Information processing system, information processing method, and program

Abstract

To provide an information processing system, method and program for predicting customer lifetime value, for a job seeker.SOLUTION: An information processing method includes: an acquisition step of acquiring, by a server device, registration information of a job seeker and attribute information including action history related to job seeking, from an employer terminal; and a calculation step of calculating recruitment success rate and estimated sales of the job seeker, based on attribute information and the time elapsed from registration of registration information, and calculates a customer lifetime value using the product thereof. The recruitment success rate is calculated using a probability calculation function that uses the time elapsed, as a variable, and converges over time from an initial value to a value greater than the initial value. The estimated sales are calculated based on a probability distribution of sales generated upon successful recruitment. Before calculating the recruitment success rate and estimated sales, a parameter output model, which is a learning model, receives attribute information as input and outputs a converged value of the probability calculation function, a specific constant included in the probability calculation function, and parameters of the probability distribution.SELECTED DRAWING: Figure 8

Description

[Technical Field]

[0001] The present invention relates to an information processing system, an information processing method, and a program. [Background technology]

[0002] As disclosed in Patent Document 1, a technique is known for calculating a customer lifetime value, which indicates the future profitability of a customer, using customer data. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Laid-Open No. 2003-22359 Summary of the Invention [Problem to be solved by the invention]

[0004] For example, in a job search and recruitment service, it takes a certain amount of time from the time a job seeker registers as a customer until the time sales are generated based on the successful conclusion of a job offer. As such, depending on the nature of the service and the customer, it is difficult to calculate customer lifetime value using conventional technology.

[0005] In view of the above circumstances, the present invention provides an information processing system and the like that can predict the customer lifetime value of a job seeker. [Means for solving the problem]

[0006] According to one aspect of the present invention, an information processing system is provided. The information processing system includes at least one processor. The processor is configured to execute the following steps by reading a program: In the acquisition step, attribute information about a job seeker is acquired. Here, the attribute information includes at least one of the job seeker's registered information and the job seeker's behavioral history related to the job seeker's job search. In the calculation step, a job offer success probability and a predicted sales amount for the job seeker are calculated based on the attribute information and the elapsed time since the registration of the registered information, and a customer lifetime value for the job seeker is calculated using the product of the job offer success probability and the predicted sales amount. Here, the job offer success probability is calculated using a probability calculation function that uses elapsed time as a variable and converges from an initial value to a convergence value greater than the initial value as the elapsed time increases. The predicted sales amount is calculated based on a probability distribution to which sales resulting from the job offer success follow. In the calculation step, before calculating the job offer success probability and the predicted sales amount, the attribute information is input to a parameter output model, and the parameter output model is caused to output a convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution. Here, the parameter output model is a learning model that is trained so as to be able to input attribute information and output convergence values, specific constants, and parameters.

[0007] According to this aspect, the customer lifetime value of a job seeker at any elapsed time can be calculated by using the probability of closing a job offer calculated from a probability calculation function in which parameters are set based on the job seeker's attribute information and in which the elapsed time since registration of the registration information is a variable, and the predicted sales amount calculated from a probability distribution in which parameters are set based on the job seeker's attribute information. Furthermore, since the job seeker's registered information can be used as learning data to be used for training the parameter output model regardless of the elapsed time since registration, a large amount of registered information can be used for training, and information registered immediately after registration can also be used as learning data. This improves the accuracy of calculating the customer lifetime value. [Brief explanation of the drawings]

[0008] [Figure 1] 1 is a configuration diagram illustrating an information processing system 1. FIG. [Figure 2] 2 is a block diagram showing the hardware configuration of the server device 10. FIG. [Figure 3] FIG. 2 is a block diagram showing the hardware configuration of a recruiting party terminal 20 and a job seeker terminal 30. [Figure 4] 1 is a block diagram showing functions realized by a server device 10 (control unit 11), a recruiting party terminal 20 (control unit 21), and a job seeker terminal 30 (control unit 31). [Figure 5] FIG. 2 is a diagram showing an example of a graph of a probability calculation function p(t) expressed by equations (1) and (2). [Figure 6] FIG. 10 is a diagram showing an example of a graph of the probability calculation function p(t) expressed by equations (3) and (4). [Figure 7] FIG. 10 is a diagram illustrating an example of a neural network included in a parameter output model. [Figure 8] 1 is an activity diagram showing the flow of information processing (processing for calculating a customer lifetime value) executed by the information processing system 1. FIG. DETAILED DESCRIPTION OF THE INVENTION

[0009] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below with reference to the accompanying drawings. Various features shown in the following embodiments can be combined with each other.

[0010] Incidentally, the program for realizing the software appearing in one embodiment may be provided as a non-transitory computer-readable medium, or may be provided so that it can be downloaded from an external server, or may be provided so that the program is started on an external computer and its functions are realized on a client terminal (so-called cloud computing).

[0011] Furthermore, various information processing according to an embodiment may realize input and output corresponding to the input. Here, the form of information referenced in such information processing (hereinafter referred to as reference information) is not limited as long as an output is obtained as a result of the input. The reference information may be, for example, rule-based information such as a database, a lookup table, or a predetermined function (including a decision formula such as a regression formula constructed using a statistical method), a trained model that has previously trained the correlation between input and output, or a large-scale language model that can output a desired result by inputting a prompt.

[0012] In one embodiment, a "unit" may include, for example, a combination of hardware resources implemented by a circuit in the broad sense and software information processing that can be specifically realized by these hardware resources. In one embodiment, various information is handled, and this information is represented, for example, by physical values ​​of signal values ​​representing voltage and current, high and low signal values ​​as a binary bit set consisting of 0 or 1, or quantum superposition (so-called quantum bits), and communication and calculations can be performed on a circuit in the broad sense.

[0013] Furthermore, a circuit in the broad sense is a circuit realized by at least an appropriate combination of a circuit, circuitry, processor, memory, etc. The processor may be a general-purpose processor or a dedicated circuit. That is, it includes an application specific integrated circuit (ASIC), a programmable logic device (e.g., a simple programmable logic device (SPLD), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA)), etc.

[0014] 1. Hardware Configuration This section explains the hardware configuration.

[0015] <Information Processing System 1> Fig. 1 is a configuration diagram showing an information processing system 1. The information processing system 1 comprises a communication line 2, a server device 10, a plurality of recruiter terminals 20, and a plurality of job seeker terminals 30. The server device 10, the recruiter terminals 20, and the job seeker terminals 30 are configured to be able to communicate with each other via the communication line 2. The connection between the server device 10, the recruiter terminals 20, and the job seeker terminals 30 may be wired or wireless.

[0016] The information processing system 1 constitutes part of a recruitment and job search system used by multiple recruiters (first recruiter U1 and second recruiter U2) and multiple job seekers (first job seeker U3 and second job seeker U4). The information processing system 1 mainly manages the registered information of job seekers. In one embodiment, the information processing system 1 is made up of one or more devices or components. These components will be described below.

[0017] <Server device 10> 2 is a block diagram showing the hardware configuration of server device 10. Server device 10 includes a control unit 11, a storage unit 12, a communication unit 13, and a communication bus 14. Control unit 11, storage unit 12, and communication unit 13 are electrically connected within server device 10 via communication bus 14.

[0018] <Control unit 11> The control unit 11 processes and controls the overall operations related to the server device 10. The control unit 11 is, for example, a central processing unit (CPU). The control unit 11 realizes various functions related to the server device 10 by reading out predetermined programs stored in the storage unit 12. In other words, information processing by software stored in the storage unit 12 is specifically realized by the control unit 11, which is an example of hardware, and can be executed as each functional unit included in the control unit 11. These will be described in more detail in the next section. Note that the control unit 11 is not limited to being single, and multiple control units 11 may be provided for each function. A combination of these may also be used.

[0019] <Storage section 12> The memory unit 12 stores various pieces of information defined above. This can be implemented, for example, as a storage device such as a solid state drive (SSD) that stores various programs and the like related to the server device 10 executed by the control unit 11, or as a memory such as a random access memory (RAM) that stores temporarily required information (arguments, arrays, etc.) related to the program operations. The memory unit 12 stores various programs, variables, etc. related to the server device 10 executed by the control unit 11.

[0020] <Communications Department 13> The communication unit 13 is preferably a wired communication means such as USB, IEEE1394, Thunderbolt (registered trademark), or wired LAN network communication, but may also include wireless LAN network communication, mobile communication such as 3G / LTE / 5G, or BLUETOOTH (registered trademark) communication as needed. That is, it is more preferable to implement it as a collection of multiple communication means. That is, the server device 10 may communicate various information from the outside via the communication unit 13 and the network.

[0021] The server device 10 may be an on-premise server or a cloud server. The cloud server device 10 may provide the above-described functions and processes in the form of, for example, SaaS (Software as a Service) or cloud computing.

[0022] <Recruiter Terminal 20> Fig. 3 is a block diagram showing the hardware configuration of the recruiting party terminal 20 and the job seeker terminal 30. As shown in Fig. 3A, the recruiting party terminal 20 comprises a control unit 21, a memory unit 22, a communication unit 23, an input unit 24, an output unit 25, and a communication bus 26. The control unit 21, the memory unit 22, the communication unit 23, the input unit 24, and the output unit 25 are electrically connected within the recruiting party terminal 20 via the communication bus 26. The explanation of the control unit 21, the memory unit 22, and the communication unit 23 is the same as the explanation of each unit in the server device 10, and will therefore be omitted. Note that the recruiting party terminal 20 may also be a terminal operated by a recruitment agency that interacts with job seekers on behalf of the recruiter.

[0023] <Input section 24> The input unit 24 accepts operation inputs made by the user. The operation inputs are transferred as command signals to the control unit 21 via the communication bus 26. The control unit 21 can execute predetermined control or calculations based on the transferred command signals as necessary. The input unit 24 may be included in the housing of the recruiter terminal 20 or may be attached externally. For example, the input unit 24 may be implemented as a touch panel integrated with the output unit 25. When the input unit 24 is implemented as a touch panel, the user can input tap operations, swipe operations, etc. to the input unit 24. Instead of a touch panel, a switch button, a mouse, a trackpad, a QWERTY keyboard, etc. can be used as the input unit 24.

[0024] <Output section 25> The output unit 25 displays a screen of a graphical user interface (GUI) that can be operated by the user. The output unit 25 may be included in the housing of the recruiter terminal 20, or may be attached externally. Specifically, the output unit 25 may be implemented as a display device such as a CRT display, a liquid crystal display, an organic EL display, or a plasma display. It is preferable that these display devices are implemented by selectively using them depending on the type of recruiter terminal 20.

[0025] <Job Seeker Terminal 30> 3B, the job seeker terminal 30 includes a control unit 31, a memory unit 32, a communication unit 33, an input unit 34, an output unit 35, and a communication bus 36. The control unit 31, the memory unit 32, the communication unit 33, the input unit 34, and the output unit 35 are electrically connected via the communication bus 36 inside the job seeker terminal 30. The explanation of the control unit 31, the memory unit 32, the communication unit 33, the input unit 34, and the output unit 35 will be omitted as they are the same as the explanation of each unit in the recruiter terminal 20.

[0026] 2. Functional configuration This section describes the functional configuration of this embodiment. Information processing by software stored in the storage unit 12 is specifically realized by the control unit 11, which is an example of hardware, and can be executed as each functional unit included in the control unit 11 (a processor provided in the information processing system 1).

[0027] FIG. 4 is a block diagram showing functions realized by the server device 10 (controller 11), the recruiter terminal 20 (controller 21), and the job seeker terminal 30 (controller 31).

[0028] As shown in Fig. 4A, the server device 10 (control unit 11) includes a basic display control unit 111, an acquisition unit 112, a calculation unit 113, and an artificial intelligence unit 120. As shown in Fig. 4B, the recruiting party terminal 20 (control unit 21) includes a display unit 211 and an operation acquisition unit 212. As shown in Fig. 4C, the job seeker terminal 30 (control unit 31) includes a display unit 311 and an operation acceptance unit 312.

[0029] <Basic display control unit 111> The basic display control unit 111 is configured to display various information on the recruiter terminal 20 and the job seeker terminal 30. For example, the basic display control unit 111 displays a resume and curriculum vitae prepared by the job seeker, a job advertisement and scouting document prepared by the recruiter, etc. on the display unit 211 of the recruiter terminal 20 or the display unit 311 of the job seeker terminal 30.

[0030] Recruiters include organizations such as for-profit corporations (such as companies), non-profit corporations (such as cooperatives and foundations), and public corporations (such as local governments). Recruiters also include recruitment agencies that act as agents of organizations to mediate between job seekers and organizations. Recruitment agencies are also called headhunters or agents.

[0031] <Acquisition part 112> The acquisition unit 112 is configured to acquire attribute information related to job seekers. The attribute information includes at least one of the job seeker's registration information and the job seeker's behavioral history related to the job seeker's job search. The job seeker's registration information and behavioral history are registered in a job seeker database stored in the storage unit 12, for example. The acquisition unit 112 acquires the attribute information from the job seeker database.

[0032] The job seeker's registration information includes the job seeker's resume, curriculum vitae, and other profile information. A "resume" is a document that mainly describes the job seeker's profile, current situation, educational background, work history, desired working conditions, etc., while a "curriculum vitae," also known as a résumé, is a document in which the job seeker conveys to the employer his or her work history, experience, skills, qualifications, etc. related to his or her past work. The job seeker's registration information may also include the job seeker's desired conditions (desired industry, desired job type, etc.).

[0033] "Job-seeking related behavioral history" includes various behaviors related to job seeking, such as logging in to the information processing system 1 and editing one's own registered information, as well as behaviors related to specific job postings, such as applying for a job, registering the job posting in a bookmark list, receiving a scouting document sent based on the job posting, and replying to the scouting document. The bookmark list is a list of job postings that a job seeker has bookmarked (with a specific label), and by referring to the bookmark list, the job seeker can select a job posting to view, apply for, or take other actions. The job seeker database registers information about job postings applied for (job postings), information about job postings registered in the bookmark list, information about job postings replied to scouting documents, the date and time (or number of times) of login to the information processing system 1, the date and time of editing the registered information, and the like, as behavioral history for each job seeker.

[0034] The job posting contains information about the job vacancy, divided into multiple categories such as the name of the position being recruited for, job content and working conditions (annual salary, job type, industry, work location, work style, work environment, etc.), application qualifications (skills), desired personality, and selling points. The job posting may also include items such as the job title, headline, and information about the employer (company size (sales, number of employees, etc.), industry, etc.).

[0035] The acquisition unit 112 receives, for example, input (selection) of a job seeker who is to be the subject of calculation of a customer lifetime value (described later) (hereinafter referred to as a "target job seeker") from the employer terminal 20, and acquires attribute information of the job seeker. The acquisition unit 112 may also automatically acquire attribute information of all job seekers registered in the job seeker database as target job seekers for the customer lifetime value.

[0036] <Calculation unit 113> The calculation unit 113 is configured to calculate the probability of successful recruitment of the target job seeker and the predicted sales amount of the target job seeker based on the attribute information acquired by the acquisition unit 112 and the elapsed time since registration of the target job seeker's registration information, and to calculate the customer lifetime value of the target job seeker using the product of the probability of successful recruitment and the predicted sales amount. In other words, the calculation unit 113 is configured to receive the attribute information and elapsed time of the target job seeker as input, and to output the customer lifetime value of the target job seeker.

[0037] The timing at which sales occur as a result of a job offer being concluded (time elapsed since registration) differs for each job seeker, but by calculating customer lifetime value using a formula that includes elapsed time as a variable, it is possible to predict customer lifetime value at any elapsed time.In addition, sales generation data at any elapsed time can be applied as learning data for the parameter output model described below, rather than being fixed to a specific elapsed time.

[0038] The calculation unit 113 may determine the product of the job offer success probability and the predicted sales amount as the customer lifetime value of the target job seeker, or may determine the customer lifetime value as a value obtained by further calculating the product of the job offer success probability and the predicted sales amount (for example, a value obtained by multiplying the product by another numerical value, or a value obtained by adding or subtracting another numerical value).

[0039] <Probability of successful recruitment> The probability of closing a job offer is calculated using a probability calculation function that uses elapsed time as a variable and converges from an initial value to a convergence value greater than the initial value as the elapsed time increases. That is, the calculation unit 113 calculates the probability of closing a job offer by inputting the elapsed time of the target job seeker into such a probability calculation function.

[0040] The probability calculation function may be a monotonically increasing function. That is, the probability calculation function may be a function in which the job offer success probability increases without decreasing as the elapsed time increases. That is, the probability calculation function may be a function that increases without the job offer success probability decreasing as the elapsed time increases and converges to a constant value (convergent value) at infinite time. Thereby, along with the actual job seeker's success tendency that the longer the elapsed time from the registration of the job seeker information, the higher the success probability (the number of job seekers reaching success increases), the job offer success probability of the target job seeker can be predicted. Therefore, the accuracy of the customer lifetime value calculated by the calculation unit 113 can be improved.

[0041] Examples of the probability calculation function that is a monotonically increasing function include the following formulas (1), (2), (3), (4), etc. In formulas (1), (2), (3), and (4), p is the convergent value, α is a specific constant of a positive value, t is the elapsed time (variable), and t0 is a specific constant of a positive value. The convergent value p is the job offer success probability (sales occurrence probability (upper limit probability) at infinite time) when the elapsed time t is infinite time, has a value larger than the initial value p(0) of the probability calculation function, and typically 0 < p ≤ 1. The method of setting the parameters (convergent value and specific constant) in each formula will be described later. p(t)=p[1-exp(-αt)] ···(1) p(t)=p[1-(1+t) -α ···(2) p(t)=p / {1+exp[-α(t-t0)]} ···(3) p(t)=p / [1+(t / t0) -α ···(4)

[0042] FIG. 5 is a diagram showing an example of a graph of the probability calculation function p(t) expressed by equations (1) and (2). FIG. 6 is a diagram showing an example of a graph of the probability calculation function p(t) expressed by equations (3) and (4). The probability calculation function p(t) of equation (1) converges to a convergence value p at an infinite time (t → ∞) as shown in FIG. 5A. The probability calculation function p(t) of equation (2) converges to a convergence value p at an infinite time (t → ∞) as shown in FIG. 5B. The probability calculation function p(t) of equation (3) converges to a convergence value p at an infinite time (t → ∞) as shown in FIG. 6A. The probability calculation function p(t) of equation (4) converges to a convergence value p at an infinite time (t → ∞) as shown in FIG. 6B.

[0043] Furthermore, it is preferable that the initial value of the probability calculation function is zero. This makes it possible to predict the probability of a job offer being concluded for a target job seeker in line with the actual conclusion tendency, which can be assumed to be such that no job offer is concluded immediately after the job seeker's information is registered. This improves the accuracy of the customer lifetime value calculated by the calculation unit 113. In the above formulas (1)-(4), the probability calculation functions p(t) in formulas (1), (2), and (4) have an initial value p(0), which is the value when t=0, of zero.

[0044] Furthermore, the derivative of the probability calculation function is preferably a monotonically decreasing function. That is, the derivative of the probability calculation function, which is the rate of increase in the job offer success probability with respect to elapsed time, is preferably smaller (or at least not larger) as the elapsed time increases. This makes it possible to predict the job offer success probability of the target job seeker in accordance with a success trend in which the success probability increases significantly in the early stages when the elapsed time is short, and then the rate of increase in the success probability gradually decreases thereafter. This improves the accuracy of the customer lifetime value calculated by the calculation unit 113. Among the probability calculation functions p(t) in the above formulas (1)-(4), the derivatives p'(t) of formulas (1) and (2) are monotonically decreasing functions. That is, the probability calculation functions of formulas (1) and (2) do not include any change points in the slope of the graph, and the job offer success probability increases exponentially with elapsed time.

[0045] The probability calculation function p(t) of the above formula (1) is preferably used as the probability calculation function by which the calculation unit 113 calculates the job offer success probability. By using formula (1) and setting the parameters of formula (1) using a learning model described later, it is possible to improve the prediction accuracy of the job offer success probability.

[0046] <Forecasted sales amount> The predicted sales amount is calculated based on the probability distribution that the sales resulting from the conclusion of a job offer follow. Specifically, the calculation unit 113 predicts the probability distribution of sales resulting from the conclusion of a job offer (i.e., a probability distribution in which sales are a random variable) consisting of a large number of job seekers who have the same or similar attributes as the target job seeker, and calculates the expected value of sales in this probability distribution as the predicted sales amount of the target job seeker. Note that the "sales resulting from the conclusion of a job offer" is the amount of commission (performance-based compensation) that the recruiter pays to the provider of the job-seeking / recruiting service each time a job offer is concluded, and is a positive value. Note that the method for setting the parameters (parameters) that determine the probability distribution will be described later.

[0047] The probability distribution for calculating the predicted sales amount is preferably a continuous probability distribution. In other words, the probability distribution of sales is preferably one that handles continuous sales values. This makes it possible to predict the predicted sales amount of the target job seeker using a probability distribution that reflects the distribution of sales that changes depending on the attributes of the job seeker. As a result, the accuracy of the customer lifetime value calculated by the calculation unit 113 is improved.

[0048] The probability distribution for calculating the predicted sales amount may be a log-normal distribution, a gamma distribution, a truncated normal distribution, or a Pareto distribution. This makes it easier to reproduce the distribution of sales resulting from successful job offers, improving the accuracy of the predicted sales amount for the target job seeker. As a result, the accuracy of the customer lifetime value calculated by the calculation unit 113 is improved.

[0049] The probability distribution for calculating the forecast sales amount is preferably a log-normal distribution. By using the log-normal distribution as the probability distribution and setting the parameters (mean and variance) of the log-normal distribution using a learning model (described later), the forecast accuracy of the forecast sales amount can be improved.

[0050] When the calculation unit 113 uses a log-normal distribution as the probability distribution, the predicted sales amount of the target job seeker is calculated by using the mean μ and the variance σ as μ·exp(1+σ 2 / 2). Therefore, for example, when the calculation unit 113 uses a combination of the probability calculation function p(t) of formula (1) and a log-normal distribution, the customer lifetime value (LTV) of the target job seeker is calculated by the following formula (5). LTV=p[1-exp(-αt)]×μ·exp(1+σ 2 / twenty five)

[0051] <Parameter settings> Before calculating the job offer success probability and the predicted sales amount, the calculation unit 113 inputs the attribute information of the target job seeker into the parameter output model of the artificial intelligence unit 120, and causes the parameter output model to output the convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution. The parameter output model is a learning model that receives attribute information as input and is trained so as to be able to output the convergence value, the specific constant, and the parameter.

[0052] The convergence value and specific constant of the job offer success probability and the parameter of the predicted sales amount are output from the parameter output model according to the attribute information of the target job seeker, and are therefore parameters that depend on the attribute information of the target job seeker. Furthermore, these parameters do not depend on the elapsed time of the target job seeker's registration information.

[0053] The attribute information of the target job seeker to be input to the parameter output model includes first information acquired from the registration information of the target job seeker and second information acquired from the behavioral history of the target job seeker. The calculation unit 113 may input only the first information to the parameter output model, may input only the second information to the parameter output model, or may input both the first information and the second information to the parameter output model.

[0054] The first information includes, for example, feature amounts extracted from the registered information. These feature amounts are data (numerical values) derived from the content described in the registered information, regardless of the format of the registered information. For example, a feature vector obtained by converting the content of the registered information into a vector expressing its features is used as the feature amount. Here, multiple feature vectors may be obtained from one piece of registered information. For example, a feature vector may be obtained for each category in the registered information. Examples of categories in the registered information include age, educational background, current organization, current industry, current job type, current position, current annual salary, possessed skills, possessed qualifications, organizations experienced (years of tenure at each organization), experienced industries (years of experience per industry), experienced job types (years of experience per job type), desired industry, desired job type, desired annual salary, and desired work location.

[0055] The feature vector of the first information includes a first vector obtained by vectorizing a sentence included in the registered information and a second vector obtained by vectorizing an attribute included in the registered information. The calculation unit 113 may use only the first vector, only the second vector, or both the first vector and the second vector as the feature to be input to the parameter output model.

[0056] The calculation unit 113 converts a sentence included in the registered information into a first vector, for example, by the following procedure. First, the calculation unit 113 performs morphological analysis on the text data included in the registered information and divides the sentence into words. Furthermore, the divided words are filtered to remove stop words (functional words such as particles and auxiliary verbs) and extract only nouns. The stop words are determined, for example, based on dictionary definitions, frequency of appearance, etc. Next, the calculation unit 113 performs word normalization (absorbing spelling variations) on the extracted nouns. Word normalization includes procedures such as standardizing character sets, replacing numbers, and standardizing words using a dictionary. Character set standardization includes standardizing uppercase letters to lowercase letters, standardizing half-width kana to full-width kana, etc. Number replacement is replacing numbers included in words with a representative symbol representing a number (for example, "0").

[0057] Thereafter, the calculation unit 113 acquires a first vector by using the sentence of the registration information and the words that have been processed as described above, by any vectorization method such as the TF-IDF method.

[0058] The TF-IDF method is a technique for quantifying importance using the frequency of occurrence of words in a sentence. The calculation unit 113 calculates the number of occurrences of each word contained in the sentence of the registered information in the registered information. The calculation of the number of occurrences is performed, for example, using a sentence analysis model such as Bag of Words. Next, the calculation unit 113 converts the number of occurrences of a word in one registered information into the term frequency (tf) of the word. Furthermore, based on the term frequency (tf) of the word in each registered information, the calculation unit 113 calculates the inverse document frequency (idf), which is the rarity of the occurrence frequency of each word in all registered information from which the data was acquired. Finally, the calculation unit 113 calculates the importance (tf-idf) of each word in each job posting from the term frequency (tf) and the inverse document frequency (idf), and generates a first vector whose components are the importance of each word. The dimension of the first vector is the number of words whose importance is being calculated.

[0059] In addition to the TF-IDF method described above, LSI, LDA, and other methods may be used to quantify each word. Furthermore, the calculation unit 113 may generate a first vector from the text of the job posting using a model that performs vectorization based on distributed representations of words, such as Word2vec or BERT. The TF-IDF method has the advantage of being strong in keyword processing and lightweight. Since keywords that appear in job seeker registration information are characteristic, the TF-IDF method is preferable. Furthermore, the TF-IDF method reduces the weight of words that appear in a large number of registrations, regardless of whether or not a job offer has been concluded, and therefore can suitably determine the feature quantities of the registration information.

[0060] The calculation unit 113 converts the attributes included in the registration information into a second vector, for example, by the following procedure. The calculation unit 113 converts the numerical data or categorical (qualitative) data that are the attributes into a vector, for example, by a method such as One-Hot Encoding. In One-Hot Encoding, the calculation unit 113 converts the numerical data or categorical (qualitative) data that indicate the attributes into a One-Hot vector, each component of which is 0 or 1. The calculation unit 113 generates the second vector by combining the One-Hot vectors of the attributes. The "attributes" are information that are input by selecting from predetermined categories or numerical values, for example, gender, age, annual income, industry, etc.

[0061] The calculation unit 113 may vectorize the registered information using a vectorization model, which is a generative AI including a large-scale language model. In this case, the calculation unit 113 receives the registered information as input, inputs a prompt including an instruction to vectorize and output the registered information to the vectorization model, and causes the vectorization model to output a first vector and / or a second vector. Furthermore, the calculation unit 113 may input, to the vectorization model, a prompt in which, as input and output samples, for example, one or more samples of the registered information and one or more samples of the corresponding first vector and / or second vector are inserted, in addition to the registration information and the instruction to vectorize and output the registered information.

[0062] The feature amount of the registered information may be extracted by a method other than vectorization. Furthermore, the first information may include the amount of information (number of characters) of the registered information as a feature amount. The amount of information of the registered information may be the amount of information of the entire registered information or the amount of information for each item (category). Therefore, the amount of information of the registered information may be, for example, the number of characters in a job seeker's resume.

[0063] The second information includes, for example, feature amounts extracted from the content of the target job posting (job posting) included in the job seeker's behavioral history. Target job postings include job postings to which the job seeker has applied, job postings that the job seeker has added to his / her bookmark list, job postings for which the job seeker has received a scouting document, job postings for which the job seeker has replied to a scouting document received, and job postings that have passed screening (document screening, interview screening, etc.). These feature amounts are data (numerical values) derived from the content described in the job posting, regardless of the format of the job posting. For example, feature vectors obtained by converting the content of the job posting into vectors expressing its features are used as feature amounts. Here, multiple feature vectors may be obtained from one job posting. For example, a feature vector may be obtained for each category in the job posting. Examples of categories in a job posting include employer (organization name), recruiting industry, recruiting job type, recruiting position, required skills, required qualifications, expected annual salary, etc.

[0064] Like the feature vectors of the first information, the feature vectors of the second information include a first vector obtained by vectorizing the sentences included in the job advertisement and a second vector obtained by vectorizing the attributes included in the job advertisement. The calculation unit 113 may use only the first vector, only the second vector, or both the first and second vectors as the feature quantities to input to the parameter output model. The procedure for obtaining the first and second vectors in the second information is the same as the procedure for obtaining the first and second vectors in the first information, except that the registered information is replaced with the job advertisement.

[0065] The feature amount of the behavioral history may be extracted by a method other than vectorization. The second information may also include, as feature amounts, the number of times the job seeker logs in to the information processing system 1, the number of times the job seeker applies for jobs, the number of scouting letters received by the job seeker, the number of replies to the scouting letters by the job seeker, the number of times the job seeker passes screenings (document screening, interview screening, etc.), etc., during a predetermined period.

[0066] The parameter output model is constructed by machine learning (first learning) using attribute information of a large number of job seekers used for learning at the time of registration, and whether or not a job offer has been concluded and the amount of sales actually generated, each of which is linked to the registration information, as first learning data. This allows the parameter output model to be trained so as to output optimal parameters for predicting the probability of a job offer being concluded and the amount of sales, based on the registration information of a large number of job seekers. This improves the accuracy of the customer lifetime value calculated by the calculation unit 113.

[0067] In the first learning data, whether or not a job offer has been concluded may be expressed by whether or not sales are zero. In other words, the first learning data may be data in which the sales of job seekers who have not concluded a job offer are set to zero. Therefore, the first learning data may be composed only of attribute information at the time of registration of the registered information and the amount of actual sales associated with each registered information. Furthermore, the first learning data measures (records) the amount of sales for each job seeker at an arbitrary point in time. The arbitrary point in time at which the amount of sales is measured may be different for each job seeker.

[0068] Specifically, the parameter output model includes a neural network that receives feature quantities of job seeker attribute information as input (explanatory variables) and outputs convergence values, specific constants, and parameters as output (objective variables). FIG. 7 is a diagram illustrating an example of a neural network included in the parameter output model. The neural network has an input layer, a hidden layer (intermediate layer), and an output layer. Note that the neural network may have two or more hidden layers. The neural network in FIG. 7 outputs, in response to an input of four different pieces of attribute information, a predicted value of the convergence value p of the probability calculation function, predicted values ​​of μ (mean) and σ (variance), which are parameters of the sales probability distribution, and a predicted value of the specific constant α of the probability calculation function. In the first learning, learning is performed so that the neural network can output parameters (μ, σ), which are parameters of the lognormal distribution to which sales amount follows, and parameters (p, α) for calculating the contract probability, from the job seeker attributes.

[0069] The number of pieces of attribute information input to the neural network of the parameter output model is not limited to four. Furthermore, if there are multiple specific constants in the probability calculation function, the neural network of the parameter output model outputs multiple specific constants. For example, if the calculation unit 113 uses the probability calculation function of equation (3) or equation (4), the neural network of the parameter output model outputs two specific constants, α and t0, thereby outputting parameters of the sales probability distribution for calculating the predicted sales amount. Therefore, the number of parameters output by the neural network varies depending on the probability calculation function and the shape of the sales probability distribution used by the calculation unit 113.

[0070] When a log-normal distribution is used as the probability distribution of sales, the calculation unit 113 causes the neural network of the parameter output model to output the mean μ and variance σ of the probability distribution as parameters, as shown in FIG. 7. When a probability distribution other than the log-normal distribution is used, the neural network of the parameter output model outputs parameters according to the probability distribution from the output layer. Therefore, the number and type of parameters output by the neural network vary depending on the type of probability distribution (the number and type of parameters that serve as parameters). For example, when the probability distribution is a gamma distribution, the shape parameter and scale parameter are output as parameters; when the probability distribution is a truncated normal distribution, the mean, spread, upper limit, and lower limit are output as parameters; and when the probability distribution is a Pareto distribution, the shape parameter and scale parameter are output as parameters.

[0071] In the first learning, the weighting (weight coefficient) of the neural network is adjusted so that the first loss related to the job offer success probability and the second loss related to sales are each reduced. The first loss is the loss (e.g., the amount of deviation between the two, or a value evaluated by a loss function) between the convergence value (job offer success probability over infinite time) of the probability calculation function output to the output layer when attribute information of the first learning data is input to the input layer of the neural network and the job offer success probability calculated or predicted from the presence or absence of job offer success in the first learning data. The second loss is the loss between the expected value (predicted sales value) of the probability distribution defined by the parameters output to the output layer and the actual sales amount in the first learning data.

[0072] Furthermore, the parameter output model (neural network) undergoes second learning using second learning data. Specifically, the parameter output model is trained to minimize the difference between the predicted value of the job closing probability and the actual job closing probability contained in the second learning data. The predicted value of the job closing probability is a value obtained by inputting elapsed time into a probability calculation function that uses a convergence value and a specific constant output by the parameter output model using attribute information contained in the second learning data as input. The actual job closing probability is a value linked to the attribute information input into the parameter output model when the predicted value is obtained and the elapsed time input into the probability calculation function. The actual job closing probability may be synonymous with the actual number of job closings and is a value indicating how many job closings actually occurred in a specific elapsed time. This makes it possible to construct a parameter output model that can output a specific constant for predicting the job closing probability according to elapsed time without incorporating elapsed time into the neural network's explanatory variables. That is, when a parameter output model is trained using elapsed time as an input (explanatory variable), it is difficult to take into account the characteristic that customer lifetime value increases monotonically with elapsed time, making it difficult to improve the accuracy of prediction, but according to the above-described training, the characteristic of customer lifetime value can be reproduced by not using the elapsed time from the job seeker's registration as an explanatory variable in the parameter output model.As a result, the probability of closing a job offer taking elapsed time into consideration can be predicted with high accuracy, thereby improving the accuracy of the customer lifetime value calculated by calculation unit 113.

[0073] The elapsed time is not used as a parameter to be input to the input layer of the neural network, but is used at the stage of calculating the predicted value of the success probability when calculating the loss to adjust the weighting parameters of the learning model (neural network).This allows for accurate parameter setting that takes into account the sales amount distribution that changes with elapsed time.

[0074] The second learning data in the second learning includes attribute information for learning and data on the change in the probability of a job offer being concluded over time for a job seeker having the attribute information (output of a probability calculation function). The data on the change in the probability of a job offer being concluded is constructed from data on whether or not a job offer has been concluded over time for a large number of job seekers.

[0075] The attribute information for learning contained in the second learning data may be the same as the attribute information for learning contained in the first learning data. In other words, the data of job seekers used in the first learning may be the same as the data of job seekers used in the second learning.

[0076] In the second learning, when attribute information of the second learning data is input to the input layer of the neural network, the parameter output model is trained so as to reduce (minimize) the loss (for example, the amount of deviation between the two, the value evaluated by the loss function, etc.) between the job offer contract probability calculated from the probability calculation function p(t) using the convergence value p and specific constant α (and further t0 as necessary) of the probability calculation function output to the output layer and the elapsed time t in the second learning data, and the job offer contract probability in the second learning data. In other words, the parameter output model is trained by adjusting the weighting (weight coefficients) of the neural network so that the degree of agreement between the contract probability calculated based on the parameters output by the neural network and the actual contract probability is increased.

[0077] The second learning is performed, for example, after the weights of the neural network have been adjusted to a certain extent by the first learning. The first learning and the second learning may be performed alternately.

[0078] The first learning data used in the first learning and the second learning data used in the second learning can each cover all job seekers registered in the job seeker database. That is, the first learning data and the second learning data include attribute information, etc., of both job seekers who have been hired and job seekers who have not been hired. The first learning data and the second learning data also include attribute information, etc., of multiple job seekers whose registration information has been registered for different periods of time. Furthermore, the second learning data may include multiple data for the same job seeker at different elapsed times.

[0079] The parameter output model is successively updated through the first learning and / or second learning using new learning data, such as when the registration information of a new job seeker is registered in the job seeker database, when a new job offer is concluded and sales are confirmed, etc.

[0080] The customer lifetime value output by calculation unit 113 can be used, for example, to manage customers (job seekers) by scoring based on the customer lifetime value, estimate the effectiveness of advertising in job-seeking and recruitment services, etc. Customer management includes, for example, preferentially sending scouting documents to job seekers with high customer lifetime values, sorting a list of job seekers by the magnitude of their customer lifetime values, etc.

[0081] <Artificial Intelligence Department 120> The artificial intelligence unit 120 is configured to receive input from each functional unit and return the instructed output. The artificial intelligence used by the server device 10 in each functional unit may be a common one, or may be prepared individually for each functional unit.

[0082] The artificial intelligence unit 120 is an AI (Artificial Intelligence) equipped with learning models such as Transformers including GPT (Generative Pretrained Transformer, including GPT-1, GPT-2, and GPT-3), BERT (Bidirectional Encoder Representations from Transformers), and BART (Bidirectional and Auto-regressive Transformer), and language models such as Recurrent Neural Networks (RNNs), and may include generative AI.

[0083] The language model is an example of a learning model based on a machine learning algorithm. Specific examples of machine learning algorithms include nearest neighbor methods, naive Bayes methods, decision trees, support vector machines, and deep learning using neural networks. The artificial intelligence unit 120 can apply the above algorithms as appropriate.

[0084] The artificial intelligence unit 120 may have a trained model constructed by a learning method such as supervised learning, unsupervised learning, or self-supervised learning. In supervised learning, machine learning is performed using training data (training data). The training data consists of pairs of input data for learning and output data (correct answer data). Furthermore, the language model may not only be trained for a specific task, but also be a general-purpose model that can be used for a wide range of tasks.

[0085] The artificial intelligence unit 120 may be a general-purpose natural language processing learning model, such as a large language model (LLM), which has learned a huge amount of data. An LLM is a learning model that has previously learned a large amount of data, such as text data (e.g., (i) web content on the Internet, or (ii) data stored in a specified database), and can execute various language processing tasks when given a task. Such a general-purpose learning model includes a language model that can handle various tasks without fine-tuning, using one-shot learning, few-shot learning, etc. Furthermore, a general-purpose learning model may also be configured to handle various tasks using zero-shot learning. The artificial intelligence used in each functional unit of the control unit 11 may be a separate learning model, or a common general-purpose learning model.

[0086] The learning models included in the artificial intelligence unit 120 (learning models used in each functional unit, such as parameter output models) can undergo additional learning as transfer learning or fine tuning. For example, each time new job seeker registration information and job posting registrations occur, the artificial intelligence unit 120 may perform additional learning and fine tuning using these as new training data. This improves the accuracy of the information output from the learning model.

[0087] The learning model included in the artificial intelligence unit 120 may be a learning model (distilled model) obtained by knowledge distillation using an original learning model. In knowledge distillation, a trained model such as a large-scale language model is used as a teacher model, and the parameters of the student model are adjusted to reduce the output loss (Soft Target Loss) of the student model (distilled model) relative to the output (Soft Target) of the teacher model, thereby learning the student model, which becomes the distilled model. Alternatively, the student model may be learned to reduce the output loss (Hard Target Loss) of the student model relative to the correct label (Hard Target) of the teacher data (combination of input data and output data of the learning model). Compared to the original learning model (teacher model), the distilled model has a smaller number of parameters and a smaller processing load while maintaining performance similar to the learning model. Therefore, using a distilled model can reduce the cost of the information processing system 1.

[0088] For example, the parameter output model etc. may be a distilled model trained using a combination of input data and output data in a large-scale language model as training data. Furthermore, when the information processing system 1 is introduced, a large-scale language model may be used as the parameter output model etc., and when training data from the large-scale language model is accumulated, a distilled model obtained by knowledge distillation using the training data may be used as the parameter output model etc.

[0089] <Display> The display unit 211 of the recruiting party terminal 20 and the display unit 311 of the job seeker terminal 30 each display a screen indicated by the screen data transmitted from the server device 10.

[0090] <Operation acquisition part> The operation acquisition unit 212 of the recruiting party terminal 20 accepts operations by a user (recruiter) who uses the recruiting party terminal 20. The operation acceptance unit 312 of the job seeker terminal 30 accepts operations by a user (job seeker) who uses the job seeker terminal 30.

[0091] 3. Information Processing Method This section describes an information processing method of the server device 10. This information processing method is executed by a computer, with each unit of the server device 10 acting as each step.

[0092] This information processing includes an acquisition step and a calculation step. In the acquisition step, attribute information related to a target job seeker is acquired. In the calculation step, a job offer success probability and a predicted sales amount for the target job seeker are calculated based on the attribute information and the time elapsed since the registration of the registration information, and the customer lifetime value of the target job seeker is calculated using the product of the job offer success probability and the predicted sales amount. In addition, in the calculation step, before calculating the job offer success probability and the predicted sales amount, the attribute information is input into a parameter output model, and the parameter output model is caused to output a convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution.

[0093] 8 is an activity diagram showing the flow of information processing (customer lifetime value calculation processing) executed by the information processing system 1. Below, the information processing will be explained along with each activity in this activity diagram.

[0094] The process of calculating a customer lifetime value begins, for example, with a recruiter selecting a target job seeker for whom the customer lifetime value is to be calculated. The recruiter inputs the target job seeker in the recruiter terminal 20 (activity A101). The server device 10 receives the input of the target job seeker in the recruiter terminal 20 and acquires attribute information of the target job seeker (activity A102).

[0095] The server device 10 inputs the acquired attribute information of the target job seeker into a parameter output model to acquire a convergence value, a specific constant, and a parameter (activity A103). This determines the probability calculation function and the probability distribution of sales. Next, the server device 10 inputs the elapsed time of the target job seeker's registration information into the probability calculation function to calculate the probability of closing a job offer (activity A104). Furthermore, the server device 10 calculates the customer lifetime value using the product of the calculated probability of closing a job offer and the predicted sales amount based on the probability distribution (activity A105).

[0096] 4. Effect The operation of this embodiment can be summarized as follows. Specifically, by using the probability of closing a job offer calculated from a probability calculation function in which parameters are set based on the job seeker's attribute information and in which the elapsed time since registration of the registration information is a variable, and the predicted sales amount calculated from a probability distribution in which parameters are set based on the job seeker's attribute information, the customer lifetime value of a job seeker at any elapsed time can be calculated. Furthermore, since the job seeker's registered information can be used as learning data for training the parameter output model regardless of the elapsed time since registration, a large amount of registered information can be used for training regardless of the elapsed time since the job seeker's registration, and the latest registered information immediately after registration can also be used as learning data. This allows the latest trends in closings (sales generation) to be taken into account as learning data to build a learning model, thereby improving the accuracy of calculating the customer lifetime value.

[0097] In particular, for services where the distribution of sales is biased toward zero and where it takes a long time for sales to be observed, this embodiment makes it possible to use data from the most recent period for learning, thereby improving the accuracy of predicting job offers (sales).In addition, when predicting customer lifetime value, the future period to be predicted (the period until a job offer is concluded) can be set arbitrarily, thereby increasing the flexibility of the learning model for calculating customer lifetime value.

[0098] For example, in the case of recruitment and job-seeking services, it generally takes several months or more from the time a job seeker registers until sales are generated, resulting in a distribution with a large proportion of members with zero sales resulting from the job seeker's successful contract (employment decision). Therefore, conventional customer lifetime value prediction requires learning data from job seekers who have been registered for a long period of time (e.g., one year or more). In contrast, this embodiment calculates the probability of sales generation from a function that changes over time, not just from the job seeker's attributes, and predicts the parameters of this function from the job seeker's attributes, thereby taking into account the time dependency of sales. Furthermore, sales data for each job seeker observed at any time point can be used to train the learning model, and any prediction period can be set even at the time of predicting customer lifetime value.

[0099] Although the embodiment of the present invention has been described above, the present invention is not limited to this and can be modified as appropriate within the scope of the technical idea of ​​the invention.

[0100] 5.Other In the above embodiment, the server device 10 performs various storage and control functions. However, multiple external devices may be used instead of the server device 10. That is, various information and programs may be distributed and stored in multiple external devices using blockchain technology or the like. In particular, the artificial intelligence unit 120 may be an external component of the server device 10. In this case, the external artificial intelligence unit 120 may be provided, for example, by an artificial intelligence service server and configured to receive inputs from each functional unit of the server device 10, receive requests to execute artificial intelligence services, and return the instructed output as a processing result to the server device 10. The artificial intelligence service server may be a server that provides services using a language model as a learning model, or a server that executes language processing tasks using a language model. The artificial intelligence service server may be constructed using LLM. The artificial intelligence service server receives inputs of prompts such as text, images, and voice, and generates and responds to the prompts.

[0101] The aspect of this embodiment is not limited to the information processing system 1, and may be an information processing method or a program. The information processing method includes steps executed by the information processing system 1. The program causes a computer to execute the steps of the information processing system 1.

[0102] This embodiment may be applied to predicting customer lifetime value in other customer services, not just recruitment and job search services. That is, in the above embodiment, a "job seeker" may be a "customer," a "job search behavioral history" may be a "customer service behavioral history," a "job offer success" may be a "sales occurrence," and a "job offer success probability" may be a "sales occurrence probability." In this case, the sales occurrence probability in customer service may be calculated from a function that varies depending on the customer's attributes and the time elapsed since the customer registered with the service, and the parameters of the function may be calculated using a learning model such as a neural network based on the customer's attribute information, including the customer's registration information or behavioral history. This allows for the time dependency of customer sales to be taken into account, and sales data for each customer observed at any time point may be used to train the learning model.

[0103] It may be provided in the following manner.

[0104] (1) An information processing system, comprising at least one processor, the processor being configured to execute the following steps by reading a program: in an acquisition step, attribute information about a job seeker is acquired, wherein the attribute information includes at least one of registered information about the job seeker and a behavioral history related to the job search of the job seeker; in a calculation step, a job offer success probability and a predicted sales amount for the job seeker are calculated based on the attribute information and an elapsed time since the registration of the registered information, and a customer lifetime value of the job seeker is calculated using a product of the job offer success probability and the predicted sales amount; wherein the job offer success probability is calculated using the elapsed time as a variable, and an information processing system in which the attribute information is input to a parameter output model before calculating the job closing probability and the predicted sales amount, and the parameter output model is caused to output the convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution, and the attribute information is input to the parameter output model before calculating the job closing probability and the predicted sales amount, and the parameter output model is caused to output the convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution, and the parameter output model is input to the attribute information and is trained to be able to output the convergence value, the specific constant, and the parameter.

[0105] (2) In the information processing system described in (1) above, the parameter output model is constructed by machine learning using the attribute information at the time of registering the registration information, and the presence or absence of a job offer and the amount of sales actually generated, linked to the registration information, as learning data.

[0106] (3) In the information processing system described in (2) above, the parameter output model is trained to reduce the loss between the predicted value of the job offer conclusion probability and the actual job offer conclusion probability, wherein the predicted value is a value obtained by inputting the elapsed time into the probability calculation function that uses the convergence value and the specific constant output by the parameter output model using the attribute information as input, and the actual job offer conclusion probability is a value linked to the attribute information input into the parameter output model when the predicted value is obtained and the elapsed time input into the probability calculation function.

[0107] (4) In the information processing system according to any one of (1) to (3) above, the probability calculation function is a monotonically increasing function.

[0108] (5) In the information processing system described in (4) above, the initial value of the probability calculation function is zero.

[0109] (6) In the information processing system described in (4) or (5) above, the derivative of the probability calculation function is a monotonically decreasing function.

[0110] (7) In the information processing system according to any one of (4) to (6) above, the probability calculation function is expressed by p(t) in the following formula (1), where p is the convergence value, α is the specific constant of a positive value, and t is the elapsed time: p(t)=p[1-exp(-αt)] (1)

[0111] (8) The information processing system according to any one of (1) to (7) above, wherein the probability distribution is a continuous probability distribution.

[0112] (9) In the information processing system described in (8) above, the probability distribution is a log-normal distribution, a gamma distribution, a truncated normal distribution, or a Pareto distribution.

[0113] (10) In the information processing system described in (9) above, the probability distribution is a log-normal distribution, and in the calculation step, the parameter output model is caused to output the mean and variance of the probability distribution as the parameters.

[0114] (11) An information processing method, comprising steps executed by the information processing system according to any one of (1) to (10) above.

[0115] (12) A program for causing a computer to execute each step of the information processing system described in any one of (1) to (10) above. Of course, this is not the case.

[0116] Finally, while various embodiments of the present disclosure have been described, they are presented as examples and are not intended to limit the scope of the invention. The novel embodiments may be embodied in various other forms, and various omissions, substitutions, and modifications may be made without departing from the spirit of the invention. Such embodiments and modifications are intended to be included within the scope and spirit of the invention, as well as within the scope of the inventions and their equivalents as defined in the claims. [Explanation of symbols]

[0117] 1: Information processing system 2: Communication line 10: Server device 11: Control section 12: Storage section 13: Communications Department 14: Communication bus 20: Recruiter terminal 21: Control unit 22: Storage section 23: Communications Department 24: Input section 25: Output section 26: Communication bus 30: Job seeker terminal 31: Control unit 32: Storage section 33: Communications Department 34: Input section 35: Output section 36: Communication bus 111: Basic display control section 112: Acquisition Department 113: Calculation section 120: Artificial Intelligence Department 211:Display section 212: Operation acquisition section 311: Display section 312: Operation reception section

Claims

1. An information processing system, at least one processor; The processor is configured to execute the following steps by reading the program: In the acquisition step, attribute information about job seekers is acquired, wherein the attribute information includes at least one of registration information of the job seeker and a behavioral history of the job seeker related to job seeking, In the calculation step, a job offer success probability of the job seeker and a predicted sales amount of the job seeker are calculated based on the attribute information and the elapsed time since the registration of the registration information, and a customer lifetime value of the job seeker is calculated using the product of the job offer success probability and the predicted sales amount; Here, the job offer contract probability is calculated using a probability calculation function that uses the elapsed time as a variable and converges from an initial value to a convergence value greater than the initial value as the elapsed time increases, The predicted sales amount is calculated based on a probability distribution to which sales resulting from the conclusion of a job offer will follow; In the calculation step, before calculating the job offer contract probability and the predicted sales amount, the attribute information is input to a parameter output model, and the parameter output model is caused to output the convergence value of the probability calculation function, at least one specific constant included in the probability calculation function, and a parameter of the probability distribution; Here, the parameter output model is a learning model that has been trained so as to be able to input the attribute information and output the convergence value, the specific constant, and the parameter.

2. 2. The information processing system according to claim 1, The parameter output model is an information processing system constructed by machine learning using the attribute information at the time of registering the registration information, and the presence or absence of a job offer and the amount of sales actually generated, which are linked to the registration information, as learning data.

3. 3. The information processing system according to claim 2, The parameter output model is trained so as to reduce the loss between the predicted value of the job offer contract probability and the actual job offer contract probability, wherein the predicted value is a value obtained by inputting the elapsed time into the probability calculation function that uses the attribute information as input and the convergence value output by the parameter output model and the specific constant, An information processing system, wherein the actual job offer closing probability is a value linked to the attribute information input into the parameter output model when the predicted value is obtained and the elapsed time input into the probability calculation function.

4. 2. The information processing system according to claim 1, An information processing system, wherein the probability calculation function is a monotonically increasing function.

5. 5. The information processing system according to claim 4, An information processing system, wherein the initial value of the probability calculation function is zero.

6. 5. The information processing system according to claim 4, An information processing system, wherein the derivative of the probability calculation function is a monotonically decreasing function.

7. 5. The information processing system according to claim 4, The probability calculation function is expressed by p(t) in the following formula (1), where p is the convergence value, α is the specific constant of a positive value, and t is the elapsed time. p(t)=p[1-exp(-αt)]...(1)

8. 2. The information processing system according to claim 1, The information processing system, wherein the probability distribution is a continuous probability distribution.

9. 9. The information processing system according to claim 8, The information processing system, wherein the probability distribution is a log-normal distribution, a gamma distribution, a truncated normal distribution, or a Pareto distribution.

10. 10. The information processing system according to claim 9, the probability distribution is a log-normal distribution, In the calculation step, the information processing system causes the parameter output model to output the mean and variance of the probability distribution as the parameters.

11. An information processing method, comprising: An information processing method comprising the steps executed by the information processing system according to any one of claims 1 to 10.

12. A program, A program for causing a computer to execute each step of the information processing system according to any one of claims 1 to 10.

Images