Information processing device, information processing method, and program

Abstract

To solve the problem of providing a user with advice for enabling an AI platform to be used more appropriately, this information processing device comprises an acquisition means (21) that acquires log information and a resource usage rate relating to a target user in an environment to be analyzed by machine learning, a generation means (22) that generates advice information for the target user by inputting the log information and resource usage rate relating to the target user, which are acquired by the acquisition means, to an advice information generation model trained to generate advice information from log information and a resource usage rate in an environment to be analyzed; and an output means (23) that outputs the advice information generated by the generation means.

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program that can provide a user with advice to enable the user to more appropriately utilize an AI platform. [Background technology]

[0002] In recent years, companies have been actively developing services and systems that incorporate AI, but building analytical models that utilize AI often requires large investments.

[0003] This is where AI platforms come in. AI platforms provide the equipment and software necessary to build analytical models, and by utilizing the AI ​​platform, the burden on users can be significantly reduced.

[0004] However, there are cases where short-term and small-scale projects in AI platform services fail. In such cases, service users may end up canceling the service. For this reason, there is a demand for technology that can provide users with advice to help them make better use of the AI ​​platform.

[0005] On the other hand, from the viewpoint of churn prediction, Patent Document 1 proposes a technique for generating a sales activity plan using churn prediction, and allowing sales representatives to carry out sales activities efficiently at the appropriate time. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application No. 2021-64406 Summary of the Invention [Problem to be solved by the invention]

[0007] In the case of an AI platform, it is difficult to prevent cancellations unless progress is made in a user's project. For this reason, even if the technology of Patent Document 1 is used, it is difficult to generate suitable advice regarding the AI ​​platform.

[0008] One aspect of the present invention has been made in consideration of the above-mentioned problems, and one example of its objective is to be able to provide users with advice that will enable them to make better use of an AI platform. [Means for solving the problem]

[0009] An information processing device according to one aspect of the present invention comprises an acquisition means for acquiring log information and resource usage rate of a target user in an analysis environment where analysis is performed using machine learning, a generation means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment, and an output means for outputting the advice information generated by the generation means.

[0010] An information processing method according to one aspect of the present invention includes acquiring log information and resource usage of a target user in an analysis environment where analysis is performed using machine learning, generating advice information about the target user by inputting the log information and resource usage of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage in the analysis environment, and outputting the generated advice information.

[0011] A program relating to one aspect of the present invention causes a computer to function as an information processing device that includes an acquisition means for acquiring log information and resource usage rate of a target user in an analysis environment where analysis is performed using machine learning, a generation means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment, and an output means for outputting the advice information generated by the generation means. Effect of the Invention

[0012] According to one aspect of the present invention, it is possible to provide a user with advice that enables the user to more appropriately utilize an AI platform. [Brief description of the drawings]

[0013] [Figure 1] 1 is a block diagram showing an example of the configuration of an information processing device according to a first exemplary embodiment of the present invention. [Diagram 2] 1 is a flowchart showing a flow of an information processing method according to a first exemplary embodiment of the present invention. [Diagram 3] FIG. 11 is a block diagram showing an example of the configuration of a churn prediction system according to an exemplary embodiment 2 of the present invention. [Figure 4] FIG. 4 is a block diagram showing an example of the configuration of a churn prediction device. [Diagram 5] FIG. 13 is a diagram for explaining a prediction method of the advice information generation model 81. [Figure 6] FIG. 11 is a diagram illustrating an example of advice information. [Figure 7] 13 is a flowchart illustrating the flow of advice information output processing by the churn prediction device 10. [Figure 8] FIG. 1 is a block diagram showing an example of the configuration of a cancellation learning and prediction device 10A. [Figure 9] 11 is a flowchart illustrating the flow of a learning process performed by a cancellation learning and prediction device 10A. [Figure 10]FIG. 1 is a block diagram showing an example of the configuration of a cancellation learning and prediction device 10A. [Figure 11] FIG. 1 illustrates an example of a computer that executes instructions of a program to realize each function. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0014] Example 1 A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is a basic form of the exemplary embodiments described later.

[0015] <Overview of information processing device 20> The information processing device 20 according to this exemplary embodiment is, broadly speaking, a device that generates information for an operator providing an AI platform to give advice to users using operation monitoring data.

[0016] More specifically, the information processing device 20, for example, An acquisition means for acquiring log information and resource usage of a target user in an analysis environment in which analysis by machine learning is performed; a generating means for generating advice information regarding a target user by inputting the log information and the resource usage rate of the target user acquired by the acquiring means into an advice information generation model trained to generate advice information from the log information and the resource usage rate in an analysis environment; an output means for outputting the advice information generated by the generation means; It is equipped with:

[0017] <Configuration of information processing device 20> The configuration of an information processing device 20 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing an example of the configuration of the information processing device 20.

[0018] 1, the information processing device 20 includes an acquisition unit 21, a generation unit 22, and an output unit 23. The acquisition unit 21 is a component that realizes an acquisition means in this exemplary embodiment. The generation unit 22 is a component that realizes a generation means in this exemplary embodiment. The output unit 23 is a component that realizes an output means in this exemplary embodiment.

[0019] The acquisition unit 21 acquires log information and resource usage rates of a target user in an analysis environment in which analysis by machine learning is performed. Here, the analysis environment in which analysis by machine learning is performed is, for example, an AI platform.

[0020] Log information includes, for example, user access history, execution history when analysis is performed on the AI ​​platform, error history, etc.

[0021] The resource utilization rate is, for example, the utilization rate of the CPU, memory, GPU, etc.

[0022] The generation unit 22 generates advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition unit 21 into an advice information generation model that has been trained to generate advice information from log information and resource usage rate in the analysis environment.

[0023] The advice information generation model is, for example, a prediction model including a decision tree rule and a linear regression model. For example, the log information and resource usage rate of the target user are classified into cases based on the rule, and prediction is performed for each case using the linear regression model.

[0024] The advice information generation model is, for example, a model trained by machine learning with reference to training data including user log information and a plurality of pairs of resource usage rates and advice information.

[0025] For example, the time when a user will cancel the AI ​​platform service is predicted as a prediction using a linear regression model, and the factors that will cause the user to cancel the AI ​​platform service are predicted by rule-based case classification.

[0026] Then, advice information including advice for eliminating the cause of cancellation is generated.

[0027] The output unit 23 outputs the advice information generated by the generation unit 22 .

[0028] <Flow of information processing method by information processing device 20> The flow of the information processing method executed by the information processing device 20 configured as above will be described with reference to Fig. 2. Fig. 2 is a flowchart showing the flow of the information processing method. As shown in the figure, the information processing includes steps S11, S12, and S13.

[0029] In step S11, the acquisition unit 21 acquires log information and resource usage rates of a target user in an analysis environment in which analysis by machine learning is performed.

[0030] In step S12, the generation unit 22 generates advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition unit 21 into an advice information generation model that has been trained to generate advice information from log information and resource usage rate in the analysis environment.

[0031] In step S13, the output unit 23 outputs the advice information generated by the generation unit 22.

[0032] <Effects of the information processing device 20 and the information processing method> According to the information processing device 20 according to this exemplary embodiment, the log information and resource usage rate of the target user are acquired, and advice information regarding the target user is generated by inputting the log information and resource usage rate of the target user into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment. In this way, advice that allows the user to more appropriately utilize the AI ​​platform can be provided to the user. For example, advice for performing a more accurate analysis can be provided to the user, and the project can be prevented from stalling.

[0033] Exemplary embodiment 2 A second exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first exemplary embodiment are given the same reference numerals, and the description thereof will be omitted as appropriate.

[0034] <Configuration of churn prediction system 1> The configuration of the churn prediction system 1 according to this exemplary embodiment will be described with reference to Fig. 3. Fig. 3 is a diagram illustrating the churn prediction system. As shown in Fig. 3, the churn prediction system includes a churn prediction device 10 and an AI platform 30.

[0035] (AI Platform 30) The AI ​​platform 30 provides devices and software necessary for building an analytical model using AI, and can be used, for example, by users who have signed up.

[0036] A user who has signed a contract with a business operator obtains, for example, a user ID and a password, and is permitted to access the AI ​​platform 30. This allows the user to build an analytical model using the AI ​​platform. In this manner, the AI ​​platform service is provided by the business operator.

[0037] As an example, the user is a data scientist who, at the request of a retail company, aims to build an analytical model for predicting product sales as accurately as possible. At this time, the retail company provides information such as past sales records, customer attributes, and sales stores of the product, and a project is started to complete the analytical model within a deadline specified by the retail company.

[0038] The AI ​​platform 30 includes a storage 31. The storage 31 stores learning data, prediction data, and the like. The learning data is data including, for example, information on past accumulated product sales records, customer attributes, sales stores, and the like. The prediction data is data including, for example, information on sales conditions of a product to be predicted, and the like.

[0039] The AI ​​platform 30 also includes an analysis execution environment 32 and an analysis model 33. The analysis execution environment 32 is configured, for example, by a computer, and executes calculation processing related to the analysis. The analysis execution environment 32 may be realized, for example, by cloud computing.

[0040] The analytical model 33 is generated by processing executed in the analysis execution environment 32, and is, for example, a predictive model configured by model parameters stored in the storage 31.

[0041] A user inputs data on the sales conditions of the product to be predicted into an analysis model 33 to obtain a prediction result 34.

[0042] Furthermore, the AI ​​platform 30 includes a monitoring infrastructure 35 and storage 36. Users of the AI ​​platform cannot access the monitoring infrastructure 35 and storage 36, but businesses can access the monitoring infrastructure 35 and storage 36.

[0043] The monitoring platform 35 monitors the processing executed in the analysis execution environment 32. The monitoring platform 35 records, for example, the date and time of a user's login to the analysis execution environment 32 and the date and time of a user's logout from the analysis execution environment 32 in the storage 36 as an access history to the AI ​​platform 30. The monitoring platform 35 also records the resource usage rate of the analysis execution environment 32 in the storage 36. The resource usage rate is, for example, a CPU usage rate and / or a memory usage rate.

[0044] In addition, the monitoring platform 35 records in the storage 36 the start date and time and end date and time of an analysis job executed by a user in the analysis execution environment 32, as well as an error history indicating whether the job ended normally or not.

[0045] The access history and error information described above may be recorded as log information in the storage 36. That is, the log information includes the number of times the target user has accessed the analysis execution environment and an error history related to the analysis.

[0046] The log information and resource usage rate as described above are stored in the storage 36 as the monitoring results. Furthermore, in addition to the log information and the resource usage rate, the monitoring results may include other information acquired by the monitoring board 35. The monitoring results stored in the storage 36 are provided to the churn prediction device 10 as learning data and prediction data.

[0047] (Canceled subscription prediction device) The churn prediction device 10 is used, for example, by a business operator that provides the AI ​​platform 30. The churn prediction device 10 is a device that predicts the time and reasons for a user who uses the AI ​​platform 30 to cancel a contract for the AI ​​platform service, and generates advice to be provided by the business operator to the user.

[0048] The churn prediction device 10 includes a storage 11. Learning data, prediction data, and the like are stored in the storage 11. The learning data is, for example, data including monitoring results accumulated in the past.

[0049] Furthermore, contract information data is stored in the storage 12 of the cancellation prediction device 10. As an example, the contract information data is data including information indicating the user's service usage start date, the user's cancellation date, and the user's cancellation reason, and is stored, for example, as a database consisting of records for each user. Note that, when a user cancels a contract related to the AI ​​platform service, the business operator confirms the reason for cancellation with the user and records a record of the contract information data including the reason for cancellation.

[0050] The prediction data is data that includes information such as monitoring results relating to the user who is the subject of the prediction, for example.

[0051] The information processing device 20 predicts the cancellation time and cancellation factors, which are the time and the cause of a user canceling the AI ​​platform service, using the advice information generation model. In addition, the information processing device 20 generates advice information using the advice information generation model. The advice information generation model is configured with model parameters obtained by machine learning performed using learning data.

[0052] The prediction result 15 by the information processing device 20 includes the cancellation time and cancellation cause, as well as a countermeasure for eliminating the cancellation cause, and advice information including the cancellation time and cancellation cause is generated based on the prediction result 15.

[0053] <Configuration of the churn prediction device 10> 4 is a block diagram showing a configuration example of the churn prediction device 10. As shown in the figure, the churn prediction device 10 includes an information processing device 20, a storage unit 100, a communication unit 121, an external input unit 122, and an external output unit 123.

[0054] The information processing device 20 is a functional block having the same functions as the information processing device 20 described in the first exemplary embodiment.

[0055] The storage unit 100 is configured, for example, by a semiconductor memory device or the like, and stores data. In this example, prediction data and contract information data are stored in the storage unit 100. The storage unit 100 is a functional block corresponding to the storage 11 and the storage 12 shown in FIG. 3.

[0056] 4, prediction data, contract information data, and model parameters are stored in the storage unit 100. The model parameters are used for prediction by the advice information generation model 81.

[0057] The communication unit 121 is an interface for connecting the churn prediction device 10 to a network. The specific configuration of the network does not limit the present exemplary embodiment, but as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0058] The external input unit 122 accepts various inputs to the churn prediction device 10. The specific configuration of the external input unit 122 is not limited to this exemplary embodiment, but as an example, the external input unit 122 may be configured to include input devices such as a keyboard and a touchpad. The external input unit 122 may also be configured to include a data scanner that reads data via electromagnetic waves such as infrared rays and radio waves, and a sensor that senses the environmental state.

[0059] The external output unit 123 is a functional block that outputs the processing results by the churn prediction device 10. Although the specific configuration of the external output unit 123 is not limited to this exemplary embodiment, as an example, it is configured by a display, a speaker, a printer, etc., and displays various processing results by the churn prediction device 10 on a screen or outputs them as sounds or figures.

[0060] As described in the first exemplary embodiment, the acquisition unit 21 of the information processing device 20 acquires log information and resource usage rates of a target user in the analysis execution environment 32. As described above with reference to Fig. 3, the log information and resource usage rates are stored in the storage 36 as monitoring results and are used as prediction data.

[0061] It should be noted that since resource usage cannot be obtained on a per-user basis, for example, the CPU usage and / or memory usage in the analysis execution environment 32 during the execution time of a job identified from the log information is obtained.

[0062] The acquisition unit 21 may further acquire contract information of the target user. For example, the acquisition unit 21 may acquire the time when the target user started using the service, which is part of the contract information data stored in the storage unit 100.

[0063] In the example of FIG. 4, the generation unit 22 of the information processing device 20 has an advice information generation model 81.

[0064] The advice information generation model 81 classifies the log information and resource usage rate of the target user acquired by the acquisition unit 21 into cases based on rules, and executes prediction by a linear regression model for each case. Note that the data input to the advice information generation model 81 may include contract information of the target user.

[0065] The output unit 23 refers to the predicted cause of cancellation and generates advice information including advice for resolving the cause.

[0066] (Prediction based on advice information generation model) 5 is a diagram for explaining a prediction method of the advice information generation model 81. As an example, the advice information generation model is configured as a heterogeneous mixture prediction model. The heterogeneous mixture prediction model classifies input data into cases according to rules in the form of a decision tree, and makes predictions using a linear model in which different explanatory variables are combined for each case.

[0067] Here, the input data corresponds to prediction data and contract information data, such as the target user's log information and resource usage rate, as well as the time when the target user signed up for the AI ​​platform service.

[0068] In the example of Fig. 5, the input data is classified into cases according to condition A. If the input data does not satisfy condition A (condition A=N), it is further classified into cases according to condition B. In this way, the input data is first classified into cases according to rules in the form of a decision tree.

[0069] When the input data satisfies condition A (condition A=Y), prediction is made using prediction formula 1. That is, some or all of the information contained in the input data is used as explanatory variables, and prediction is made by calculating a linear prediction formula to find the objective variable.

[0070] Furthermore, when the input data satisfies condition B (condition B=Y), prediction is performed using prediction formula 2. Prediction formula 2 may be, for example, a linear prediction formula with a weighting variable different from prediction formula 1. Alternatively, prediction formula 2 may be a linear prediction formula with both weighting variables and explanatory variables different from prediction formula 1. Alternatively, prediction formula 2 may be a linear prediction formula that predicts a response variable different from prediction formula 1.

[0071] Furthermore, when the input data satisfies condition B (condition B=Y), prediction is performed using prediction formula 3. Prediction formula 3 may also be a linear prediction formula with a different weighting variable from prediction formula 1 and prediction formula 2, or may be a linear prediction formula with different weighting variables and explanatory variables. Alternatively, prediction formula 3 may be a linear prediction formula that predicts a different objective variable from prediction formula 1 and prediction formula 2.

[0072] In this way, a linear model corresponding to each prediction formula is selected according to the rules of the decision tree, and prediction is performed using the linear model.

[0073] For example, the time when a user will cancel the AI ​​platform service is predicted by prediction using a linear model corresponding to Prediction Formula 1 to Prediction Formula 3. In addition, the factors that cause a user to cancel the AI ​​platform service are predicted by case classification based on the rules of a decision tree.

[0074] Alternatively, the timing of cancellation may be predicted by case classification based on decision tree rules, and the factors that will cause a user to cancel the AI ​​platform service may be predicted by prediction using a linear model corresponding to Prediction Formulas 1 to 3.

[0075] Alternatively, a linear model corresponding to prediction formula 1 to prediction formula 3 selected through case classification based on the rules of one decision tree, and a linear model corresponding to prediction formula 4 to prediction formula 6 selected through case classification based on the rules of another decision tree may be generated. Then, the time of cancellation may be predicted by each of the linear models corresponding to prediction formula 1 to prediction formula 3, and the cause of cancellation may be predicted by each of the linear models corresponding to prediction formula 4 to prediction formula 6.

[0076] In the real world, predictive analysis requires that the basis for a prediction be easily explained. Complex nonlinear predictions tend to be treated as a black box, even if they have high prediction accuracy. On the other hand, while linear regression and decision trees are simple and easy to understand, they are unable to capture the behavior of complex big data, resulting in low prediction accuracy.

[0077] As described above, by using the advice information generation model 81 of the heterogeneous mixture prediction model, the grounds for prediction can be explained in an easily understandable manner, and highly accurate prediction can be performed.

[0078] Next, an example of the prediction result 15 by the churn prediction device 10 will be described.

[0079] (Example of prediction results 1) For example, when a prediction is made using advice information generation model 81 using prediction data related to user U1, it is predicted that the cancellation will occur within one month, and the reason for cancellation is predicted to be a lack of knowledge about how to use the AI ​​platform.

[0080] In this case, it is confirmed by case analysis based on the rules of the advice information generation model 81 that errors or process stops during job execution by user U1 have occurred frequently in the last 14 days, and that since then, the access frequency has decreased to 50% or less of the level before the occurrence. As a result of this case analysis, the advice information generation model 81 predicts that the cause of cancellation is a lack of knowledge about how to use the AI ​​platform.

[0081] The cancellation time is predicted by a prediction calculation using a linear prediction formula selected as a result of the above-mentioned case classification.

[0082] The output unit 23 refers to the predicted cause of cancellation, refers to the startup guide, and generates advice information including advice to review the setting items specified during analysis. The advice is associated with the cause of cancellation by a pre-stored table or the like as a countermeasure to eliminate the cause of cancellation. In other words, the output unit 23 selects the countermeasure corresponding to the cause of cancellation obtained as a prediction result of the advice information generation model 81 as advice to the target user.

[0083] (Example of prediction results 2) Also, for example, when a prediction is made using the advice information generation model 81 using prediction data related to user U2, it is predicted that the cancellation will occur at the end of the fiscal year, and the reason for cancellation is high costs due to inefficient resource use.

[0084] In this case, job execution by user U2 is concentrated between 17:00 and 19:00 on weekdays, with the memory usage rate exceeding 80%. In addition, after a high load state for 2 to 3 hours, it is confirmed by case discrimination based on the rules of the advice information generation model 81 that there are no operation logs until 9:00 a.m. the following morning.

[0085] The output unit 23 refers to the predicted cause of cancellation and generates advice information including advice to reduce costs by scaling down the server.

[0086] (Example of prediction results 3) Furthermore, for example, when a prediction is made using the advice information generation model 81 using prediction data related to user U3, it is predicted that the cancellation will occur at the end of the fiscal year, and the reason for the cancellation is predicted to be insufficient utilization of the functions of the AI ​​platform 30.

[0087] In this case, it is confirmed by case discrimination based on the rules of the advice information generation model 81 that the frequency of access by the user U3 is on a downward trend and that there is not even one log of the use of the new function / engine of the AI ​​platform 30 released six months ago.

[0088] The output unit 23 refers to the predicted cause of cancellation and generates advice information including advice to try the new function / engine.

[0089] (Example of prediction results 4) Also, for example, when a prediction is made using the advice information generation model 81 using prediction data related to user U4, it is predicted that the cancellation will occur within one month, and the reason for the cancellation is predicted to be a failure of analysis due to a lack of data to analyze.

[0090] In this case, it is confirmed by case discrimination based on the rules of the advice information generation model 81 that the job execution time by user U4 is short, the number of data records is small, and there are many missing values ​​in the data used for the variables.

[0091] The output unit 23 generates advice information that includes advice to collect a sufficient amount of valid data before analysis, with reference to the predicted causes of cancellation.

[0092] (Example of prediction results 5) Also, for example, when a prediction is made using the advice information generation model 81 using prediction data related to user U5, it is predicted that the cancellation will occur at the end of the fiscal year, and the reason for cancellation is predicted to be dissatisfaction with the cost-effectiveness.

[0093] In this case, it is confirmed by case discrimination based on the rules of the advice information generation model 81 that the job executed by the user U5 is executed only once every few months, and the processing takes about one day.

[0094] The output unit 23 refers to the predicted cause of cancellation and generates advice information including advice to select a rate plan suited to the actual situation of the analysis.

[0095] In this way, the advice information generation model 81 is a model that generates advice information and predicts cancellation factors and cancellation times.

[0096] In addition, the advice information generation model 81 may be configured to include a first model that predicts the cancellation factors and cancellation timing from the target user's log information and resource usage rate, and a second model that generates advice information by referring to the cancellation factors and cancellation timing predicted by the first model.

[0097] In this case, for example, the generator 22 is provided with an advice information generation model 81-1 and an advice information generation model 81-2.

[0098] The advice information generation model 81-1 predicts the cancellation cause and cancellation time, for example, using as input the log information and resource usage rate of the target user acquired by the acquisition unit 21. The advice information generation model 81-2 predicts advice information, for example, using as input the cancellation cause and cancellation time predicted by the advice information generation model 81-1.

[0099] 6 is a diagram for explaining an example of advice information. The advice information shown in the figure shows an example of a case where the advice information is displayed on a display of a computer, a smartphone, etc. This display may be a display constituting the external output unit 123 of the churn prediction device 10.

[0100] 6, a customer name display area 201 is displayed on the display 200, and in this example, the customer name "ABC" is displayed. Also displayed on the display 200 is a cancellation time / cause for cancellation display area 202. In this example, the cancellation time / cause for cancellation display area 202 displays "Cancellation time: within one month" and "Cause for cancellation: lack of knowledge about how to use the AI ​​platform."

[0101] Thus, the advice information includes predictions regarding cancellation causes and cancellation timing.

[0102] It is also possible for the display 200 not to display the cancellation time / cause for cancellation display area 202 .

[0103] Furthermore, on the display 200, a business advice display area 203 is displayed, and in this example, "Please present the startup guide to the customer again for advice" is displayed in the business advice display area 203. The business advice display area 203 is an area where advice to be presented to the user from the business's perspective is displayed.

[0104] That is, the advice information includes advice to the provider of the analysis execution environment 32 .

[0105] Also displayed on the display 200 is a customer advice display area 204, and in this example, "Refer to the startup guide and review the setting items specified during analysis" is displayed in the customer advice display area 204. The customer advice display area 204 is an area where advice that the user should directly refer to is displayed.

[0106] That is, the advice information includes advice for the target user.

[0107] The advice information may not include the customer advice display area 204. In this case, the advice information is displayed on a display constituting the external output unit 123 of the churn prediction device 10, for example.

[0108] Alternatively, the advice information may not include the business-oriented advice display area 203. In this case, the advice information is to be displayed, for example, on the display of a smartphone of the user who is the subject of the advice.

[0109] <Flow of advice information output process by the churn prediction device 10> Next, an example of advice information output processing by the churn prediction device 10 will be described with reference to the flowchart of FIG.

[0110] In step S31, the acquisition unit 21 acquires log information.

[0111] The specific process of the acquisition unit 21 is as described above, and at this time, for example, the access history and error information recorded in the storage 36 are acquired as log information. In addition, in step S31, contract information of the target user may be further acquired. For example, the start time of the target user's use of the service, which is part of the contract information data stored in the storage unit 100, may be acquired.

[0112] In step S32, the acquisition unit 21 acquires the resource usage rate. At this time, for example, as described above, the resource usage rate of the analysis execution environment 32 recorded in the storage 36 is acquired.

[0113] In step S33, the generation unit 22 executes prediction using the advice information generation model 81. The specific processing of the generation unit 22 is as described above, and at this time, the log information acquired in step S31 and the resource usage rate acquired in step S32 are input to the advice information generation model 81.

[0114] In step S34, the advice information generation model 81 predicts the cancellation time.

[0115] In step S35, the advice information generation model 81 predicts the cancellation factor.

[0116] In step S36, the advice information generation model 81 predicts advice.

[0117] The specific processing of the advice information generation model 81 is as described above, and at this time, advice for business operators and / or advice for customers are predicted by the advice information generation model. Note that the advice for business operators is, for example, information displayed in the advice display area 203 for business operators in Fig. 6, and the advice for customers is, for example, information displayed in the advice display area 204 for customers in Fig. 6.

[0118] In step S37, the output unit 23 outputs advice information including the advice predicted in step S36. The specific processing of the output unit 23 is as described above, and at this time, for example, the advice information is presented by a screen as described above with reference to FIG.

[0119] The advice information may be printed on a medium such as paper and output, or may be output as sound from a speaker or the like.

[0120] In this manner, the advice information output process is executed.

[0121] Exemplary embodiment 3 Next, a third exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first and second exemplary embodiments are given the same reference numerals, and the description thereof will be omitted as appropriate.

[0122] <Configuration of the cancellation learning and prediction device 10A> The configuration of a churn learning and prediction device 10A according to this exemplary embodiment will be described with reference to Fig. 8. The churn learning and prediction device 10A is a device that further has a function of learning model parameters of an advice information generation model 81 in addition to the functions of the churn prediction device 10.

[0123] Fig. 8 is a block diagram showing a configuration example of a churn learning and prediction device 10A. The churn learning and prediction device 10A shown in Fig. 10 differs from the churn prediction device 10 shown in Fig. 4 in that the information processing device 20 is provided with a training data acquisition unit 24 and a learning unit 25, and that learning data and evaluation data are stored in the storage unit 100.

[0124] In the cancellation learning and prediction device 10A, learning data is stored in the storage unit 100. As described above, the learning data is data including, for example, monitoring results accumulated in the past, and more specifically, includes log information including the number of accesses to the analysis execution environment by each of a plurality of users, and an execution history and an error history related to the analysis. The log information is stored for each user.

[0125] The learning data also includes resource usage, which may be, for example, the CPU usage, memory usage, and / or GPU usage in the analysis execution environment 32 during the execution time of a job executed by each user.

[0126] Moreover, in the cancellation learning and prediction device 10A, evaluation data is stored in the storage unit 100. The evaluation data is, for example, data for verifying whether or not the learned advice information generation model 81 can perform appropriate prediction.

[0127] The training data acquisition unit 24 acquires training data that is data for learning the advice information generation model 81. The training data includes log information and resource usage rate of each user, and attribute information of each user.

[0128] The attribute information is, for example, for a user who has already canceled the service, and includes the target user's service use start date, the target user's cancellation date, the target user's reason for cancellation, and cancellation prevention measures to be presented to the target user.

[0129] The target user's service usage start date, the target user's cancellation date, and the target user's cancellation reason are included in, for example, the contract information data. The cancellation prevention measures to be presented to the target user are, for example, cancellation prevention measures that the business operator believes should have presented in order to prevent the user from canceling. These cancellation prevention measures correspond, for example, to advice for business operators displayed in the business operator advice display area 203 and / or advice for customers displayed in the customer advice display area 204 in FIG. 6.

[0130] The cancellation prevention measures to be presented to the target user are, for example, input by the business operator via the external input unit 122 and stored as a part of the contract information data.

[0131] The training data acquisition unit 24 acquires, for each user, a set of the service usage start date, log information, and resource usage rate, and the cancellation date, cancellation reason, and cancellation prevention measures in order to train the advice information generation model 81. Note that the service usage start date may not be acquired. That is, the training data acquisition unit 24 acquires training data including a plurality of sets of the log information and resource usage rate in the analysis execution environment 32, and advice information.

[0132] The training data may be acquired by the acquiring unit 21. In this case, the training data acquiring unit 24 may not be provided in the cancellation learning and prediction device 10A.

[0133] The learning unit 25 learns the advice information generation model 81 by performing machine learning with reference to the training data acquired by the training data acquisition unit 24 or the acquisition unit 21. More specifically, the learning unit 25 learns the advice information generation model 81 by updating model parameters of the advice information generation model 81 with reference to the training data.

[0134] As described above, the advice information generation model 81 may be configured to include a first model that predicts the cancellation cause and cancellation time, and a second model that generates advice information by referring to the cancellation cause and cancellation time predicted by the first model. In this case, the training data acquisition unit 24 or the acquisition unit 21 acquires training data including a plurality of pairs of the log information and resource usage rate in the analysis environment, and the cancellation cause and cancellation time, and the learning unit 25 causes the first model included in the advice information generation model to learn using the training data.

[0135] <Learning process flow by cancellation learning and prediction device 10A> Next, the learning process performed by the cancellation learning and prediction device 10A will be described with reference to the flowchart in FIG.

[0136] In step S101, the training data acquisition unit 24 acquires training data. The specific processing of the training data acquisition unit 24 is as described above, and at this time, a set of the service usage start date, log information, and resource usage rate, and the cancellation date, cancellation reason, and cancellation prevention measures are acquired for each user. Note that the service usage start date may not be acquired.

[0137] In step S102, the learning unit 25 learns the decision tree of the advice information generation model 81. As a result, appropriate thresholds are learned for variables obtained from the log information and the resource usage rate.

[0138] An example of variables obtained from log information and resource usage is shown below.

[0139] (1) Number of accesses to the analysis execution environment: Number of accesses in the last 30 days, number of accesses in the last 7 days, and the difference between them (2) Errors during analysis: Errors during the last 28, 21, 14, and 7 days, and the difference between them (3) Number of process stops: The number of times in the last 28, 21, 14, and 7 days, and the difference between them (4) CPU utilization exceeded a specified threshold: the time of day, the duration of the threshold, and the number of times the threshold was exceeded (5) Memory utilization exceeded a specified threshold: the time of day, the duration of the threshold, and the number of times the threshold was exceeded (6) Quantity / quality of learning data when creating an analytical model: number of records, data type In step S102, for example, the optimum threshold value when performing case classification using a decision tree using the variables shown in (1) to (6) above is learned.

[0140] In step S103, the learning unit 25 learns a linear model of the advice information generation model 81. As a result, appropriate weighting coefficients are learned for variables obtained from the log information and the resource usage rate. For the variables obtained from the log information and the resource usage rate used at this time, different combinations of variables are selected depending on the result of case classification using a decision tree, for example. That is, as described above with reference to FIG. 5, different linear models are selected depending on the result of case classification using a decision tree, and therefore different combinations of variables are selected depending on each linear model.

[0141] In step S104, the learning unit 25 evaluates the advice information generation model 81 learned through the processes in steps S102 and S103. At this time, the evaluation is performed using evaluation data.

[0142] In step S105, the learning unit 25 refers to the evaluation result of step S104, and deletes unnecessary case classification in the decision tree and the linear model selected as a result of the case classification.

[0143] In step S106, the learning unit 25 updates the model parameters of the advice information generation model 81 by reflecting the processing result of step S105.

[0144] In this manner, the learning process is carried out.

[0145] Exemplary embodiment 4 Next, a fourth exemplary embodiment of the present invention will be described in detail with reference to the drawings. Note that components having the same functions as those described in the first to third exemplary embodiments are given the same reference numerals, and the description thereof will be omitted as appropriate.

[0146] <Configuration of learning device 10B> The configuration of a learning device 10B according to this exemplary embodiment will be described with reference to FIG.

[0147] Learning device 10B is a device configured to have a function related to learning model parameters, unlike cancellation learning and prediction device 10A described above with reference to Fig. 10, and does not have a function related to generating advice information. That is, learning device 10B does not have an acquisition unit 21, a generation unit 22, and an output unit 23 in an information processing device 20. The other configurations of learning device 10B are similar to those of cancellation learning and prediction device 10A, and detailed description thereof will be omitted.

[0148] In the learning device 10B, the learning process described with reference to Fig. 9 is executed, and the model parameters in the storage unit 100 are learned. After the learning is executed with reference to a sufficient number of sets of training data, the model parameters stored in the storage unit 100 of the learning device 10B are stored in a storage medium such as a USB memory. Then, the model parameters stored in the storage medium are used by another device (for example, the churn prediction device 10 in Fig. 3). Alternatively, the model parameters of the learning device 10B may be transferred to another device via a network.

[0149] In this way, for example, model parameters obtained by learning with reference to different training data can be provided to the other devices.

[0150] [Software implementation example] Some or all of the functions of the information processing device 20, the churn prediction device 10, the churn learning and prediction device 10A, and the learning device 10B may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0151] In the latter case, the information processing device 20, the churn prediction device 10, the churn learning and prediction device 10A, and the learning device 10B are realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter, referred to as computer C) is shown in FIG. 11.

[0152] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for operating the computer C as the information processing device 20, the churn prediction device 10, the churn learning and prediction device 10A, or the learning device 10B. In the computer C, the processor C1 reads the program P from the memory C2 and executes it to realize each function of the information processing device 20, the churn prediction device 10, the churn learning and prediction device 10A, or the learning device 10B.

[0153] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a microcontroller, or a combination of these. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination of these.

[0154] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may further include a communication interface for transmitting and receiving data to and from other devices. The computer C may further include an input / output interface for connecting input / output devices such as a keyboard, a mouse, a display, and a printer.

[0155] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can obtain the program P via such a recording medium M. Furthermore, the program P can be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also obtain the program P via such a transmission medium.

[0156] [Additional Note 1] The present invention is not limited to the above-described embodiment, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the above-described embodiment are also included in the technical scope of the present invention.

[0157] [Additional Note 2] Some or all of the above-described embodiments can be described as follows. However, the present invention is not limited to the following described aspects.

[0158] (Appendix 1) An acquisition means for acquiring log information and resource usage of a target user in an analysis environment in which analysis by machine learning is performed; a generating means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquiring means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; an output means for outputting the advice information generated by the generation means; An information processing device comprising:

[0159] (Appendix 2) The log information includes: The number of times the target user accesses the analysis environment; and Error history for the analysis 2. The information processing device according to claim 1,

[0160] (Appendix 3) The advice information includes predictions regarding cancellation causes and cancellation timing. 3. The information processing device according to claim 1 or 2.

[0161] (Appendix 4) The advice information generation model is A first model for predicting a cancellation cause and a cancellation time from log information and a resource usage rate of the target user; a second model that generates advice information by referring to the cancellation cause and cancellation time predicted by the first model; 4. The information processing device according to claim 3,

[0162] (Appendix 5) The advice information includes advice to a provider that provides the analysis environment. 5. An information processing device according to any one of claims 1 to 4.

[0163] (Appendix 6) The advice information includes advice to the target user. 6. An information processing device according to any one of claims 1 to 5.

[0164] (Appendix 7) A learning unit for learning the advice information generation model. 7. The information processing device according to claim 1, further comprising:

[0165] (Appendix 8) The acquisition means includes: Acquire training data including a plurality of pairs of log information, resource usage rates, and advice information in the analysis environment; The learning unit causes the advice information generation model to learn using the training data. 8. The information processing device according to claim 7.

[0166] (Appendix 9) The acquisition means includes: Acquire training data including a plurality of pairs of log information and resource usage rates in the analysis environment, and cancellation causes and cancellation times; The learning unit learns a first model included in the advice information generation model by using the training data. 9. The information processing device according to claim 8.

[0167] (Appendix 10) Obtaining log information and resource usage of target users in an analysis environment where machine learning analysis is performed; generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; and outputting the generated advice information. Information processing methods.

[0168] (Appendix 11) Computer, An acquisition means for acquiring log information and resource usage of a target user in an analysis environment in which analysis by machine learning is performed; a generating means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquiring means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; and outputting means for outputting the advice information generated by the generating means. program.

[0169] [Additional Note 3] A part or all of the above-described embodiments can be further expressed as follows.

[0170] At least one processor, the processor comprising: A process of acquiring log information and resource usage of a target user in an analysis environment where analysis using machine learning is performed; generating advice information about the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; and outputting the advice information generated by the generating means.

[0171] The information processing device may further include a memory that stores a program for causing the processor to execute the acquisition process and the output sequence generation process. The program may be recorded on a computer-readable, non-transitory, tangible recording medium. [Explanation of symbols]

[0172] 1. Cancellation prediction system 10 Cancellation prediction device 10A Cancellation learning and prediction device 10B Learning Device 20 Information processing device 21 Acquisition Department 22 Generation part 23 Output section 24 Training data acquisition section 25 Learning Department 30 AI Platform 32 Analysis Execution Environment 35 Monitoring infrastructure 81 Advice Information Generation Model 100 Storage section 121 Communications Department 122 External input section 123 External Output Section

Claims

1. An acquisition means for acquiring log information and resource usage of a target user in an analysis environment in which analysis by machine learning is performed; a generation means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquisition means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; an output means for outputting the advice information generated by the generation means; An information processing device comprising:

2. The log information includes: The number of times the target user accesses the analysis environment; and Error history for the analysis The information processing device according to claim 1 ,

3. The advice information includes predictions regarding cancellation causes and cancellation timing.

3. The information processing device according to claim 1 or 2.

4. The advice information generation model is A first model for predicting a cancellation cause and a cancellation time from log information and a resource usage rate of the target user; a second model that generates advice information by referring to the cancellation cause and cancellation time predicted by the first model; The information processing apparatus according to claim 3 ,

5. The advice information includes advice to a provider that provides the analysis environment. The information processing device according to claim 1 .

6. The advice information includes advice to the target user. The information processing device according to claim 1 .

7. A learning unit for learning the advice information generation model. The information processing device according to claim 1 , further comprising:

8. The acquisition means includes: Acquire training data including a plurality of pairs of log information, resource usage rates, and advice information in the analysis environment; The information processing apparatus according to claim 7 , wherein the learning unit learns the advice information generation model using the training data.

9. One or more processors, Obtaining log information and resource usage of target users in an analysis environment where machine learning analysis is performed; generating advice information about the target user by inputting the acquired log information and resource usage rate of the target user into an advice information generation model trained to generate advice information from log information and resource usage rate in the analysis environment; and Outputting the generated advice information. An information processing method comprising:

10. Computer, An acquisition means for acquiring log information and resource usage of a target user in an analysis environment in which analysis by machine learning is performed; a generating means for generating advice information regarding the target user by inputting the log information and resource usage rate of the target user acquired by the acquiring means into an advice information generation model trained to generate advice information from the log information and resource usage rate in the analysis environment; and outputting means for outputting the advice information generated by the generating means. program.

Images