1}. For this training data input, the SVM has the following linear discriminant function: f (x) = sign (w ^T x + b). Here, w and b are parameters for determining the discrimination surface g (x). The function sign (t) is a sign function that takes 1 when t> 0 and -1 when t ≦ 0. Furthermore, w ^T indicates the transpose of the vector w.

Assuming that the training data is linearly separable, the problem of finding the parameters w and b that maximize the margin is, after all, the constraints y _i (w ^T x _i + b) ≧ 1, (i = 1, 2,. .., n) is equivalent to the problem of finding a parameter that maximizes 1 / ‖w‖ (optimization problem P1). In practice, training data is rarely linearly separable, and in this case, a soft margin method can be used that relaxes the constraints to allow some discrimination errors.

The soft margin method allows some training data to enter the other side while maximizing the margin 1 / ‖w‖. The problem of finding the optimal discriminant surface in the soft margin method is the constraint condition y _i (w ^T x _i + b) ≧ 1-ξ _i , (i = 1, 2, ..., n, ξ _i ≧ 0, i = 1, 2, ..., n), equivalent to the problem of finding a parameter that maximizes 1 / ‖w‖-C ₁ Σξ _i , (C ₁ ≧ 0) (optimization problem P2) . Here, Σ is the sum from i = 1 to i = n, and ξ _i is a parameter indicating the amount of protrusion. C ₁ is a weight parameter for adjusting the balance between the margin of the first term and the amount of protrusion of the second term.

  The optimization problems P1 and P2 are known as quadratic programming problems in the field of mathematical programming, and various numerical calculation methods have been proposed, and their solutions are well-known, and thus description thereof is omitted here. The optimization problems P1 and P2 can be solved by using a solver such as an open source SVM-QP.

 If each non-customer with a non-customer label is a target that is clearly not a customer in the future as a result of sales activities, etc. or as a result of some survey, SVM or software described above The discriminant surface in the margin method can be used. However, as in this embodiment, when using customer information of a customer excluding a customer for a service user from a customer for a service provider as non-customer information, that is, guessing or assuming an object to be a non-customer When doing so, the discriminant plane calculation unit 230 calculates the discriminant plane using the biased SVM. Alternatively, the discriminant plane calculation unit 230 may calculate the discriminant plane using a soft margin method in biased SVM.

  Biased SVMs are disclosed in B. Liu, Y. Dai, X. Li, W. S. Lee, P. S. Yu, "Building Text Classifiers Using Positive and Unlabeled Examples", ICDM-03, pp19-22, 2003. Therefore, biased SVM will be described.

First, it is assumed that a set of training data is given as follows. (x ₁ , y ₁ ), (x ₂ , y ₂ ), ..., (x _n , y _n ). Here, x _i is an input vector, and y _i is a class label that satisfies y _i ε {1, -1}. However, (k-1) training data from i = 1 to i = k-1 has a class label of +1, while the remaining i = k to i = n (n-k + 1 ) Pieces of training data are assumed to be unlabeled training data given a label of -1. Here, it is assumed that the size of the training data is sufficiently large. Then, an appropriate classifier can be obtained by correctly classifying training data having a class label of +1 and minimizing the number of training data without a label classified as +1.

When training data having a label of +1 other than unlabeled training data can be correctly classified, the problem of obtaining the optimum discriminant plane in the biased SVM is that the constraint condition (w ^T x _i + b) ≧ 1, (i = 1 , 2, ..., k-1) and (w ^T x _i + b) ≤-1 + d _i , (i = k, k + 1, ..., n), (d _i ≥0, i = k, k + 1, ..., n), 1 / ‖w‖-C ₂ Σd _i , (C ₂ ≧ 0, Σ is the sum from i = k to i = n) Is equivalent to the problem of obtaining parameters to be optimized (optimization problem P3). A method for determining an optimum discriminant plane in the biased SVM will be described with reference to FIG. 4 using the feature vector of the first target in the present embodiment as training data.

  In FIG. 4, the triangle symbol is training data (feature vector) having a non-customer label (corresponding to -1 label), and the circle symbol is training data (feature vector) having a customer label (corresponding to +1 label). ). Here, the discriminant plane for classifying the training data into two is the amount of protrusion of the training data having the label of the non-customer that has entered the opposite side beyond the discriminant plane while maximizing the margin (the part indicated by the arrow in the figure). ) Is minimized.

In addition, when discriminating errors are allowed even for training data having a label of +1, the soft margin method in the biased SVM is used. In this case, the problem of finding the optimal discriminant plane is the constraint condition y _i (w ^T x _i + b) ≧ 1-d _i , (i = 1, 2, ..., n, d _i ≧ 0, i = 1, 2, ..., under n), 1 / ‖w‖-C + Σd i -C - Σd i, (C +, C - ≧ 0, 2 is Σ items from i = 1 i = This is equivalent to the problem of obtaining a parameter that maximizes the sum up to k−1, and Σ of the three items indicates the sum from i = k to i = n (optimization problem P4). C ₊ is a weight parameter for the amount of protrusion for the training data having a label of +1, and C _- is a weight parameter for the amount of protrusion for the training data having a label of -1.

The above formula optimization problem P3, for example, rewritten using the Manhattan distance L ₁ is as follows.
minimize: ‖w‖ ₁ + C ₂ Σd _i , (C ₂ ≧ 0, Σ is the sum from i = k to i = n)
subject to: w ^T x _i + b ≧ 1, (i = 1, 2, ..., k-1)
w ^T x _i + b ≦ -1 + d _i , (i = k, ..., n), d _i ≧ 0, (i = k, ..., n)

Furthermore, since the above equation can be converted into a linear programming problem by rewriting as follows, an optimum solution can be obtained.
minimize: Σ (w ⁺ _J + w ^- _j ) + C ₂ Σd _i , (C ₂ ≧ 0, Σ for one item is the sum from j = 1 to j = n, Σ for the two items is i = k to i = indicates the total up to n)
subject to: (w ⁺ _J + w ^- _j ) x _ij + b ≧ 1, (i = 1, 2, ..., k-1)
(w ⁺ _J + w ^- _j ) x _ij + b ≦ -1 + d _i , (i = k, ..., n), d _i ≧ 0, (i = k, 2, ..., n)
The linear programming problem can be solved by using a commercial solver such as CPLEX of ILOG and an open source solver. The optimization problem P4 can be solved similarly.

The distance calculation unit 240 calculates the distance between the first target and the second target, which is the target of the customer candidate, as a length obtained by projecting the distance between the feature vectors onto the normal vector of the discrimination surface. That is, the distance calculation unit 240 reads the feature vectors x _i and x _j of the first object and the second object from the feature vector storage unit 225, and uses the parameter w read from the discrimination plane storage unit 235 to The distance between the object and the second object is determined as | w (x _i -x _j ) | (see FIG. 5). Thus, in the present invention, a new metric space is defined using the discrimination surface.

  The determination unit 245 extracts, for each of the second objects, the first object that is located in the vicinity of the second object and has the customer label according to each distance obtained by the distance calculation unit 240, and the second object is extracted. It is determined as evaluation information for evaluation. Then, the determination unit 245 stores the extracted identification information of each target in the customer candidate information storage unit 215 in association with the corresponding second target in the customer candidate list.

  Preferably, the vicinity of the second object is a distance that reaches the top Kth when the first objects are arranged in order of increasing distance from the second object. Here, K is a positive integer, for example, 100. FIG. 6 shows an example when K = 5 for simplicity. In FIG. 6, a square symbol indicates an object of a customer candidate, a triangle symbol indicates an object having a non-customer label, and a circle symbol indicates an object having a customer label. Since K = 5, the first objects having customer labels extracted by the determining unit 245 are three inside the circle indicated by the broken line. Note that each object in FIG. 6 is shown in the new metric space described above.

The present invention uses the k-nearest neighbor method in evaluating each target customer candidate based on each target being a customer and each target being a non-customer. The k-nearest neighbor method is a statistical classification method based on the nearest training data in the feature space. In the conventional k-neighbor method, the Manhattan distance L ₁ , the Euclidean distance L ₂ , and the Mahalanobis distance are used as the distance between samples. For this reason, in the conventional k-neighbor method, when considering the neighborhood, all the samples are treated equally and the labels of the samples are not considered.

  However, as in the present invention, when the objective is to evaluate each candidate customer based on each customer target and each non-customer target, the distance between the two samples is the label of the sample. It is desirable to decide in consideration. That is, when customers E and F and non-customer G are given, use the k neighborhood in a distance space where the distance between customer E and customer F is short and the distance between customer E and non-customer G is long. Is desirable. Therefore, in the present invention, a new metric space is defined using the discriminant plane as described above, and the k-nearest neighbor method is used in the metric space.

  The evaluation unit 250 evaluates each of the second objects based on the number of the first objects having the customer label determined as the evaluation information for the second object. Instead of this or in addition to this, the evaluation unit 250 adds the attribute values of the specific attributes of the first object having the customer label determined as the second object evaluation information for each of the second objects. You may evaluate by. As the specific attribute, for example, sales information on the first target may be used.

  Further, the evaluation unit 250 determines each of the second objects from the top to the Mth in the order of the shortest distance from the second object among the first objects having the customer label determined as the evaluation information of the second object. You may evaluate based on the 1 or more 1st object contained in. Here, M is a positive integer. When M is 1, the evaluation unit 250 sets each of the second objects as the first object closest to the second object among the first objects having the customer label determined as the evaluation information for the second object. That is, the evaluation is performed on the first object having the label of the nearest customer.

  In addition, for example, the evaluation unit 250 may evaluate the second target in which the excellent customer of the service user is included in the one or more first targets thus determined. Furthermore, the evaluation unit 250 may treat one or more pieces of information of the first target thus obtained as evaluation results. As an example of how to use such information itself, a sales representative of a customer candidate who is the second target is a sales representative of any one of the above-described one or more first targets determined for the second target. And so on.

  In addition, the evaluation unit 250 scores the second target based on the evaluation result, and stores the second target list arranged in the score order in the score storage unit 255. Instead, the evaluation unit 250 may store the evaluation result in the score storage unit 255 as it is. FIG. 7 shows an example of a score table stored in the score storage unit 255. In the example shown in FIG. 7, each company that is a customer candidate is ranked according to the number of customer companies that exist in the vicinity. However, when the number of customer companies in the vicinity is the same, each company that is a customer candidate is ranked according to the average sales of the nearby customer companies. The effectiveness of the present invention when ranking customer candidates by such a method was verified by conducting a comparative experiment by cross-validation.

  FIG. 8 shows the results of a comparative experiment by cross validation. Here, data having a customer or non-customer label divided into two is called data 1 and the rest is called data 2. Then, Fig.8 (a) shows the experimental result at the time of making data 1 into training data which has a label, and data 2 as unknown data without a label. FIG. 8B is the opposite, and shows experimental results when data 1 is unknown data with no label and data 2 is training data with a label. In both cases, the lower data shown as kNN shows the result of the k-nearest neighbor method using the L1 distance, and the upper data shown as proposedAlgorithm shows the result of using the present invention.

  In FIG. 8, the horizontal axis indicates the number of companies in the top ranking, and the vertical axis indicates the actual number of customer companies. For example, when the value on the horizontal axis is 200, the value on the vertical axis represents the number of customer companies actually included in the top 200 companies when that method is selected. When the horizontal axis is around 400, according to the present invention, there are more than 70 correct answers, but the conventional k-nearest method has only about 60 correct answers, and the present invention gives about 15 to 20% more correct answers. I understand that

  Next, with reference to FIG. 9, an example of the flow of processing for supporting the evaluation work for new customer candidates according to an embodiment of the present invention will be described. In FIG. 9, the processing executed in the computer starts at step 1000, and the computer creates the above-described feature vector for each object listed in the respective lists of customers, non-customers and customer candidates. Then, the computer uses the calculation process to convert feature vectors of a plurality of objects having customer or non-customer labels as training data in any one of SVM, soft margin method, biased SVM, and biased SVM soft margin method. A discrimination plane is calculated (step 1005).

  Next, the computer calculates the distance between each object listed in the customer candidate list and each of a plurality of objects having either a customer or non-customer label, and the distance between each feature vector. The length projected onto the normal vector of the discriminant plane is calculated (step 1010).

  Then, the computer extracts, for each target in the customer candidate list, according to each calculated distance that is located in the vicinity of the target and has a customer label, and identifies the extracted identification information for each target in the correspondence to the customer candidate list Recording in association with the target to be performed (step 1015)

  Finally, the computer evaluates each target in the customer candidate list based on a target having one or more customer labels associated with the target (step 1020). Then, the process ends.

  As described above, according to the present invention, it is possible to obtain the degree to which a new customer can be obtained even for an individual, a company, an organization, or the like who has not been contacted in the past and has not received any reaction. In addition, it is possible to evaluate the appropriateness of a new candidate as a new customer based on current or past customer information or both.

  As mentioned above, although this invention was demonstrated using embodiment, the technical scope of this invention is not limited to the range as described in the said embodiment. It will be apparent to those skilled in the art that various modifications or improvements can be added to the above-described embodiments. Therefore, it is a matter of course that embodiments with such changes or improvements are also included in the technical scope of the present invention.

It is the figure which showed an example of the hardware constitutions of the information processing apparatus suitable for implement | achieving the assistance apparatus 200 which supports the evaluation operation | work of the new customer candidate which concerns on one embodiment of this invention. It is a figure which shows an example of a function structure of the assistance apparatus 200 which concerns on one embodiment of this invention. It is a figure which shows an example of the mutual relationship of each information of a non-customer, a customer, and a customer candidate. It is a figure which shows an example of the discrimination | determination surface in biased SVM. It is a figure which shows an example of the distance between the feature vectors in the metric space defined by this invention. It is a figure which shows an example of k neighborhood method utilization in the metric space defined by this invention. It is a figure which shows an example of the score table | surface obtained by applying this invention. (A) shows an example of the result of the comparative experiment by cross validation. (B) shows another example of the result of the comparative experiment by cross validation. It is a figure which shows the flowchart which shows an example of the flow of a process which supports evaluation work of the new customer candidate which concerns on one embodiment of this invention.

Claims (12)

A device that supports evaluation work for new customer candidates,
Attribute information about a plurality of preselected attributes obtained for each of a plurality of first objects having a label of either customer or non-customer and a plurality of second objects listed in the customer candidate list A feature vector storage unit for storing feature vectors
Discriminant plane calculating means for calculating a discriminant plane in a support vector machine using a plurality of feature vectors of the first target read from the feature vector storage unit as training data;
Distance calculating means for calculating the distance between the second object and the first object as a length obtained by projecting the distance between the feature vectors onto the normal vector of the discrimination surface;
For each of the second objects, the first object that is located in the vicinity of the second object and has the customer's label is extracted according to the calculated distances, and is used for evaluating the second object Determining means for determining as information;
Including the device.   The apparatus according to claim 1, wherein the discriminant plane is a discriminant plane obtained by applying a biased support vector machine.   The apparatus according to claim 1, wherein the discriminant plane is a discriminant plane obtained by applying a soft margin method in a biased support vector machine.   The attribute information related to the plurality of attributes that are elements of the feature vector of each target includes attribute information related to third-party evaluation of the target, attribute information related to the scale and sales of the target, and attributes related to the business type and region of the target The apparatus of claim 1, comprising any plurality of attribute information in the information.   The apparatus of claim 1, wherein the feature vector of each object is a normalized feature vector.   The apparatus according to claim 1, wherein the vicinity of the second object is a distance that enters the top Kth when the first objects are arranged in order of increasing distance from the second object.   The apparatus according to claim 1, further comprising: an evaluation unit that evaluates each of the second objects according to the number of the first objects having the customer label determined as the evaluation information of the second object.   And further comprising an evaluation unit that evaluates each of the second objects by a sum of attribute values of specific attributes of the first object having the customer label determined as the evaluation information of the second object. Item 2. The apparatus according to Item 1.   The apparatus according to claim 8, wherein the specific attribute is sales information of the first target.   From the top to the Mth in descending order of the distance from the second object among the first objects having the customer label determined as the evaluation information of the second object The apparatus according to claim 1, further comprising an evaluation unit that evaluates based on one or more of the first objects included. In a computer, a method for supporting evaluation of new customer candidates,
In a support vector machine, a set of feature vectors whose elements are attribute information about a plurality of preselected attributes, which are obtained for each of a plurality of objects having either a customer or non-customer label, are used as training data. Calculating the discriminant plane by a computer calculation process;
The distance between each object listed in the customer candidate list and each of the plurality of objects having the label of either the customer or the non-customer, and the distance between the respective feature vectors Calculating the length projected onto the line vector by a computer calculation process; and
For each object in the customer candidate list, an object located in the vicinity of the object and having the customer label is extracted according to the calculated distances, and the identification information of each extracted object corresponds to the customer candidate list Storing in the storage unit in association with the object;
Including methods. A program for supporting evaluation of new customer candidates using a computer, the program being stored in the computer,
In a support vector machine, a set of feature vectors whose elements are attribute information about a plurality of preselected attributes, which are obtained for each of a plurality of objects having either a customer or non-customer label, are used as training data. Calculating a discriminant plane;
The distance between each object listed in the customer candidate list and each of the plurality of objects having the label of either the customer or the non-customer, and the distance between the respective feature vectors Calculating a length projected onto a line vector;
For each object in the customer candidate list, an object located in the vicinity of the object and having the customer label is extracted according to the calculated distances, and the identification information of each extracted object corresponds to the customer candidate list A program for executing the step of storing in the storage unit in association with the target.

Images