JP2010108404A - Computer system for performing risk prediction in project and method and computer program thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To cope with a problem in which a method for predicting a risk factor latent in a project is required. <P>SOLUTION: A computer system is provided for performing risk prediction in a project. The computer system includes: a collection part which collects answer data to a plurality of questions related to each project; a degree-of-risk assignment part which assigns, based on answer data to each question for each key item, a degree of risk to the key item; a learning part which learns relations between a plurality of answer data and a plurality of degrees of risk in a neural network; and a narrowing part which gives answer data including intentional answer data to the learnt neural network to determine a degree of risk, and narrows a question corresponding to the intentional answer data with high degree of risk as an important question; and a detection part which selects a project including answer data which increases the degree of risk of answer data to the important question, and performs risk prediction of a predetermined period with respect to the selected project. <P>COPYRIGHT: (C)2010,JPO&INPIT

Description

  The present invention relates to a computer system, a method, and a computer program for predicting risk in a plurality of projects having risk management.

One recent requirement for project management systems is the early detection of risk projects. As the scale of software increases, the risk countermeasures in the conventional project management system are problematic in that the timing of risk detection is delayed and the scale of influence becomes large.
The following Patent Document 1 describes a project management apparatus that performs objective and highly accurate evaluation and prediction that does not depend on the experience and subjectivity of a project manager. However, the project management apparatus has little room for incorporating qualitative judgment according to the project (see, for example, paragraph 0020).
Non-Patent Document 1 below describes a structured analogy prediction method for capturing potential risk factors as signs in consideration of the state and environment of a project. The structured analogy prediction method is composed of a method in which a qualitative questionnaire and a quantitative analysis are mixed. In other words, it is more risky to find a potential sign from a qualitative one that only compiles the questionnaire, or to find a sign from the quantitative data shown by the project. It has been reported that the capture probability can be improved. However, in the structured analogy prediction method, there are many qualitative analyzes compared with the part that quantitatively aggregates the questionnaire results from multiple experts, and the result is included in the first situation description. Is likely to be dragged (see, for example, 3.5 Derive forecasts lines 1 to 3).

JP 2006-79555 A K.C.Green and J.S. Armstrong, “Structured analogies for forecasting”, International Journal of Forecasting, pp.365-376, Vol.23, No.3, 2007

  Qualitative questionnaire responses are mixed with the qualitative analysis of quantitative analysis according to project characteristics for situation descriptions such as project questionnaire responses in order to detect signs of potential risks in the project. Therefore, there is a need for a method for appropriately processing the subjective blurring that occurs in the system.

The present invention provides a computer system for predictive detection of risks in a plurality of projects having risk management. The computer system
A collection unit that collects answer data for a plurality of questions related to each project, wherein each of the plurality of questions is classified into one of a plurality of key items;
Based on the answer data for each question for each key item, a risk degree giving unit that assigns a risk degree to each key item, wherein the answer data is a numeric value or a numeric value converted from answer data, A risk level assignment unit;
Answer data is input as input data of the neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. A learning unit that learns the relationship between multiple degrees,
A narrowing unit that narrows down the question corresponding to the artificial answer data when the obtained risk level is high by giving the answer data including artificial answer data to the learned neural network, and narrows down the question corresponding to the artificial answer data when the obtained risk degree is high ,
A detection unit that selects a project including response data that raises a risk level of response data for the important question, and detects a risk sign for a predetermined period for the selected project.

In one embodiment of the present invention, the detection unit is
If the answer data for each important question in the first period is higher in risk than the answer data for each important question in the second period immediately before the first period, the project includes answer data that raises the risk degree Select
With respect to the selected project, the answer data (hereinafter referred to as the first answer data) for each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for the respective questions for each key item in each period, the first answer data and the second answer in the period 1 to n A project whose total difference value for each key item of data is equal to or greater than a predetermined threshold is extracted as a project that may have a risk. Here, n is an integer of 2 or more.

  In one embodiment of the invention, the answer data is collected from at least two of the plurality of projects or from different periods of the same project.

  In one embodiment of the present invention, the risk degree assigning unit sums up the numerical values of the answer data for the questions, and assigns the risk degree based on the summed value. Alternatively, the risk degree assigning unit scores the numerical value of the answer data for each question using a nonlinear function, and assigns the risk degree based on the score. Giving the risk level is giving a color associated with the risk level to each key item. The color associated with the risk level is determined according to the sum of the numerical values of the answer data.

  In one embodiment of the invention, the neural network is a multilayer perceptron.

  In one embodiment of the present invention, in the neural network, the learning unit uses the input data as a first node, the output data as a second node, and the first node as a multi-layer number (n ) -1 to divide the first node into multiple layers and to associate the first node with the second node.

  In one embodiment of the present invention, the artificial answer data is answer data artificially given for one or two of the plurality of questions.

  In one embodiment of the present invention, selecting the project selects a project whose level value is the number of questions corresponding to the number of answer data that increases the risk level, and the level value is 1 or more. It is to be.

In one embodiment of the present invention, when the value of the level in the first period is 1,
If the total value is greater than or equal to the first threshold, the project is extracted as a project that may have a risk,
In the case where the value of the level in the first period is 2 or more,
When the total value is equal to or greater than the second threshold, the project is extracted as a project that may have a risk.

In one embodiment of the invention, the first and second thresholds are
(Decrease function with respect to the desired number of projects extracted from the selected projects that may have the risk) + (Increase function with respect to the value of the level)
It is requested from.

In one embodiment of the invention, the threshold is
7 x exp (-5 x / 100) + 0.5 x exp (2.5 / y)
It is. Here, x is a selection ratio (%) of a project selected from projects associated with the answer data, and y is a value of the above level.

The present invention also provides a method for predictive risk detection in multiple projects having risk management. The method includes causing a computer system to perform the following steps. The step is
Collecting answer data for a plurality of questions related to each project in a storage unit, wherein each of the plurality of questions is classified into any one of a plurality of key items; ,
A step of assigning a risk level to each key item based on answer data for each question for each key item, wherein the answer data is a numerical value or a numerical value converted from the answer data. When,
Answer data is input as input data of the neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. Learning the relationship with multiple degrees,
Applying the answer data including artificial answer data to the learned neural network to determine the risk level, and narrowing down the question corresponding to the artificial answer data when the calculated risk level is high as an important question;
Selecting a project including answer data that raises the risk level of answer data for the important question, and detecting a predictor of risk for a predetermined period for the selected project.

In one embodiment of the present invention, the step of performing the sign detection includes
If the answer data for each important question in the first period is higher in risk than the answer data for each important question in the second period immediately before the first period, the project includes answer data that raises the risk degree A step of selecting
With respect to the selected project, the answer data (hereinafter referred to as the first answer data) for each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for the respective questions for each key item in each period, the first answer data and the second answer in the period 1 to n A step of extracting a project in which a total difference of each key item of data is equal to or greater than a predetermined threshold as a project having a risk, wherein n is an integer of 2 or more, and includes the above-described extraction step .

In one embodiment of the invention, the method includes causing the computer system to further perform the following steps. The step is
Repeating the step of collecting the answer data,
Repeating the step of assigning the risk level;
Repeating the step of learning the relationship between the plurality of answer data and the plurality of risk degrees in the learned neural network;
Repeating the narrowing step to update the important question.

  In one embodiment of the present invention, in the step of performing the sign detection, the updated answer data for each important question in the first period is updated in the second period immediately before the first period. When the risk level is higher than the response data for the important question, the method further includes a step of selecting a project including the response data for increasing the risk level.

  In one embodiment of the present invention, the step of selecting the project with the number of questions corresponding to the number of answer data increasing the degree of risk as the level value selects a project whose level value is 1 or more. Including the steps of:

In one embodiment of the invention, the extracting step comprises
In the case where the value of the level in the first period is 1,
Extracting the project as a project that may have a risk when the total value is equal to or greater than a first threshold; and
In the case where the value of the level in the first period is 2 or more,
Extracting the project as a project that may have a risk when the total value is equal to or greater than a second threshold;
including.

In one embodiment of the invention, the method further comprises causing the computer system to further perform the following steps: In this step, the first and second threshold values are set as follows.
(Decrease function for the desired number of projects that could have the risk extracted from the selected project) + (Increase function for the value of the level)
Including the step of obtaining from

In one embodiment of the invention, the neural network is a multilayer perceptron,
In the neural network, the learning step includes setting the input data as a first node and the output data as a second node, dividing the first node into a multilayer number (n) −1, Associating between multiple layers of first nodes and between the first node and the second node.

  In one embodiment of the present invention, the step of assigning the risk level includes the step of summing up the numerical values of the answer data for the questions and assigning the risk level based on the summed value.

The present invention also provides a method for predictive risk detection in multiple projects having risk management. The method includes causing a computer system to perform the following steps. The step is
Collecting answer data for a plurality of questions related to each project in a storage unit, wherein each of the plurality of questions is classified into any one of a plurality of key items; ,
A step of assigning a risk level to each key item based on answer data for each question for each key item, wherein the answer data is a numerical value or a numerical value converted from the answer data. When,
Answer data is input as input data of the neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. Learning the relationship with multiple degrees,
Applying the answer data including artificial answer data to the learned neural network to determine the risk level, and narrowing down the question corresponding to the artificial answer data when the calculated risk level is high as an important question;
Repeating the step of collecting the answer data,
Repeating the step of assigning the risk level;
Repeating the step of learning the relationship between the plurality of answer data and the plurality of risk degrees in the learned neural network;
Repeating the refinement step to update the important question,
A step of detecting a predictor of risk in a project for a predetermined period,
When the response data for each updated important question in the first period has a higher risk level than the response data for each updated important question in the second period immediately before the first period, the risk level is Selecting a project that contains the response data
With respect to the selected project, the answer data (hereinafter referred to as the first answer data) for each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for the respective questions for each key item in each period, the first answer data and the second answer in the period 1 to n A step of extracting a project in which a total difference of each key item of data is equal to or greater than a predetermined threshold as a project having a risk, wherein n is an integer of 2 or more, and includes the above-described extraction step And a step of performing the above sign detection.

  The present invention further provides a computer program for detecting predictive risk in a plurality of projects having risk management. The computer program causes a computer system to execute each step of the method described in any one of the above.

  According to the embodiment of the present invention, it is possible to detect a project including a potential risk even from a qualitative answer to a project having a variation by a project manager.

  In the embodiment of the present invention, a “project” is a project having risk management. In the embodiment of the present invention, a plurality of projects are targeted. In the embodiment of the present invention, a project that may potentially have a risk is detected from among a plurality of projects.

  In the embodiment of the present invention, “risk management” refers to a process for systematically managing risk, avoiding the source or cause of harm, loss, etc., or reducing them. “Risk management” includes predictive detection of risk factors potentially present in the project.

  In the embodiment of the present invention, “predictive detection” refers to detecting a project including a risk that will occur in the future in the project.

  In the embodiment of the present invention, “a plurality of questions regarding a project” relates to a content that is considered to be important for a person skilled in the art of project management, for example, to empirically grasp the situation of the project. The question is, for example, a questionnaire format, and may be either a paper-based format or an electronic-based format. The content of the question may overlap between questions.

In the embodiment of the present invention, questions regarding a project are classified into a plurality of key items. The plurality of key items are composed of an arbitrary number such as 5, 7, 10, and the like. One key item includes at least one question. Key items are items that are provided from the viewpoint necessary for project risk management. Key items include, but are not limited to, stakeholder (stakeholder) items, business benefit items, work and schedule items, team items, scope items, risk items, and benefit items for service providing organizations.
A stakeholder item is an item related to whether a stakeholder is committed, and relates to identifying, evaluating, providing information, and inspiring individuals and groups that are affected by or affecting the project. Includes questions.
The business profit item is an item related to whether the business profit of the customer is realized, and includes questions related to predicting, measuring, and monitoring the profit that the project gives to the client company. Profit takes into account expected outcomes and associated costs in both financial and non-financial aspects.
Work and schedule items are items related to whether work and schedules are predictable, manage the provision and acceptance of project services and deliverables, and ensure that performance specifications and acceptance criteria are met. Includes questions about what to do.
The team item is an item relating to whether the project team is performing well, and includes questions relating to identifying, setting up and educating the personnel required for the project team.
The scope item is an item relating to whether the scope is realistic and properly managed, and includes questions related to approving and maintaining the work scope of the project.
A risk item is an item relating to whether the risk has been mitigated, and includes questions relating to identifying and evaluating risks and issues, and performing activities such as avoidance, mitigation, and resolution.
The benefit item for the service providing organization is an item relating to whether or not the benefit for the service providing organization has been realized, and the benefit provided by the service providing organization from the project is determined, agreed, and managed. Benefits for service providers are, for example, financial compensation, knowledge transfer, and skill development.

  In the embodiment of the present invention, “answer data for a question” is a numerical value or a numerical value converted from the answer data. If the answer data is not a numerical value, it is digitized according to a predetermined standard. Even if the answer data is a numerical value, the numerical value is digitized according to a predetermined standard. For digitization, for example, discretized scoring may be used. As the discretization scoring, for example, a three-stage discretization scoring can be used.

In one embodiment of the present invention, “answer data for a question” is obtained, for example, by the manager of each project selecting a choice of 3 or 5 choices. When the answer is in an option format, the obtained answer data is a numerical value. For example, in the case of 3 choices, the answer data is a numerical value of 1, 2 or 3, and in the case of 5 choices, the answer data is a numerical value of 1, 2, 3, 4 or 5. The manager of each project makes an answer by, for example, writing an answer on a mark sheet or checking options displayed on the screen of the application software. The answer data entered on the mark sheet is digitized by passing the mark sheet through a mark sheet reader.
In another embodiment of the present invention, the “answer data for a question” is a numerical value expressed as a percentage, a ratio, or a score. The numerical value is, for example, a project progress rate, a progress rate, or a score related to progress.
In another embodiment of the present invention, “answer data for a question” is a character string such as “yes” or “no”.

In another embodiment of the present invention, the “numerical value converted from the answer data” refers to a value obtained by converting the answer data into a numerical value when the answer data is the percentage, the ratio, or the score, or is a character string. .
When the answer data is a percentage, the percentage is converted into a numerical value such as 1, 2, 3, 4 or 5 for each predetermined range. For example, if the percentage is from 0% to less than 20%, it is converted to 1, if the percentage is from 20% to less than 40%, it is converted to 2, and if the percentage is from 40% to less than 60%, it is converted to 3. If the percentage is from 60% to less than 80%, it is converted to 4, and if the percentage is from 80% to 100%, it is converted to 5. In addition, it is good also considering the said converted numerical value as 1, 2, 4, 6, or 8 using discretization scoring, for example.
When the answer data is a ratio, the ratio is converted into a numerical value such as 1, 2, 3, 4 or 5 for each predetermined range. For example, if it is a ratio of 0% to less than 20%, it is converted to 1. If it is a ratio of 20% to less than 40%, it is converted to 2. If it is a ratio of 40% to less than 60%, it is converted to 3. If it is a ratio of 60% to less than 80%, it is converted to 4. If it is a ratio of 80% to 100%, it is converted to 5. In addition, it is good also considering the said converted numerical value as 1, 2, 4, 6 or 8 using discretization scoring, for example.
When the answer data is a score, the score is converted into a numerical value of 1, 2, 3, 4, or 5 for each predetermined range. For example, if the score is 0 or more and less than 20 points, the score is converted to 1. If the score is 20 or more and less than 40 points, the score is converted to 2. If the score is 40 or more and less than 60 points, the score is 3. If the score is from 60 points to less than 80 points, the score is converted to 4. If the score is from 80 points to 100 points, the score is converted to 5. In addition, it is good also considering the said converted numerical value as 1, 2, 4, 6 or 8 using discretization scoring, for example.
If the answer data is a character string such as "Yes" or "No", give a low value to the answer to "Yes" or give no value, and answer "Yes" to the answer to "No" A value higher than the assigned value is assigned. In this example, the higher the numerical value, the higher the risk of the project. For example, a score value of 0 is assigned to a “yes” answer (ie, no score value is assigned), while a score value of 3 is assigned to a “no” answer.

In the embodiment of the present invention, the “risk level” is a standard for the sound state of the project. For example, the higher the risk level, the greater the risk in the project. The level of risk level is indicated by a color associated with the risk level, for example. The colors are, for example, three colors of green, yellow, and red. The reference values of green, yellow, and red can be arbitrarily set for each key item.
Green indicates that the project is on track, for example that the project is progressing as planned. A good project means that the general situation of the project is good.
Yellow indicates a warning to the project and indicates that action is needed soon to improve the situation.
Red indicates that the project is urgent and requires immediate action.

In the embodiment of the present invention, “giving a risk level” means giving a risk level for each key item based on a numerical value that is answer data to a question for each key item of the project.
When the answer data is an option, the risk level may be given based on the numerical value of the option itself, but the risk level may be given based on a total value obtained by adding the numerical values for each key item.
When the response data is a percentage, the risk level may be assigned based on the numerical value of the percentage itself, but the percentage is converted into a numerical value for each predetermined range, and the risk level is assigned to the converted numerical value. May be. Further, the degree of risk may be assigned to each key item based on the total value obtained by summing up the numerical values of the percentages or the sum of the converted numerical values for each key item.
When the answer data is a ratio, the risk level may be assigned based on the numerical value of the ratio itself, but the ratio is converted into a numerical value for each predetermined range, and the risk level is assigned to the converted numerical value. May be. Further, the degree of risk may be assigned to each key item based on the total value obtained by summing up the numerical values of the proportions themselves or the sum of the converted numerical values.
When the answer data is a score, the risk level may be assigned based on the numerical value of the score itself. Furthermore, the score is converted into a numerical value for each predetermined range, and the risk level is assigned to the converted numerical value. May be. Further, for the numerical value of the score itself or the converted numerical value, the risk degree may be assigned for each key item based on the total value obtained by summing the numerical values of the score itself or the total value obtained by summing the converted numerical values.
When the answer data is a character string, the character string is converted into a numerical value, and a risk level is assigned to the converted numerical value.

  A specific method for assigning the risk level is, for example, as follows. When the value of the answer data for a specific question included in a certain key item is less than a predetermined threshold, the risk degree assigning unit (FIG. 1, 103 below) determines that the value of the answer data is bad, and the key including the specific question It is determined that there is a risk in the item status. And a risk degree provision part (103) provides yellow or red risk degree with respect to this key item. On the other hand, when the value of the answer data for all questions included in a certain key item is equal to or greater than a predetermined threshold, the risk level assigning unit (103) determines that the value of the answer data is good and includes the key item including all the questions. It is determined that there is no risk in the condition. The risk degree assigning unit (103) assigns a green risk degree to the key item.

In the embodiment of the present invention, the “neural network” is a mathematical model that aims to express some characteristics found in the brain function by simulation on a computer. In an embodiment of the present invention, a multilayer perceptron using an error back propagation method may be used. In the embodiment of the present invention, the “multilayer perceptron” is a network in which simple perceptrons are connected in layers, and includes an input layer and an output layer.
In the embodiment of the present invention, a “neural network” is created for each key item of a project. The input data of the neural network is answer data for each key item. The output data of the neural network is the degree of risk corresponding to the answer data for each key item.
In the embodiment of the present invention, the “neural network” learns the relationship between the input answer data for each key item and the risk level as the output data. In the embodiment of the present invention, the “neural network” is used to specify a question corresponding to input data that influences output data, that is, answer data after the learning.

  In the embodiment of the present invention, for example, a three-layer perceptron is used as the “multilayer perceptron”. The three-layer perceptron includes two input layers and one output layer.

In the embodiment of the present invention, “artificial answer data” means answer data that is artificially given to narrow down important questions that are important for predicting risk in a project from a plurality of questions. It is. The input of artificial answer data is to input answer data that intentionally improves or deteriorates only the answer corresponding to the question whose influence is to be investigated.
In one embodiment of the present invention, inputting “answer data for each question including artificial answer data” is, for example, a value that makes the answer data value for one question worse, that is, one question Enter a value that increases the risk of the answer to the question, and the value of the answer data for the other questions is good or does not change, that is, the value that does not increase the risk of the answer to the other question. input.
In another embodiment of the present invention, inputting “answer data for each question including artificial answer data” is, for example, a value that causes an output result to deteriorate only at one input node, that is, a risk level. A high value is input, and a value that improves or does not change the output result, that is, a value with a low risk level, is input to other input data.
In one embodiment of the present invention, inputting “answer data for each question including artificial answer data” is, for example, a value that makes the value of the answer data for each of the two questions worse, ie, two Enter the value that increases the risk of the answer to each question, and the value of the answer data for other questions is good or does not change, that is, the answer to other questions does not increase the risk Enter a value.
In another embodiment of the present invention, inputting “answer data for each question including artificial answer data” means that, for example, in all combinations of two input nodes, only one combination has a poor output result. And a value that improves the output result or does not change is input to the other input nodes.

  In the embodiment of the present invention, the “important question” is a question that is important in order to detect a sign of risk in a project from among a plurality of questions. In the actual form of the present invention, it is possible to effectively quantify qualitative answer data by narrowing down important questions.

  In the embodiment of the present invention, the “predetermined period” is, for example, a period in days, weeks, months, or years.

  Embodiments of the present invention will be described below with reference to the drawings. It should be understood that this embodiment is for the purpose of illustrating a preferred aspect of the present invention and is not intended to limit the scope of the invention to what is shown here. Further, throughout the following drawings, the same reference numerals refer to the same objects unless otherwise specified.

FIG. 1 is a configuration diagram of a computer system according to an embodiment of the present invention.
The computer system (101) includes a collection unit (102), a risk level assignment unit (103), a learning unit (104), a narrowing unit (105), a detection unit (106), and a central processing unit (hereinafter referred to as CPU). (107), a storage unit (108), and a memory (109). The computer system (101) also stores software that implements the neural network (110) in the storage unit (108). Alternatively, the computer system (101) is connected to another computer (not shown) that includes software that implements the neural network (110).
The collection unit (102) collects response data for a plurality of questions regarding the project in response to input from the project manager. The answer data is collected from multiple projects or from different periods of the same project. The answer data is digitized as necessary before being collected by the collection unit (102).
The risk degree assigning unit (103) assigns a risk degree for each key item based on answer data for each question for each key item of each project. In particular, the risk degree assigning unit (103) sums the values of the answer data for each key item, and assigns a risk degree to each key item based on the total value. The answer data is a numerical value or a numerical value converted from the answer data.
The learning unit (104) inputs answer data for each key item as input data of the neural network, inputs a risk degree corresponding to the answer data input as output data of the neural network, and for each key item in the neural network. Learn the relationship between the answer data entered by and the risk level. In the neural network, the learning unit (104) also sets the input data for each key item as the first node and the output data as the second node, and divides the first node into the number (n) −1 of the multilayer. , Between the first node multilayers, and between the first node and the second node.
The narrowing-down unit (105) gives the answer data for each question including artificial answer data to the learned neural network to obtain the risk level as output data, and the artificial answer when the obtained risk degree is high Narrow down the questions corresponding to the data as important questions.
The detection unit (106) detects a risk sign in each project for a predetermined period. The detection unit (106) also increases the answer data that the answer data for each important question in the first period is worse than the answer data for each important question in the second period immediately before the first period, that is, the degree of risk. Then, a project including answer data that increases the degree of risk is selected. The detection unit (106) further includes, for the selected project, response data for each question for each key item in each of the first to nth periods (where n is an integer of 2 or more). Answer data that is worse than the answer data for each question in each of the second to (n + 1) periods immediately before the period, that is, when the risk level is increased, the number of response data for increasing the risk level is set for each period of 1 to n. Summing up and extracting a project whose sum is equal to or greater than a predetermined threshold as a project that may have a risk.
The CPU (107) executes a computer program for detecting a risk sign that is an embodiment of the present invention.
The memory (109) is used to execute the computer program.

FIG. 2 shows an example of a list of key items in the embodiment of the present invention.
In the example, key items include stakeholders, business benefits, work and schedules, teams, scopes, risks, and benefits for service providers. The key item may be selected from any of these, or other key items may be added to the example key item.

  Each key item includes at least one question, and the number of questions is arbitrary. For example, 5 to 20 questions may be included for one key item.

FIG. 3A shows an example of learning of a neural network in the embodiment of the present invention.
The neural network (302) is a three-layer perceptron. As the neural network (302), one neural network is prepared for one key item. Therefore, when there are seven key items, for example, the number of prepared neural networks (302) is seven.
The input data of the neural network (302) is answer data for the question. Therefore, for example, the input data of the neural network whose key item is a stakeholder is answer data for a question classified as a stakeholder.
The output data of the neural network (302) is the degree of risk assigned based on the answer data. The degree of risk can be represented by a numerical value.
The risk degree is converted into a risk degree judgment color according to the risk degree value so that the project manager can easily grasp the risk degree. The determination color is represented by green, yellow, and red. The numerical value corresponding to the color of the risk degree determination color can be arbitrarily determined. For example, green is 1, yellow is 2, red is 3.
In order to learn the relationship between the answer data and the risk level, the combination data (301) of the response data and the risk level is input to the neural network (302) for each key item. The data of the combination (301) is data collected in the past before the learning. Data collected in the past is data for a predetermined period of each of a plurality of projects, for example, monthly data. For example, in a three-layer perceptron, the first and second layers of the three layers are input data, and the third layer is output data.
The neural network (302) learns the relationship between the answer data and the risk degree based on the combination of the answer data and the risk degree. The learning result is expressed as a relationship between input and output data.

FIG. 3B shows narrowing a question corresponding to artificial answer data with a high degree of risk as an important question using the learned neural network in the embodiment of the present invention.
In order to select an important question, answer data for each question including artificial answer data is input as input data of the learned neural network (303). As a result, the learned neural network (303) outputs the degree of risk as output data. The degree of risk is represented by a numerical value, for example. Further, the risk level can be displayed on the display device in three stages of colors corresponding to the risk level, for example, green, yellow, or red, in order to make it easy for the manager of the project to visually recognize.
Artificial answer data is answer data that is artificially given to narrow down important questions that are important for predicting the risk of a project from a plurality of questions. In the example of FIG. 3B, a combination of answer data for each question including the following two types of artificial answer data is prepared.
The combination of the first answer data is a value that makes the answer data value for one question worse, and a value that makes the answer data values for other questions better or does not change. Answer data.
The combination of the second answer data is the answer data that makes the answer data values for each of the two questions worse, and the answer data values for the other questions are improved or do not change. It is some answer data.
As a result, the output data is output in descending order (304). The output data in FIG. 3B is an example of output data for stakeholders.
In the output data (304), since the output data for the first answer data combination did not have a large value, a part of the output data for the second answer data combination is shown in descending order. The combination of questions (Q) in the output data (304) indicates a combination of artificial response data in the second combination of response data. For example, (Q1, Q3) indicates Question 1 and Question 3 corresponding to a combination of artificial response data. In the example of FIG. 3B, the top five are combinations of important questions. Thus, the important questions are Q1, Q3, Q2, Q4, Q6, and Q8.

である場合、
又は
リスク度レベルｋが２以上であり、且つ

である場合、
プロジェクトＸは、第２基準リスクプロジェクトであるとして抽出される。
（ここで、リスク度レベルｋは、プロジェクトＸにおいて、リスク度を上げる回答データの数に対応する重要質問の数である。

は、キー項目毎の差分（Ｓｉ，ｔ）の総差分値である。但し、差分は正の整数の場合である。差分が０又は負の場合、差分は０である。
７は、キー項目の数が７であることを示す。
Ｓi,tは、キー項目毎の今月（ｔ）と前月（ｔ−１）の差分を示す。
ｉは、キー項目を示す。）
次に、閾値は、下記の基準関数の考え方を用いて下記式により決定されうる。
閾値は、
閾値＝（リスクを有しうる第２基準リスクプロジェクトの第１基準リスクプロジェクトから抽出される希望数に関する減少関数）＋（上記リスク度レベルの値に関する増加関数）
で求められる。リスク度レベルは重要質問の下降該当数を表しているので、リスク度レベルだけでも、第１基準リスクプロジェクトのうちから第２基準リスクプロジェクトを抽出する効果がある。
上記式により、最終的に選択する第２基準リスクプロジェクトの数を少なくする場合には閾値は大きくなり、一方、最終的に選択する第２基準リスクプロジェクトの数を多くする場合には閾値は小さくなる。
場合によっては、閾値の最適値を求めることも可能である。しかしながら、一般に、最終的な第２基準リスクプロジェクトの数はおおよそ決まっているので、リスク度レベルが小さいときの閾値は大きく、リスク度レベルが大きいときの閾値は小さくなる。例えば、リスク度レベルが１であるときの閾値は１０であり、リスク度レベルが２以上であるときの閾値は５である。
様々な基準関数が、閾値を求めるために用いられうる。実験的には、例えば、下記の指数関数によって閾値を決定することが分かっている。
閾値が、７×exp（-5ｘ/100）＋0.5×exp（2.5/ｙ）で表される。
ここで、ｘは、上記回答に関連付けられたプロジェクトから選択するプロジェクトの選択比率（％）であり、ｙは、上記レベルの値である。
例えば、第２基準リスクプロジェクトが選択されうる希望数を第１基準リスクプロジェクトの１０％に絞り込む場合、閾値はリスク度レベルが１の場合１０．３であり、リスク度レベルが２の場合５．９であり、それ以上のリスク度レベルでは飽和する。よって、上記した選択する式において、リスク度レベルが１の場合の閾値が１０であり、リスク度レベルが２以上の場合の閾値が５である。
ステップ４０７では、リスクを有しうるプロジェクトを判定する。
過去数ヶ月分の上記総差分値の合計値が閾値以上であれば、該プロジェクトは第２基準リスクプロジェクトであるとして抽出される。該第２基準リスクプロジェクトが、予兆検知を行いたい期間において潜在的なリスクを含みうるプロジェクトである。該潜在的なリスクを含みうるプロジェクトが、最終的な予兆検知の結果である。 FIG. 4 shows a flowchart for detecting a risk sign in a project in the embodiment of the present invention.
In step 401, the collection unit (FIG. 1, 102) collects answer data for a plurality of questions regarding each project. Answers to questions are mainly given by the project manager of each project. The project manager answers the current situation for the question. Response data is collected for each project for multiple projects. In addition, for the same project, response data is collected every predetermined period, for example, every month. The collection unit (102) stores the collected answer data in the storage unit (FIG. 1, 108). The answer data is a numerical value or a numerical value converted from the answer data.
In step 402, the risk level assignment unit (FIG. 1, 103) assigns a risk level for each key item based on the answer data for each question for each key item of each project.
In step 403, the learning unit (FIG. 1, 104) inputs the combination of the answer data and the risk level to the neural network. The answer data is input data of the neural network. The risk degree is output data of the neural network. The learning unit (104) causes the neural network to learn the relationship between the input data and the output data, that is, the causal relationship between the input data and the output data. The reason why the neural network learns the relationship between the input data and the output data is that the variation of the response data by the project manager is considered as noise, and after the above learning, it corresponds to the input data that affects the output data, that is, the response data This is to identify the question. Response data variability is, for example, response data variability derived from the manager when the project manager is changed, or replies derived from changes in the content when the project content is changed Variation.
In step 404, questions corresponding to artificial answer data with a high degree of risk are narrowed down as important questions.
Questions are narrowed down using a learned neural network. The narrowing-down unit (FIGS. 1 and 105) inputs the answer data for each question including the answer data that is artificial in the learned neural network. As a result, the narrowing-down unit (105) outputs a risk degree that is output data using the learned neural network. The narrowing-down unit (105) narrows down the question corresponding to the answer data when the output result has a high degree of risk as an important question. Output data that increases the risk level is, for example, when the output data is greater than or equal to a predetermined value (assuming that the larger the value, the greater the risk), or the risk level is expressed in red or yellow Is the case.
When using the learned neural network to detect potential risks in the project, the impact of this narrowed-down important question on the output data is significant. The narrowing down makes it possible to narrow down to quantitative important questions according to the project based on qualitative answer data from a questionnaire. The narrowing down of important questions enables dimension compression of answer data (high-dimensional answer data space) and removal of noise, which is a variation in answer data by a project manager. “Noise” means response data that results in different answers between the respondent A who is the project manager and the respondent B who is the project manager, even if the progress is objectively the same. .
In step 405, the tendency of the important question is determined.
The detection unit (106) increases the risk level when the answer data for each important question in the first period increases the risk level than the answer data for each important question in the second period immediately before the first period. Select the project that contains the response data.
The first period is a period immediately before a period when it is desired to detect a risk sign in the project, and the second period is a period immediately before the first period. For example, if the period in which the sign of risk in the project is desired to be detected is January 2008, the first period is December 2007, and the second period is November 2007.
The detection unit (106) sets the number of important questions corresponding to the number of answer data to increase the risk level as a risk level. For example, if the risk level is 1 or more, the detection unit (106) has increased the risk level of the answer data for the important question. Therefore, the project corresponding to the response data with the increased risk level is selected as the first standard. Select as a risk project. A risk level of 1 or more is a case where the risk level is increased even for one of the answers to an important question. However, if the answer to the important question in the first period is lower than the answer to the important question in the second period or has not changed, the response data with an increased risk is obtained by the discretization score. Only the corresponding project is considered, and the project corresponding to the response data with a reduced risk level and no change is not considered as the first standard risk project. Because, according to the first standard, a better score may be temporary, and conversely, a worse score may be temporary, so it covers a little risky project Because. Therefore, the project is further selected based on the second standard.
As described above, selecting the first reference risk project based on the key questions allows the project manager to change between the same projects, taking into account the subjective impression of the answers to the project manager's questions. It becomes possible to suppress the variation in the response when the project manager becomes different or the project manager differs between projects.
In step 406, the history of the project is determined.
The detection unit (106) calculates the score difference (deteriorating value) of each key item from the first reference risk project. Then, it is examined whether the total value of the score differences has increased during the past several months, for example, three months. Specifically, the detection unit (106) accumulates and adds a score difference on a monthly basis for the scores for the past several months to obtain a total difference value. Next, a total value of total difference values for the past several months, for example, three months is obtained. Then, a second reference risk project is further extracted from the first reference risk projects based on the total value of the total difference values using a classification rule using a threshold based on the concept of the reference function.
The formula for selecting the second standard risk project from the first standard risk project is as follows.
Consider project X in t month (where t is an integer from 1 to 12). The formula to be selected differs depending on the risk level is 1 or 2 or more.
・ When the risk level is 1 and the difference between the score for each key item in the previous month's answer data and the score for each key item in the current month's answer data is expressed as a positive integer (however, the higher the score, the more risk When the degree is high), the difference is accumulated and added every month to obtain a total difference value. Alternatively, the risk level is 1 and the difference between the previous month's response data score and the current month's response data score is expressed as a positive integer (however, the higher the score, the higher the risk level) The total difference value is obtained by accumulating and adding the difference every month. If the total value of the total difference values for the past several months is equal to or greater than the threshold value, the project X is extracted as a second reference risk project. The threshold is 10, for example.
・ When the risk level is 2 or more and the difference between the score for each key item in the previous month's answer data and the score for each key item in the current month's answer data is expressed as a positive integer (however, the higher the score is) When the degree of risk is high), the difference is accumulated and added every month to obtain a total difference value. Alternatively, when the risk level is 2 or more and the difference between the previous month's response data score and the current month's response data score is expressed as a positive integer (however, the higher the score, the higher the risk level) The total difference value is obtained by accumulating and adding the differences every month. If the total value of the total difference values for the past several months is equal to or greater than the threshold value, the project X is extracted as a second reference risk project. The threshold is 5, for example.
The above formula is expressed as follows.
About Project X
The risk level k is 1, and

If it is,
Or the risk level k is 2 or more, and

If it is,
Project X is extracted as being the second reference risk project.
(Here, the risk level k is the number of important questions corresponding to the number of answer data for increasing the risk level in the project X).

Is a total difference value of differences (Si, t) for each key item. However, the difference is a positive integer. If the difference is 0 or negative, the difference is 0.
7 indicates that the number of key items is seven.
Si, t indicates the difference between the current month (t) and the previous month (t-1) for each key item.
i indicates a key item. )
Next, the threshold value can be determined by the following equation using the concept of the following reference function.
The threshold is
Threshold = (decreasing function related to the desired number extracted from the first reference risk project of the second reference risk project that may have risk) + (increasing function related to the value of the risk level)
Is required. Since the risk level represents the number of falling important questions, the risk level alone has the effect of extracting the second reference risk project from the first reference risk project.
According to the above formula, the threshold value increases when the number of second standard risk projects to be finally selected is small, whereas the threshold value decreases when the number of second standard risk projects to be finally selected is large. Become.
In some cases, it is also possible to obtain an optimum threshold value. However, in general, since the final number of the second reference risk projects is roughly determined, the threshold when the risk level is small is large, and the threshold when the risk level is large is small. For example, the threshold when the risk level is 1 is 10, and the threshold when the risk level is 2 or more is 5.
Various criteria functions can be used to determine the threshold. Experimentally, for example, it has been found that the threshold value is determined by the following exponential function.
The threshold value is represented by 7 × exp (−5 × / 100) + 0.5 × exp (2.5 / y).
Here, x is a selection ratio (%) of a project selected from the projects associated with the answer, and y is a value of the above level.
For example, when narrowing down the desired number that the second standard risk project can be selected to 10% of the first standard risk project, the threshold is 10.3 when the risk level is 1, and when the risk level is 2. 9 and saturates at higher risk levels. Therefore, in the formula to be selected, the threshold when the risk level is 1 is 10, and the threshold when the risk level is 2 or more is 5.
In step 407, a project that may have a risk is determined.
If the total value of the total difference values for the past several months is equal to or greater than the threshold, the project is extracted as being the second reference risk project. The second reference risk project is a project that may contain potential risks in the period in which the sign detection is desired. Projects that may contain the potential risk are the result of eventual sign detection.

FIG. 5 shows a specific example of steps 405, 406 and 407 in FIG.
FIG. 5 shows two projects AB-123 (501) and AB-456 (502).
Step 406
Assume that the period for which a risk sign in the project is to be detected is, for example, January 2008. Then, the period immediately before the period is December 2007.
In Project AB-123 (501), there was answer data that raised the risk level for important questions (not related to key items) in December 2007 (results not shown in Fig. 5). Is 1 (indicated by an arrow). Therefore, project AB-123 (501) is selected as the first reference risk project.
Similarly, in Project AB-456 (502), there was answer data that raised the risk level for important questions (not related to key items) in December 2007 (results not shown in FIG. 5). Risk level is 1 (indicated by arrows). Therefore, project AB-456 (502) is selected as the first reference risk project.
Step 407 and step 408
Difference 1, difference 2, difference 3, difference 4, difference 5, difference 6, and difference 7 are differences from the previous month of each key item shown in FIG. The difference from the previous month for each key item is the sum of the differences from the previous month of the answer data for each question included in each key item.
For example, the value of the difference 1 in September 2007 of the project AB-123 (501) is 1. Therefore, the total difference between the answer data for each question included in the key item “stakeholder” in September 2007 and the answer data for each question included in the key item “stakeholder” in August 2007, which is the previous month, is 1 is shown. Moreover, the difference 1, difference 2, difference 3, difference 4, difference 5, difference 6, and difference 7 in September 2007 of Project AB-123 (501) are 0, 2, 1, 2, 2, 0, respectively. And 0, the total difference value is 7.
In the project AB-123 (501), the total difference value for each key item during the past three months (October, November, December, 2007) is 7, 1 and 2, respectively. Therefore, the total value of the total difference values for the past three months of project AB-123 (501) is 10. In Project AB-123 (501), as described above, the risk level is 1 in the previous month (December 2007) of the predictive month (January 2008), so it is used as the second standard risk project. The threshold value is 10. Therefore, the total value 10 of the total difference values in the past three months is the same as the threshold value 10. Therefore, the project AB-123 (501) is determined to be the second reference risk project.
In the project AB-456 (502), the total difference value for each key item during the past three months (October, November, December, 2007) is 1, 0, and 1, respectively. Therefore, the total value of the total difference values for the past three months of the project AB-456 (502) is 2. In Project AB-456 (502), as described above, the risk level is 1 in the previous month (December 2007) of the predictive month (January 2008), so it is used as the second reference risk project. The threshold value is 10. Therefore, the total value 2 of the total difference values in the past three months is less than the threshold value 10. Accordingly, project AB-456 (502) is not determined to be the second reference risk project.

Example 1
In accordance with an embodiment of the present invention, risk detection in a project was detected using existing data. The result of having detected the predictive detection of a certain month by making a neural network learn the relational data with the plurality of reply data of the month which consists of a certain range, and the plurality of risk degrees is shown.
According to the method according to FIG. 4, the number of second reference risk projects was 16% of the total. Therefore, the reduction rate for risk projects that should be focused on the possibility of worsening the next month is 84%. The list focused on 16% included projects whose risk level for the next month was actually yellow. Therefore, the sign detection of the second standard risk project showed 100% coverage.
In order to calculate the effect of the above-described embodiment of the present invention, a harmonic average (= 2 / ((1 / reduction rate) + (1 / coverage rate)) of the reduction rate and the coverage rate) is taken. In this case, it is 91.3%.
Note that it is difficult to accurately know the effect of the above embodiment of the present invention from the so-called hit rate. Because there are cases where the project manager is intentionally taking risks, the key item score is worse, or the project manager is aware that there is a potential risk, There is a possibility that there was no drop in the data (difference between the previous month and this month) by taking action by the time of reporting the answer to the question the next month, for example, removing the cause of the deterioration of the score Because. The intention of the project manager to take risks is, for example, that the answer data is temporarily deliberately dropped for the purpose of the final project success. It is to increase. The key item in this case is, for example, a key item related to profits or a key item related to teams.
Therefore, the effect of the present invention is calculated based on the harmonious average of the two in a trade-off relationship, such as how many percent of the target projects that are several hundreds are reduced and what percent of the cover ratio can be obtained.
Similarly, the results of harmonic averages for other months are shown below.
a Month Forecast 93.6%
b Monthly forecast 92.9%
c Month forecast 92.6%
d Month forecast 70.7%
As described above, results other than d month showed a result of 90% or more. In month d, it is added that the project was affected by the fact that the answer data for the questionnaire was inappropriate and that it started to make an accurate report from the current month.
Conventionally, predicting a potentially risky project from data measured by project management (for example, progress) has not been done quantitatively and there is no clear indicator. However, as described above, it is possible to extract the second reference risk project that potentially has the risk of the next month almost instantaneously from only the questionnaire result. Therefore, it is considered that the effect that the action for the degree of risk can be quickly applied to the second reference project having the above risk is extremely large.
As described above, according to the embodiment of the present invention, it is possible to detect a sign including a potential risk even from qualitative answer data that varies depending on the project manager.

FIG. 6 shows a block diagram of computer hardware in an embodiment of the present invention.
The computer system (601) of the above embodiment includes a CPU (603) and a main memory (602), which are connected to a bus (604). The CPU (603) is preferably based on a 32-bit or 64-bit architecture, such as Intel's Xeon (TM) series, Core (TM) series, ATOM (TM) series, Pentium (TM) series, The Celeron (TM) series, the AMD Phenom (TM) series, the Athlon (TM) series, the Turion (TM) series and the Empron (TM) can be used. A display (606) such as an LCD monitor is connected to the bus (604) via a display controller (605). The display (606) displays information about a computer connected to a network via a communication line and information about software running on the computer with an appropriate graphic interface for management of the computer system. Used to display. The bus (604) is also connected to a hard disk or silicon disk (608) and a CD-ROM, DVD drive or BD drive (609) via an IDE or SATA controller (607).

  The hard disk stores an operating system, a computer program according to an embodiment of the present invention, and other programs and data so that they can be loaded into the main memory.

  The CD-ROM, DVD or BD drive (609) is used for additionally introducing a program from the CD-ROM, DVD-ROM or BD to the hard disk as necessary. Further, a keyboard (611) and a mouse (612) are connected to the bus (604) via a keyboard / mouse controller (610).

  The communication interface (614) conforms to, for example, the Ethernet (trademark) protocol, is connected to the bus (604) via the communication controller (613), and plays a role of physically connecting the computer and the communication line (615). A network interface layer is provided for the TCP / IP communication protocol of the communication function of the computer operating system. The communication line may be a wired LAN environment or a wireless LAN environment based on a wireless LAN connection standard such as IEEE802.11a / b / g / n.

  As described above, the present invention has been described based on the embodiment. However, the content described in the embodiment is an example of the present invention, and those skilled in the art will be able to use various methods without departing from the technical scope of the present invention. It will be clear that variations can be conceived

1 shows a configuration diagram of a computer system in an embodiment of the present invention. FIG. The example of the list of key items in the embodiment of the present invention is shown. The example of learning of a neural network in embodiment of this invention is shown. In the embodiment of the present invention, using a learned neural network, a question corresponding to artificial answer data with a high degree of risk is narrowed down as an important question. The flowchart for performing the predictive detection of the risk in the project in embodiment of this invention is shown. Specific examples of steps 405, 406 and 407 in FIG. 4 are shown. FIG. 2 shows a block diagram of computer hardware in an embodiment of the present invention.

Claims (22)

A computer system for predictive detection of risks in a plurality of projects having risk management,
A collection unit that collects response data for a plurality of questions related to each project, wherein each of the plurality of questions is classified into any one of a plurality of key items;
Based on the answer data for each question for each key item, a risk degree giving unit that assigns a risk degree to each key item, wherein the answer data is a numerical value or a numerical value converted from answer data, A risk level assignment unit;
Answer data is input as input data of a neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. A learning unit that learns the relationship between multiple degrees,
A narrowing unit that narrows down the question corresponding to the artificial answer data when the obtained risk level is high by giving the answer data including artificial answer data to the learned neural network, and narrows down the question corresponding to the artificial answer data as the important question ,
The computer system comprising: a project including response data that raises a risk level of response data for the important question, and a detection unit that detects a predictor of risk for a predetermined period for the selected project. The detection unit is
When the response data for each important question in the first period has a higher risk level than the response data for each important question in the second period immediately before the first period, the project includes the response data for increasing the risk level Select
For the selected project, the answer data (hereinafter referred to as the first answer data) for each key item in each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for each question for each key item in each period, the first answer data and the second answer in the period 1 to n Projects in which the total value of differences for each key item of data is equal to or greater than a predetermined threshold are extracted as projects that may have a risk, where n is an integer of 2 or more.
The computer system according to claim 1.   The computer system according to claim 1, wherein the artificial answer data is answer data artificially given for one or two of the plurality of questions.   The selection of the project is to select a project having a level value of the number of questions corresponding to the number of answer data for increasing the risk level and having a level value of 1 or more. The computer system described. In the case where the value of the level in the first period is 1,
If the total value is greater than or equal to a first threshold, the project is extracted as a project that may have a risk;
In the case where the value of the level in the first period is 2 or more,
If the total value is greater than or equal to a second threshold, the project is extracted as a project that may have a risk;
The computer system according to claim 4. The first and second thresholds are
(Decrease function with respect to the desired number of projects extracted from the selected project that may have the risk) + (Increase function with respect to the value of the level)
The computer system according to claim 5, obtained from: The threshold is
7 x exp (-5 x / 100) + 0.5 x exp (2.5 / y)
And
Here, x is a selection ratio (%) of a project selected from the projects associated with the answer data,
y is the value of the level,
The computer system according to claim 6.   The computer system of claim 1, wherein the neural network is a multilayer perceptron.   In the neural network, the learning unit sets the input data as a first node and the output data as a second node, and divides the first node into a number (n) −1 of multiple layers, 9. The computer system of claim 8, wherein the computer system associates between multiple layers of one node and between the first node and the second node.   The computer system of claim 1, wherein the answer data is collected from at least two of the plurality of projects or from different periods of the same project.   The computer system according to claim 1, wherein the risk degree assigning unit sums up numerical values of answer data for the questions and assigns the risk degree based on the summed value. A method for predicting risk in a plurality of projects having risk management, wherein the computer system executes the following steps:
Collecting answer data for a plurality of questions relating to each project in a storage unit, wherein each of the plurality of questions is classified into one of a plurality of key items; ,
A step of assigning a risk degree to each key item based on answer data for each question for each key item, wherein the answer data is a numerical value or a numerical value converted from the answer data. When,
Answer data is input as input data of a neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. Learning the relationship with multiple degrees,
Applying the answer data including artificial answer data to the learned neural network to determine the risk level, and narrowing down the question corresponding to the artificial answer data when the calculated risk level is high as an important question;
Selecting a project that includes answer data that increases the risk level of answer data for the important question, and detecting a predictor of risk for a predetermined period for the selected project. The step of performing the sign detection comprises
When the response data for each important question in the first period has a higher risk level than the response data for each important question in the second period immediately before the first period, the project includes the response data for increasing the risk level A step of selecting
For the selected project, the answer data (hereinafter referred to as the first answer data) for each key item in each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for each question for each key item in each period, the first answer data and the second answer in the period 1 to n Extracting a project having a total difference value for each key item of data that is equal to or larger than a predetermined threshold as a project having a risk, wherein n is an integer equal to or larger than 2. The method according to claim 12. Repeating the step of collecting the answer data;
Repeating the step of assigning the degree of risk;
Repeating the step of learning the relationship between the plurality of answer data and the plurality of risk degrees in the learned neural network;
The method of claim 12, further comprising: repeating the narrowing step to update the key question.   In the step of performing the sign detection, the response data for each of the updated important questions in the first period is more risky than the answer data for each of the updated important questions in the second period immediately before the first period. The method according to claim 14, further comprising selecting a project including answer data that increases the degree of risk.   The number of questions corresponding to the number of answer data for increasing the risk level is set as a level value, and the step of selecting the project includes the step of selecting a project having the level value of 1 or more. The method described. Said extracting step comprises:
In the case where the value of the level in the first period is 1,
Extracting the project as a risky project if the total value is greater than or equal to a first threshold;
In the case where the value of the level in the first period is 2 or more,
Extracting the project as a risky project if the total value is greater than or equal to a second threshold;
The method of claim 16 comprising: The first and second thresholds are
(Decrease function for the desired number of projects that may have the risk extracted from the selected project) + (Increase function for the level value)
The method of claim 17, obtained from The neural network is a multilayer perceptron;
In the neural network, the learning step includes the input data as a first node and the output data as a second node, and the first node is divided into a number (n) −1 of multilayers, 13. The method of claim 12, comprising associating between multiple layers of first nodes and between the first node and the second node.   The method according to claim 12, wherein the step of assigning the risk level includes a step of summing up numerical values of answer data for the respective questions and assigning the risk level based on the summed value. A method for predicting risk in a plurality of projects having risk management, wherein the computer system executes the following steps:
Collecting answer data for a plurality of questions relating to each project in a storage unit, wherein each of the plurality of questions is classified into one of a plurality of key items; ,
A step of assigning a risk degree to each key item based on answer data for each question for each key item, wherein the answer data is a numerical value or a numerical value converted from the answer data. When,
Answer data is input as input data of a neural network for each key item, and a risk level corresponding to the input answer data is input as output data of the neural network. In the neural network, a plurality of the answer data and the risk are input. Learning the relationship with multiple degrees,
Applying the answer data including artificial answer data to the learned neural network to determine the risk level, and narrowing down the question corresponding to the artificial answer data when the calculated risk level is high as an important question;
Repeating the step of collecting the answer data;
Repeating the step of assigning the degree of risk;
Repeating the step of learning the relationship between the plurality of answer data and the plurality of risk degrees in the learned neural network;
Updating the important question by repeating the narrowing step;
A step of detecting a predictor of risk in a project for a predetermined period,
When the answer data for each updated important question in the first period has a higher risk level than the answer data for each updated important question in the second period immediately before the first period, the risk level is Selecting a project that contains the response data
For the selected project, the answer data (hereinafter referred to as the first answer data) for each key item in each of the key items in the first to n periods is the second to n + 1 immediately before the first to n periods. When the risk level is higher than the answer data (hereinafter referred to as second answer data) for each question for each key item in each period, the first answer data and the second answer in the period 1 to n Extracting a project having a total difference value for each key item of data that is equal to or larger than a predetermined threshold as a project having a risk, wherein n is an integer equal to or larger than 2. Performing the sign detection.   A computer program for performing predictive detection of risk in a plurality of projects having risk management, wherein the computer system executes each step of the method according to any one of claims 12 to 21. Computer program.

Images