Project success / failure prediction device, machine learning method for predictive model, and project success / failure prediction method

Abstract

To provide a technology for predicting a project success or failure by using determination materials more multifaceted than ever before.SOLUTION: A project success / failure prediction device includes a success / failure prediction unit for predicting a success / failure of a project by using a prediction model generated by machine learning, and a project information processing unit for generating a matrix at a prediction from project information of the project to a prediction execution time point, inputs an explanatory variable based on the matrix at a prediction to the success / failure prediction unit, and predicts a project success / failure by an objective variable to be outputted The project information includes a management index and a document. In the project success / failure prediction device, the project information processing unit includes a numerical index processing unit for converting the management index into the matrix at a prediction, and a natural language processing unit for converting the document into the matrix at a prediction.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a technique for predicting the success or failure of a project, for example, a technique for predicting the success or failure of a software development project. [Background technology]

[0002] The success or failure of a software development project (PJ) is judged by factors such as delays in the process, insufficient quality, and cost overruns at the end of the project. To prevent project failure, the project manager (PM) judges whether there are any signs of failure in the project based on various management indicators such as the amount of project deliverables and defects, as well as the situation on-site. It is necessary to detect such signs as early as possible and use them to discover and address the problems. PMs gain knowledge on discovering signs of failure by understanding and learning in detail what caused the failure of past projects they have been in charge of.

[0003] In organizations where many projects are ongoing simultaneously, there are often teams such as Quality Assurance (QA) and Project Management Office (PMO) that oversee multiple projects and are responsible for risk management and quality control. Unlike the PM, these teams are responsible for managing multiple projects across the board. By studying reports from the PM when a project fails and the management indicators of each project, you can gain insight into how to detect signs of failure.

[0004] In this way, PMs, PMOs, and QAs (collectively known as project managers) learn empirically from multiple failed projects they have worked on in the past about the causal relationships between indicators and situations that appeared beforehand and project failures. If similar indicators and situations appear in a new project, they will detect some kind of abnormality in the project as a sign of failure.

[0005] However, humans can only learn causal relationships empirically from a few dozen projects out of the many past projects that exist. In addition, there is a limit to the combinations of indicators and situations that humans can consider. If the projects and indicators to learn from are limited, the understanding of the project situation will be insufficient, and signs of failure will be overlooked. A supervised machine learning approach is considered to be suitable for deriving causal relationships by considering all the numerous combinations of indicators and situations for the large number of projects accumulated in the past.

[0006] Supervised machine learning is a method of determining the causal relationship between explanatory variables and objective variables from a large amount of data set in which the explanatory variables and objective variables are known. As an example, if we assume that the objective variable is calculated by multiplying the explanatory variables by a vector of any dimension, then that vector will represent the causal relationship between the input and output. This vector is called a weight vector (prediction model). The weight vector is adjusted so that it can best explain the relationship between all known input and output values. The predictive model obtained in this way is multiplied by an input whose output is unknown to calculate the output. The step of creating a predictive model is called the learning phase in machine learning, and the step of calculating the output from the input based on the predictive model is called the prediction phase.

[0007] Based on the principles of machine learning, it is possible to predict the final success or failure of an ongoing project using project management indicators at the completion of any stage. Specifically, this is achieved by performing machine learning using the success or failure of a large number of past projects as the objective variable and the project management indicators at the completion of the same stages as above as the explanatory variables. This eliminates the dependency on personal skills when discovering signs of failure, and is expected to be effective in discovering signs of failure that would be difficult for humans to find.

[0008] One related technology for achieving this is Patent Document 1. The risk assessment device in Patent Document 1 performs machine learning using multiple management indicators at the completion of any process of multiple past projects as explanatory variables and the success or failure of each project as a target variable to generate a prediction model. Using the generated prediction model, it outputs the success or failure of an ongoing project at the same time.

[0009] Using this system, project managers can determine that there is some kind of abnormality in a project that is predicted to fail, and by gaining a more detailed understanding of the project's situation, they can identify smaller-scale problems and prevent project failure from occurring.

[0010] Furthermore, Patent Document 2 shows that projects likely to experience problems can be narrowed down by estimating the probability and timing of trouble occurrence based on the probability distribution of project trouble occurrence based on the time point at which an expression pattern describing project information appears in the text of the project information, and on the statistics that represent the distribution. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] JP 2014-235654 A [Patent Document 2] JP 2009-230351 A Summary of the Invention [Problem to be solved by the invention]

[0012] Project management indicators are one of the judgment materials that PMs have used empirically to detect signs. Examples of project management indicators include numerical indicators such as the amount of defects and the amount of deliverables. In Patent Document 1, failure prediction is performed using machine learning with only the project management indicators as explanatory variables. In this way, indicators expressed as numerical values ​​can be directly input as explanatory variables, making it easy to apply a machine learning approach.

[0013] However, the PM's empirical judgment criteria for detecting signs include many documents accumulated during the project management process. For example, emails between responsible parties, regular reports, meeting materials, and lists of concerns. Such project management documents may contain written descriptions of on-site conditions that are difficult to express with numerical indicators. Therefore, a machine learning approach that uses only project management indicators does not provide enough information to make judgments compared to the PM's empirical judgment criteria for predicting success or failure and discovering signs of failure.

[0014] The system in Patent Document 2 statistically analyzes the frequency of occurrence of risk natural language pattern expressions in past project reports and the number of troubles that occurred in the projects, and quantifies the risks present in the reports. This makes it possible to use new project reports to estimate the probability and timing of troubles, narrowing down projects that are likely to become troubled with a higher likelihood than was previously possible. Patent Document 1 uses only project management indicators as a basis for judgment, while Patent Document 2 uses only project management documents as a basis for judgment.

[0015] Thus, there are existing technologies for predicting success or failure and detecting signs of failure that use either project management indicators or project management documents as decision-making materials, but there are no existing technologies that use both as decision-making materials.Compared to the empirical discovery of signs of failure by the PM, the existing technologies have an issue in that they use approaches based on machine learning and statistical methods with insufficient decision-making materials.

[0016] An object of the present invention is to provide a technique for predicting the success or failure of a project by utilizing more multifaceted decision-making information than conventional techniques. [Means for solving the problem]

[0017] A preferred aspect of the present invention is a project success / failure prediction device comprising: a success / failure prediction unit that predicts the success or failure of a project using a prediction model generated by machine learning; and a project information processing unit that generates prediction-time metrics from project information of the project up to the time of prediction execution, wherein explanatory variables based on the prediction-time metrics are input to the success / failure prediction unit and the project success or failure is predicted based on an output objective variable, the project information including management indicators and documents, and the project information processing unit comprises a numerical indicator processing unit that converts the management indicators into the prediction-time metrics and a natural language processing unit that converts the documents into the prediction-time metrics.

[0018] Another preferred aspect of the present invention is a machine learning method for a predictive model that predicts the success or failure of a project, comprising: a project information processing process that generates learning-time metrics from project information of the project; and a predictive model construction process that performs machine learning of the predictive model using explanatory variables based on the learning-time metrics and a target variable based on the project success or failure result, wherein the project information includes management indicators and documents; and the project information processing process that performs a numerical indicator processing process that converts the management indicators into the learning-time metrics and a natural language processing process that converts the documents into the learning-time metrics.

[0019] Another preferred aspect of the present invention is a method for predicting project success or failure using a prediction model trained by the above-mentioned machine learning method for a prediction model, comprising: executing a second project information processing process to generate metrics at the time of prediction from project information of the project up to the time of prediction execution; inputting explanatory variables based on the metrics at the time of prediction into the prediction model; and predicting project success or failure based on an output objective variable; the project information includes management indicators and documents; and the second project information processing process executing a second numerical indicator processing process to convert the management indicators into the metrics at the time of prediction and a second natural language processing process to convert the documents into the metrics at the time of prediction. Effect of the Invention

[0020] It is possible to provide a technology that predicts the success or failure of a project by utilizing more multifaceted decision-making information than was previously possible. [Brief description of the drawings]

[0021] [Figure 1] 2 is a block diagram showing internal data of each processing unit and a recording device constituting the software development project success / failure prediction device. FIG. [Diagram 2] 2 is a block diagram showing a hardware configuration for executing the processes of the software development project success / failure prediction device; FIG. [Diagram 3] 1 is a block diagram showing a data flow of a software development project success / failure prediction device; [Figure 4] FIG. 1 is a table showing examples of project management indicators. [Diagram 5] FIG. 1 is a table showing an example of project management documentation. [Figure 6] FIG. 13 is a table showing a past project success / failure result record table. [Figure 7] FIG. 13 is a table showing a success / failure prediction value record table. [Figure 8] FIG. 4 is a table showing an example of a keyword record table according to the first embodiment. [Figure 9]FIG. 4 is a table showing an example of a word string classification table according to the first embodiment. [Figure 10] FIG. 2 is a table showing an example of a keyword score table according to the first embodiment. [Figure 11] FIG. 13 is a table showing an example of a metrics record table. [Figure 12] FIG. 13 is a block diagram showing the data flow of the past project information processing unit. [Figure 13] 1 is a flow diagram showing a machine learning processing flow using past project information. [Figure 14] 13 is a flow diagram showing a process flow for predicting success or failure from ongoing project information. [Figure 15] FIG. 13 is a block diagram showing the data flow of the ongoing project information processing unit. [Figure 16] 13 is a flowchart showing a process flow of a project information extraction unit. [Figure 17] FIG. 11 is a table showing an example of a word string classification table according to the second embodiment. [Figure 18] FIG. 11 is a table showing an example of a keyword record table according to the second embodiment. [Figure 19] FIG. 11 is a table showing an example of a keyword score table according to the second embodiment. [Figure 20] 11 is a flowchart showing a process flow of a natural language processing unit according to the second embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0022] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. Note that the embodiment described below does not limit the invention according to the claims. Furthermore, not all of the elements and combinations thereof described in the embodiment are necessarily essential to the solution of the invention.

[0023] In the configurations of the embodiments described below, the same parts or parts having similar functions are denoted by the same reference numerals in different drawings, and duplicated explanations may be omitted.

[0024] When there are multiple elements having the same or similar functions, they may be described by using the same reference numerals with different subscripts. However, when there is no need to distinguish between multiple elements, the subscripts may be omitted.

[0025] The designations "first," "second," "third," and the like in this specification are given to identify components, and do not necessarily limit the number, order, or content. Furthermore, numbers for identifying components are used in different contexts, and a number used in one context does not necessarily indicate the same configuration in another context. Furthermore, a component identified by a certain number does not prevent the component from also having the function of a component identified by another number.

[0026] In order to facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings, etc. may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings, etc.

[0027] All publications, patents, and patent applications cited herein are hereby incorporated by reference in their entirety.

[0028] As used herein, elements referred to in the singular are intended to include the plural unless the context clearly indicates otherwise. EXAMPLES

[0029] Hereinafter, an embodiment of the present invention will be described. FIG. 1 is a diagram showing internal data of each processing unit and a recording device constituting a software development project success / failure prediction device. FIG. 2 is a diagram showing a hardware configuration for executing the processes of the software development project success / failure prediction device. FIG. 3 shows a data flow of the software development project success / failure prediction device. First, the configuration of the software development project success / failure prediction device will be described with reference to Figs.

[0030] Fig. 1 is a functional block diagram of a software development project success / failure prediction device 10 according to an embodiment. As shown in Fig. 1, the software development project success / failure prediction device 10 includes a project information processing unit 101 consisting of a natural language processing unit 103 and a numerical index processing unit 102. It also includes a project information extraction unit 104, a prediction model construction unit 105, and a success / failure prediction unit 106. The functions of each of these processing units are implemented as a program, as will be described later.

[0031] The project information recording database (DB) 110 records a past project success / failure result recording table 111, a keyword recording table 112, project information 113 including project management indicators 114 and project management documents 115, and a success / failure prediction value recording table 116. As long as the software development project success / failure prediction device 10 can use the data, the project information recording DB 110 may be a part of the software development project success / failure prediction device 10 (for example, the auxiliary storage device 203 described below) or may be configured as another device connected via a network or the like. In this embodiment, an example will be described in which it is configured as the auxiliary storage device 203.

[0032] The hardware and software configuration of the software development project success / failure prediction device 10 will be described with reference to FIG.

[0033] The software development project success / failure prediction device 10 is an information processing device (computer) used for managing a software development project, etc. The software development project success / failure prediction device 10 may be realized virtually, such as a cloud server provided by a cloud system.

[0034] The software development project success / failure prediction device 10 includes a processor 201, a main memory device 202, an auxiliary memory device 203, an input device 204, an output device 205, and a communication device 206. These components are connected to each other so as to be able to communicate with each other via a communication means such as a bus.

[0035] The processor 201 is a logic circuit realized by a semiconductor device such as a CPU (Central Processing Unit) or an MPU (Micro Processing Unit). The processor 201 executes a program stored in the main memory device 202, thereby realizing various functions of the software development project success / failure prediction device 10. The program is read from the auxiliary memory device 203 or another memory device and stored in the main memory device 202.

[0036] The main memory device 202 is a device such as a RAM (Random Access Memory) that stores programs and data.

[0037] The auxiliary storage device 203 is a hard disk drive (HDD), a solid state drive (SSD), an optical storage device (such as a compact disc (CD) or a digital versatile disc (DVD)), a storage system, a read / write device for recording media such as an IC card, an SD memory card, or an optical recording medium, a storage area of ​​a cloud server, etc. Programs and data stored in the auxiliary storage device 203 are loaded into the main storage device 202 as needed.

[0038] The input device 204 is a device for inputting information to the software development project success / failure prediction device 10, and is, for example, a keyboard, a mouse, a touch panel, a card reader, a voice input device, or the like.

[0039] The output device 205 is a device that realizes a user interface that provides the user with various information such as the processing progress and processing results in the software development project success / failure prediction device 10, and is, for example, a screen display device (liquid crystal monitor, LCD (Liquid Crystal Display), graphics card, etc.), an audio output device (speaker, etc.), a printing device, etc.

[0040] The communication device 206 is a device that realizes a wired or wireless communication interface that realizes communication with other devices via communication means such as a LAN or the Internet, and is, for example, a NIC (Network Interface Card), a wireless communication module, a USB (Universal Serial Interface) module, a serial communication module, etc. In this device, the communication device 206 is used to connect to the project information recording DB 110.

[0041] A numerical index processing program 207, a natural language processing program 208, a prediction model construction program 210, a success / failure prediction program 211, and a project information extraction program 209 are installed in the auxiliary storage device 203 of the software development project success / failure prediction device 10 of this embodiment.

[0042] In addition, a keyword score table 213, a learning time metrics record table 212, and a prediction model 215 are stored.

[0043] The numerical indicator processing program 207, the natural language processing program 208, the predictive model construction program 210, the success / failure prediction program 211, and the project information extraction program 209 are programs which, when executed by the processor 201, respectively realize the functions of the numerical indicator processing unit 102, the natural language processing unit 103, the predictive model construction unit 105, the success / failure prediction unit 106, and the project information extraction unit 104 in Figure 1.

[0044] 1 to 3, the project information extraction unit 104 is responsible for dividing the project information 113 into past project information and ongoing project information, and outputting them to the main storage device 202. When dividing the information here, it is necessary to refer to an indicator showing whether each project is ongoing or completed, but this reference indicator is omitted in this device and is stored in a separate table.

[0045] The project information processing unit 101 processes the input past and ongoing project information into a form suitable for use as explanatory variables (metrics) for machine learning, and then records it in the learning time metrics recording table 212 and the prediction time metrics recording table 214.

[0046] The predictive model construction unit 105 uses the metric values ​​of multiple projects recorded in the learning metrics record table 212 as explanatory variables and the predicted success / failure values ​​of each project recorded in the past project success / failure result record table 111 as objective variables through machine learning to construct a predictive model 215.

[0047] The success / failure prediction unit 106 uses the prediction model 215 generated above and the metrics recorded in the prediction time metrics recording table 214 to perform success / failure prediction through machine learning, and records the results in the success / failure prediction value recording table 116.

[0048] The numerical index processing unit 102 receives the project management index 114 as an input and converts the index value into metrics.

[0049] The natural language processing unit 103 receives the project management document 115 as an input and plays a role in generating metrics from the natural language in the document.

[0050] Furthermore, the project information recording DB 110 stores project information 113 consisting of project management indexes 114 and project management documents 115, a past project success / failure result recording table 111, a keyword recording table 112, and a success / failure prediction value recording table 116. Since both the project management indexes 114 and the project management documents 115 are information accumulated during the course of a project, hereinafter these will be collectively referred to as project information 113.

[0051] The project management index 114 is a table for recording project management indexes such as the amount of deliverables and the amount of defects.

[0052] FIG. 4 shows an example of project management indicators 114. It is assumed that indicator values ​​704 at the time of completion of each process 703 of the project are stored for each project number P# 701 and indicator type 702. The indicator type may be freely defined, for example, "deliverable amount" or "defect amount," but is basically an indicator expressed as a numerical value. If it can be converted into a numerical value, it may be ranked, such as A, B, or C. Each process 703 may store the completion time of the process, or may store a number or symbol that identifies the process defined in the project (for example, requirements definition, design, debugging, etc.). In this embodiment, the process 703 is defined commonly for each project.

[0053] An example of a project management document table is shown in Fig. 5. Project management document 115 is a table that records documents accumulated during the course of a project, such as email content, weekly reports, pending issues, and corrections. In reality, documents are often saved in text file format rather than in a table, but for the sake of convenience, it is assumed here that all of the text in a document is stored in one cell in a table.

[0054] Data is stored for each project number P#701. The storage date 802 stores the date when the document was stored. The document type 803 stores the name of the document type (for example, weekly report, report, email, etc.). The text 804 stores all the text within the document.

[0055] 6 is a diagram showing a past project success / failure result record table. The past project success / failure result record table 111 is a table that stores the success / failure results of past projects as values ​​within a certain range of failure degrees that are manually determined in advance. The values ​​are determined and entered by a person with knowledge of projects, such as a PM.

[0056] The past project success / failure result record table 111 has the degree of failure of the project, which is the target variable of machine learning, as an attribute of the table. The overall degree of failure of the project is recorded in the overall failure degree result value 905 for each project number P#701. In addition, like result values ​​902 to 904, the degree of failure is divided into smaller categories such as Q (quality), C (cost), and D (delivery date) and recorded.

[0057] Fig. 7 is a diagram showing a success / failure prediction value recording table. The success / failure prediction value recording table 116 is a table for recording the success / failure prediction values ​​of ongoing projects. As shown in Fig. 7, the outline of the table is the same as the past project success / failure result recording table 111 in Fig. 6. However, instead of past success / failure result values, the success / failure prediction result values ​​by the success / failure prediction unit 106 of this device are stored as failure degree prediction values ​​in the prediction values ​​1002 to 1005.

[0058] FIG. 8 is a diagram showing an example of a keyword recording table. The keyword recording table 112 is a table in which word strings and the characteristics of those word strings are recorded. It is assumed that this table is created in advance before the execution of this device. As an example, as shown in FIG. 8, a classification such as a classification type 601 (FIG. 9) described later is recorded in classification 1102. A classification is performed for a specific word string (for example, "failure," "delay," "smooth," etc.) that appears in a document, and if it fits the classification criterion 602, that classification is entered in classification 1102. The keyword recording table 112 is determined and entered by a person in charge with knowledge of the project, such as a PM, or a document expert.

[0059] Fig. 9 is a diagram showing an example of a word string classification table. The word string classification table 600 shows the classification type of word strings and the classification criteria. In Fig. 9, classification type 601 is classified as whether the meaning is good or bad from the perspective of project management. This is determined and input by a person in charge of projects such as a PM or a document expert. Although it is not a component of this device, it is described for the purpose of explaining this embodiment.

[0060] 10 is a diagram showing an example of a keyword score table. The keyword score table 213 consists of, for a word string 1203 that appears in a document, its appearance date 1202, its classification 1204, and a score value 1205 that indicates the importance of the word string to the entire document. As will be explained later, the score value is calculated by the natural language processing unit 103.

[0061] Fig. 11 is a diagram showing an example of a metrics record table. The learning-time metrics record table 212 stores the metrics generated by the project information processing unit 101 during learning in the value 1303. As will be described later, for natural language data, the aggregation results are stored as values, so as in the example of Fig. 11, the character string "aggregation results" is added after the classification, such as "Quality-aggregation results".

[0062] That is, a predetermined word is extracted by referring to a keyword record table 112 (FIG. 8) showing predetermined word classifications, and an attribute based on the word classification is assigned, and a character string showing the classification (attribute) is assigned to the score value. The classification of words can be set arbitrarily by the user. In the example of FIG. 11, the natural language data is classified based on the evaluation viewpoint, such as "Quality" and "Delivery". Classification may be performed by single classification or by assigning meanings such as "Good" and "Bad" according to the keyword record table 112 of FIG. 8. An example of detailed classification of natural language data based on viewpoints will be described in detail in Example 2.

[0063] Similarly, the metrics record table 214 at the time of prediction stores the metrics generated by the project information processing unit 101 at the time of prediction.

[0064] Since the learning-time metrics record table 212 and the prediction-time metrics record table 214 have the same table format, a conceptual diagram of the tables is shown together in FIG.

[0065] Next, the flow of the process executed by the software development project success / failure prediction device 10 according to the first embodiment will be described with reference to FIG. Figure 3 shows the overall data flow with inputs and outputs clearly indicated. FIG. 12 shows the details of the project information processing unit 101. Figure 13 shows the flow for generating a predictive model in the learning phase. FIG. 14 shows the flow for generating success / failure prediction results in the prediction phase.

[0066] In this embodiment, the learning phase and the prediction phase are realized by the same device, but the learning phase and the prediction phase can also be executed by different devices or systems.

[0067] Before explaining the function of each processing unit, the overall concept will be explained, including the outside of the device, with reference to Figure 3. The project manager 1 decides the process for executing the success / failure prediction among all the project processes of the ongoing project. Hereinafter, this will be called the time of executing the success / failure prediction.

[0068] 3 and 12, in the learning phase of this device, the project information extraction unit 104 extracts past project information 301 from project information 113 (including both project management indexes 114 and project management documents 115), and the project information processing unit 101 transforms the information up to the time of execution of success / failure prediction into an appropriate form as machine learning metrics. The natural language processing unit 103 of the project information processing unit 101 refers to the keyword record table 112 to generate metrics from sentences. The numerical value index processing unit 102 converts the numerical values ​​of the project management indexes 114 as necessary.

[0069] A prediction model 215 is generated by the prediction model construction unit 105 using the metrics recorded in the learning-time metrics record table 212 and the past project success / failure result record table 111. Specifically, supervised learning of the prediction model is performed using the generated metrics as explanatory variables (questions) and the past project success / failure results as objective variables (answers).

[0070] Once the predictive model 215 is generated, there is no need to execute the learning phase again until more projects are accumulated, and the same predictive model continues to be used in the predictive phase for a certain period of time for multiple ongoing projects.

[0071] The prediction phase differs from the learning phase in that it uses, as explanatory variables, metrics record table 214 at the time of prediction based on ongoing project information 302. First, as in the learning phase, metrics are generated from information up to the time of executing a success / failure prediction of the ongoing project. Using the generated metrics and a prediction model 215, the success / failure prediction unit 106 calculates a success / failure prediction value. Specifically, the generated metrics are input as explanatory variables, and a target variable (success / failure prediction value) is obtained as an output.

[0072] The project manager 1 recognizes projects with a high risk of failure based on the success / failure prediction values ​​of a plurality of ongoing projects recorded in the success / failure prediction value record table 116 .

[0073] Fig. 13 is a flowchart showing a method for learning past project information 301 and generating a prediction model 215. Thereafter, in steps 1501, 1502, and 1503 in Fig. 14, a success / failure prediction value for an ongoing project is generated using the generated prediction model 215. Each step leading up to the generation of the success / failure prediction value is described below.

[0074] 13, in step s1401, the project information extraction unit 104 receives project information 113 as input, and the project management index 114 filters and extracts only the lines for the period up to the time when the success / failure prediction is performed based on the process 703, and generates past project information 301. However, from the project management document 115, all sentences in the document with the storage date 802 up to the time when the success / failure prediction is performed are linked and extracted.

[0075] In step S1402, the past project information 301 is input, and the project information processing unit 101 processes the input into a form suitable for explanatory variables in machine learning, generates metrics, and records them in the learning-time metrics recording table 212.

[0076] In step s1403, in the prediction model construction unit 105, all metrics in the learning-time metrics record table 212 are input as explanatory variables, and the success / failure results in the past project success / failure result record table 111 are input as objective variables, and supervised machine learning is executed to generate a prediction model 215. The failure degree result values ​​902 to 905 of each item in the past project success / failure result record table 111 in Fig. 6 are input as the success / failure results of the objective variable, thereby generating a prediction model 215 for each item.

[0077] 14 is a flowchart showing a project success / failure prediction process by the success / failure prediction unit 106. Steps s1501 and s1502 in the prediction phase have the same flows as steps s1401 and s1402 in the learning phase, respectively.

[0078] In step s1501, the project information extraction unit 104 receives project information 113 of the project to be predicted, filters and extracts only the lines for the period up to the time when the success / failure prediction is executed, and generates ongoing project information 302. However, from the project management document 115, all sentences in the document with storage dates up to the time when the success / failure prediction of the project to be predicted is linked and extracted.

[0079] In step S1502, the ongoing project information 302 is input, and the project information processing unit 101 processes the input into a form suitable for explanatory variables for machine learning, generates metrics, and records them in the prediction time metrics recording table 214.

[0080] In step s1503, the success / failure prediction unit 106 performs machine learning prediction using each prediction model 215 generated in step s1403, with all metrics in the prediction time metrics record table 214 as explanatory variables, to generate success / failure prediction values ​​1002 to 1005 in the success / failure prediction value record table 116 in Figure 7, and record them in the success / failure prediction value record table 116.

[0081] Next, we will describe in detail the project information processing unit 101. The project information processing unit 101 plays a role in processing the past project information 301 and the ongoing project information 302 into an appropriate form as machine learning metrics.

[0082] 15 shows a flow in which the project information processing unit 101 processes ongoing project information during prediction. When the input is a project management index 114, the numerical index processing unit 102 generates metrics. When the input is a project management document 115, the natural language processing unit 103 generates metrics.

[0083] The natural language processing unit 103 extracts a word string from the input text using natural language processing, and calculates the frequency of occurrence of the word string. In this embodiment, this index is called the score value of the word string. The score values ​​are tallied for each project, and the tallied results are output as metrics.

[0084] Here, the numerical index processing unit 102 uses a method similar to that of Patent Document 1, and since it is not closely related to the claims of this patent, a description thereof will be omitted.

[0085] 16 is a diagram showing a processing flow of the natural language processing unit according to the embodiment 1. Only the processing flow in the natural language processing unit 103 related to this embodiment will be explained.

[0086] First, natural language processing is used to extract a word string from the sentences in the input project management document 115 (s1701).

[0087] All the extracted word strings are taken out one by one, and it is confirmed whether they exist in the keyword record table 112 (FIG. 8) (s1702). If they exist, the classification 1102 of the corresponding word string written in the keyword record table 112 is read (s1703).

[0088] Thereafter, a score value is calculated based on the frequency of occurrence of word strings classified as BAD in word string classification table 600 (s1704).

[0089] Each word string, its classification, and its score value are recorded in the keyword score table 213 in the auxiliary storage device 203 (s1705).

[0090] Finally, the score values ​​recorded in the keyword score table 213 are totaled. As an example of totaling, the sum of the word string score values ​​before the success / failure prediction is performed is normalized by the total number of words that appear, and this is used as a metric, which is recorded in the learning-time metrics record table 212 (s1706). In this embodiment, the word string score values ​​are totaled for each classification: Good or Bad, or QCD.

[0091] This embodiment makes it possible to predict the success or failure of a project based on both project management indicators and project management documents, which makes it possible to discover signs of failure from a wider range of factors than with existing technologies. EXAMPLES

[0092] In Example 2, an example of improving accuracy will be shown by further developing Example 1. By refining the classification of word strings, the dimensions of explanatory variables can be expanded.

[0093] Fig. 17 is a diagram showing an example of a word string classification table 1800 according to the second embodiment. It corresponds to the table of Fig. 9 in the first embodiment, but instead of classification into good and bad, classification types 1801 and classification criteria 1802 classified from different perspectives of project management are stored. These can be set arbitrarily by the administrator. In this example, words are classified into Delivery (D) indicating delivery date, Cost (C) indicating cost, and Quality (Q) indicating quality (QCD classification).

[0094] Fig. 18 is a diagram showing an example of the keyword record table 112 according to the embodiment 2. It corresponds to the table of Fig. 8 in the embodiment 1, but in addition to the good / bad classification 1102, a QCD classification 1901 based on the classification of the word string classification table 1800 is added.

[0095] Fig. 19 is a diagram showing an example of the keyword score table 213 according to the embodiment 2. This corresponds to the table of Fig. 10 in the embodiment 1, but a QCD classification 2001 is added.

[0096] 20 is a diagram showing a processing flow of the natural language processing unit 103 according to the second embodiment, which corresponds to the processing flow of FIG.

[0097] The processing flow of the natural language processing unit 103 of this embodiment shown in Fig. 20 performs processing similarly to that of Fig. 16 in the first embodiment, but is different from step s2003 onwards. In step s2003, the natural language processing unit 103 considers the QCD classification criteria as shown in Fig. 17 in addition to the good / bad classification criteria shown in Fig. 9.

[0098] 18 shows a specific example of the keyword record table 112 to be input to the natural language processing unit 103. In addition to the word strings 1101 and good / bad classifications 1102, the QCD classifications 1901 of each word string are also shown.

[0099] In the flow of FIG. 20, when a predetermined word string is extracted in s1702, the keyword record table 112 is referenced and the attributes of the extracted word string (good / bad classification 1102 and QCD classification 1901) are read (s2003).

[0100] Thereafter, for each word string in the keyword record table 112, a score value 1205 is calculated based on the frequency of occurrence for each QCD (s2004).

[0101] Each word string 1203, the BAD classification 1204, the QCD classification 2001, and the score value 1205 are recorded in the keyword score table 213 in the auxiliary storage device 203 (s2005).

[0102] The score values ​​of each word string recorded in the keyword score table 213 shown in FIG. 19 are tallied in the same manner as in the first embodiment, and the tallied results are used as metrics and are recorded in the learning-time metrics record table 212 (s2006). By increasing the number of QCD classifications as inputs, the dimensionality of the explanatory variables increases, which is expected to improve the accuracy of machine learning.

[0103] According to the above embodiment, efficient project management can be realized, which enables efficient project execution, reduces energy consumption and carbon emissions, prevents global warming, and contributes to the realization of a sustainable society. [Explanation of symbols]

[0104] Software development project success / failure prediction device 10 Project Information Processing Department 101 Numerical Index Processing Department 102 Natural Language Processing Department 103 Project information extraction unit 104 Prediction Model Building Department 105 Success / Failure Prediction Department 106 Project information record DB 110 Past project success / failure record table 111 Keyword Record Table 112 Project Information 113 Project Management Metrics 114 Project Management Documents 115 Success / failure prediction value record table 116

Claims

1. a success / failure prediction unit that predicts the success or failure of a project using a prediction model generated by machine learning; a project information processing unit that generates forecast time metrics from project information of the project up to a forecast execution time, An explanatory variable based on the prediction time metrics is input to the success / failure prediction unit, and the success / failure of the project is predicted based on the output objective variable; The project information includes: Includes management indicators and documentation, The project information processing unit includes: a numerical index processing unit that converts the management index into the prediction-time metrics, and a natural language processing unit that converts the document into the prediction-time metrics, The natural language processing unit Using a keyword record table in which word strings are classified and registered in advance according to whether they have positive or negative meanings from the perspective of project management, Extracting word strings from the input text using natural language processing, and calculating the frequency of occurrence of the word strings; The occurrence frequency is aggregated for each project as a score value, and the aggregated results are output as metrics. Project success / failure prediction device.

2. A machine learning method for a predictive model that predicts the success or failure of a project, comprising: A project information processing process for generating learning metrics from project information of the project; Executing a predictive model construction process for performing machine learning of the predictive model using explanatory variables based on the learning metrics and a target variable based on the project success / failure results; The project information includes: Includes management indicators and documentation, The project information processing process includes: Executing a numerical index processing process to convert the management index into the training metrics, and a natural language processing process to convert the document into the training metrics; The natural language processing includes: Using a keyword record table in which word strings are classified and registered in advance according to whether they have positive or negative meanings from the perspective of project management, Extracting word strings from the input text using natural language processing, and calculating the frequency of occurrence of the word strings; The occurrence frequency is aggregated for each project as a score value, and the aggregated results are output as metrics. Machine learning methods for predictive models.

3. A method for predicting project success or failure using a prediction model machine-learned by the machine learning method for predictive models according to claim 2, execute a second project information processing process to generate forecast time metrics from project information of the project up to the time of forecast execution; Input explanatory variables based on the prediction metrics into the prediction model, and predict the success or failure of the project based on the output objective variables; The project information includes: Includes management indicators and documentation, The second project information processing process includes: A second numerical index processing process is executed to convert the management index into the prediction time metrics, and a second natural language processing process is executed to convert the document into the prediction time metrics. How to predict project success or failure.

4. The second natural language processing step includes: converting words in the document into score values ​​based on the frequency of occurrence of predetermined words in the document; The method for predicting project success or failure according to claim 3.

5. The second natural language processing step includes: extracting the predetermined word by referring to a keyword record table indicating a predetermined word classification, and assigning an attribute based on the word classification; assigning said attribute to said score value; 5. The method for predicting project success or failure according to claim 4.

Images