KR100586503B1 - System for tailoring a software development process using neural network and method of the same - Google Patents

Abstract

A software process tailoring system and method using neural networks is disclosed. The database unit stores tailoring related information including information on a target process, tailoring parameters, and task extraction rules. The tailoring related information management unit stores the tailoring related information input from the outside in the database unit or sets the relation between information information. The tailoring learning material management unit generates a learning material by applying a task extraction rule to the combination values generated by combining the values of the tailoring parameters. The neural network generation / learning unit generates a neural network by setting the number of units including an input unit, an output unit, and a hidden unit, an activation function of each unit, and a connection state of the unit, and a target value derived from the performance of the process. The neural network is trained by the learning data so that an error between prediction values derived by the neural network is smaller than a predetermined threshold value. The task suitability determining unit converts the result of the process performed by the neural network into a percentage to determine the suitability of the task based on the criteria desired by the organization. According to the present invention, the experience and knowledge can be efficiently reused and the automation of the work can be further improved compared to a rule-based system that can use the experience and knowledge of the process tailoring work only in situations where it has already occurred.

Description

System for tailoring a software development process using neural network and method of the same}

1 is a block diagram showing a detailed configuration of a software process tailoring system using a neural network according to the present invention;

2 is a structural diagram of a multi-layer forward neural network applied to a software process tailoring system and method using a neural network according to the present invention;

3 is a flowchart illustrating a process of performing a software process tailoring method using a neural network according to the present invention.

The present invention relates to a software process tailoring system and method using a neural network, and more particularly, to improve the efficiency of a process for obtaining a desired level of quality in software development, and to develop a process suitable for a development environment. The present invention relates to a software process tailoring system and method.

Software process tailoring requires a lot of software process knowledge and software process tailoring experience. However, since this knowledge and experience is difficult to obtain, the software process tailoring work is performed by a small number of professionals with a lot of software process knowledge and software process tailoring experience. As such, software process tailoring work by experts is expensive because it is rare for experienced professionals, and each software process tailoring work is optimized for the situation, making it difficult to use past tailoring experiences or materials in new situations. have.

In order to solve this problem, efforts are being made to reuse experience and knowledge by acquiring and storing experience and knowledge related to software process tailoring through research of existing rule-based systems or process patterns. However, these methods are limited to situations where the tailoring of software processes has already occurred in the past, and the reuse of knowledge and experience of the task is limited.

The technical problem to be achieved by the present invention is to provide a software process tailoring system and method using a neural network to learn the experience or knowledge of experts through a neural network and to enable the software process tailoring work in a new situation.

Another technical problem to be achieved by the present invention is a program for executing a software process tailoring method using a neural network on a computer to learn the experience or knowledge of experts through a neural network and to use the software process tailoring work in a new situation. To provide a computer-readable recording medium for recording.

According to an aspect of the present invention, there is provided a software process tailoring system using a neural network, including: a database unit configured to store tailoring related information including information on a target process, a tailoring parameter, and a learning material generation rule; A tailoring related information manager configured to store the tailoring related information input from the outside in the database unit; A tailored learning material management unit generating the learning material by applying the learning material generation rule to the combined values generated by combining the values of the tailoring parameter; The neural network is generated by setting the number of units including the input unit, the output unit, and the hidden unit, the activation function of each unit, and the connection state of the unit, and the target value derived from the performance of the process and the neural network derived by the neural network. A neural network generation / learning unit for learning the neural network by the learning material such that an error between the predicted values is smaller than a predetermined threshold value; And a task suitability determining unit for converting the result of the process performed by the neural network into a percentage to determine the suitability of the task based on a criterion desired by the organization.

In order to achieve the above technical problem, the software process tailoring method using the neural network according to the present invention includes tailoring related information including information on a target process, a tailoring parameter, and a learning data generation rule to a predetermined database unit. Storing; Generating the learning material by applying the learning material generation rule to the combined values generated by combining the values of the tailoring parameter; Generating a neural network by setting the number of units including an input unit, an output unit, and a hidden unit, an activation function of each unit, and a connection state of the units; Training the neural network by the learning data such that an error between a target value derived as a result of performing the process and a predicted value derived by the neural network is smaller than a predetermined threshold value; And converting a result of the process performed by the neural network into a percentage to determine a fitness of the task based on criteria desired by the organization.

This allows for more efficient reuse of experience and knowledge than rule-based systems that can use the experience and knowledge of process tailoring tasks only in situations where they have already occurred.

Hereinafter, with reference to the accompanying drawings will be described in detail a preferred embodiment of the software process tailoring system and method using a neural network according to the present invention.

1 is a block diagram showing a detailed configuration of a software process tailoring system using a neural network according to the present invention.

Referring to FIG. 1, the software process tailoring system 100 using the neural network according to the present invention includes a tailoring related information management unit 110, a tailoring learning material management unit 120, a neural network generation / learning unit 130, and a task suitability determination unit. 140, a database unit 150 is provided.

The tailoring related information management unit 110 receives and manages information related to process tailoring such as a target process, a tailoring parameter, a learning material generation rule, and the like. The tailoring related information management unit 110 stores the input process tailoring related information in an appropriate location of the database unit 150 or sets the relation between information information. The target process can be briefly defined as the ID of the required job, the name of the job, and the worker. This embodiment describes the definition and management of process tailoring related information for activities in a Rational Unified Process (RUP) Project Management Discipline. Next, the tailoring parameters are defined. The tailoring parameter is a parameter that can evaluate a project situation where software process tailoring is required and can be defined as an item as described in the following table.

Item Nominal value Item value External elements of the organization  Project development type Contract based ordering project 4 Commercial product development projects 3 Internal project 2 Experimental project One Project Partners height One lowness 2 Internal Elements of Organization  Project scale Very large (more than 125 people) 4 Large (26 ~ 125 people) 3 Normal (6 ~ 25 people) 2 Low (~ 5 people) One Geographical distribution Dispersion 2 Concentration One  Development process maturity height One middle 2 lowness 3  Development team domain experience height One middle 2 lowness 3 Characteristics of the target system  New degree height 3 middle 2 lowness One  Architectural risks height 3 middle 2 lowness One  Technical complexity height 3 middle 2 lowness One  Management Complexity height 3 middle 2 lowness One

Referring to Table 1, items are largely divided into external and internal elements of the organization and characteristics of the target system to be developed. Each item can be adjusted according to the needs of the organization or the project situation. However, an item to consider when adjusting the item is that the process tailoring parameter is mapped to an input attribute when constructing a neural network model, and an assigned value is mapped to an input attribute value. Therefore, in order to be used as an input attribute value of the neural network model, the process tailoring parameter value must be converted into a numeric form.

In addition, when the nominal value of the process tailoring parameter is converted into a number, a larger value is assigned toward increasing the stereotype of the process. The process formality here refers to how much a process has a task, and a process having many tasks is said to have a high degree of formality.

In the present invention, since the software process tailoring material is trained using a neural network, the existing software process tailoring experience data must be obtained. However, if there is no existing data on software process tailoring, we define the rules for generating learning materials based on general knowledge about software process tailoring. Using this, general knowledge is generated as a learning material and used for neural network generation / learning.

The training material generation rule can be divided into environmental constraints and task extraction rules.

Environmental constraints are conditions that eliminate contradictions that arise between tailoring parameters defined by the organization. For example, referring to the tailoring parameters listed in Table 1, when the project type is a commercial product development project, it is likely that a large number of 'project related partnerships' will be high in product development because it targets a large number of unspecified customers in the future. Can not. This knowledge or fact can be regularized as follows:

IF (Project Development Type = 3) THEN (Project Related Partners = 2)

Task extraction rules define the knowledge about whether a particular job is used in a particular situation. Task extraction rules can be defined differently according to the process adopted by the organization, the organization's policies, and the experience and control of the process engineer. However, this rule is based on a rule that is self-explanatory by consensus within the organization intending to use this method.

The task extraction rule describes "Creating a business use case" as an example of activities in the RUP's project management phase. A "business use case" is a scenario for understanding the flow of work in an organization. In general, "business use case creation" is performed only when the nature of the project is a contract-based project and it is important to understand the work or when the project is large. In other words, if the project development type is an experimental project or the project is small, the fact that it does not apply can be created as a task extraction rule as follows.

IF {(Project Development Type = 1) & (Project Scale = 1)} THEN (Develop Business Use Case = 0)

The tailoring learning material management unit 120 generates a learning material based on the learning data generation rule defined by the tailoring related information management unit 110 or generates and adds existing process tailoring experience data as learning material, and so on. Carry out the work of turning the material into learning materials.

로 표시되고, 학습 자료는 <

>로 표시된다. 이 때,

는 테일러링 파라미터 값들,

는 해당 테일러링 파라미터 값들로 표현되는 프로젝트 상황에서 요구되는 태스크 선별 여부 값들로 주어진다.In addition, the tailored learning data management unit 120 displays the learning materials as a pair of input values and target output values according to the input, which are types of learning materials required by the neural network. Figure 2 shows the structure of a multi-layer forward neural network applied to the software process tailoring system and method using the neural network according to the present invention. Referring to Figure 2, the input value and the target output value, respectively

, And the learning material is <

Is indicated by>. At this time,

Is the tailoring parameter values,

Are given as task selection values required in the project situation represented by corresponding tailoring parameter values.

The tailoring learning material manager 120 generates all situations that may occur in the tailoring parameters by combining the values of the tailoring parameters, and removes the situations that do not occur by applying environmental constraint rules among the generated conditions. Through this process, the tailoring learning material management unit 120 applies the task extraction rule to the tailoring parameter combination values satisfying all of the environment constraint rules, and sets the unnecessary work by the rule to 0 to generate the learning material. This transformed learning material can be used to train neural networks with environmental constraints and task extraction rules.

On the other hand, since the learning material stored in the database unit 150 is a general knowledge of software process tailoring, it may not sufficiently reflect the situation that occurs in the actual software process tailoring such as a special environment or personality of an organization. To solve this problem, the actual tailoring experience data generated in the existing organization is added so that the neural network can be learned to reflect the actual situation. In addition, the process tailoring prediction performed by the neural network may have an error. Accordingly, after applying the process tailoring according to the present invention to the actual project situation, the process tailoring result is calculated, and additionally required tasks or tasks not required for the situation are reflected in the process tailoring result. The process tailoring result reflecting the execution result is generated as an additional feedback learning material, and it can be added to the existing learning material to correct the error generated by learning the neural network again.

The neural network generation / learning unit 130 performs neural network generation and learning of neural networks in accordance with the learning data generated by the tailoring learning material management unit 120. Through this process, the neural network learns the basic rules related to process tailoring, the process tailoring experience that occurred in the actual organization in the past, and feedback data of the values predicting the existing process tailoring.

을 각 층으로 전달하여 출력값 벡터

를 계산하는 전파 단계와 출력층에서 발생되는 에러를 줄이기 위해 신경망 내의 연결 가중치를 수정하는 역전파 단계로 나뉜다.The neural network generation / learning unit 130 trains the neural network in a manner to reduce the error between the desired result value of the learning data and the neural network prediction value and calculates the predicted value by using the connection strength between units of each layer. A typical example of a neural network is a multi-layered feed forward neural network using the backpropagation algorithm shown in FIG. Multi-Layered Feed Forward Neural Networks are neural networks in which one or more intermediate layers exist between the input and output layers. At this time, the input layer and the output layer are called hidden layers, and the neural network is connected in the input layer, hidden layer, and output layer directions. There is no connection in each layer and a direct connection from the output layer to the input layer. The neural network generation / learning unit 130 learns the neural network structure using a back propagation algorithm. Backpropagation algorithm is an input vector

Is passed to each layer to output vector

It is divided into a propagation step of calculating a and a back propagation step of modifying a connection weight in a neural network to reduce an error occurring in an output layer.

In the propagation phase, each unit multiplies the input values connected to them by the weight of the connection and then sums them. The formula for summing input values of each unit is as follows.

In the above equation, n is the number of input values, and x _n is the input vector of the training material. w _n means a weight connecting the input vector and the input layer unit.

When the input values of each unit are added as described above, the neural network generation / learning unit 130 generates an output value using a sigmoid activation function. The formula that generates the result by applying the activation function is as follows.

In the same manner as above, the hidden and output layers multiply the input value with the corresponding weight and generate and propagate the output value through the sigmoid activation function.

에 따라 각 출력 유닛에서 출력값 벡터

를 계산하게 된다. 이 때, 해당 입력값 벡터

에 신경망이 출력해야 하는 출력값벡터

와의 차이가 발생하게 된다. 각 출력 유닛의 오차는 다음의 수학식에 의해 계산된다.This propagated input vector

Output vector at each output unit

Will be calculated. At this time, the corresponding input vector

Output vector that neural network should output

And a difference occurs. The error of each output unit is calculated by the following equation.

In order to reduce the error of the output unit calculated by Equation 3, each connection weight of the neural network is modified through a back propagation step. First, in order to update the connection weights, calculation of an error term in each unit is necessary.

은 다음의 수학식에 의해 계산된다.Error term at the output layer

Is calculated by the following equation.

은 다음의 수학식에 의해 계산된다. Also, the error term in the hidden layer

Is calculated by the following equation.

The error term of the input layer is also calculated in the same way as the error term of the hidden layer. At this time, the weight from the input layer to the hidden layer is calculated by the following equation.

를 만족한다.here,

Satisfies.

The weight of the output layer in the hidden layer is also updated in the same manner as above.

Such a neural network may be configured with the following detailed settings. Neural network configuration and settings can be changed as needed.

Item value Explanation Neural Network Settings Input unit n pcs Tailoring Parameter Coefficient Output unit m pcs The number of tasks the selected process has in your organization Concealment unit (n + m) / 2 The number of intermediate units in the role that pass input Active function Sigmoid Activation function for each unit Connection status Feed-forward Unit connection state of neural network Set up learning to use the BackPropagation algorithm Learning rate 0.3 The amount by which weights are updated Number of learning 10 Number of cycles to learn

Next, the neural network generation / learning unit 130 learns the neural network generated by making the learning materials and the basic learning materials generated based on the rules and the experiences of the organization. At the end of the training, the connection weights in the neural network are adjusted so that the learning data can be output close to the desired target. In this process, the project status is entered using the values of the predefined tailoring parameters.

The task suitability determination unit 140 performs a process tailoring operation for a corresponding situation through neural networks learned when necessary for process tailoring. In this process, the project status is entered using the values of the predefined tailoring parameters. When the project status is input using the tailoring parameter, each input is propagated to the output layer, and the task suitability determination unit 140 converts the result generated by the percentage into a percentage to generate / determine the task suitability based on the criteria desired by the organization.

3 is a flowchart illustrating a process of performing a software process tailoring method using a neural network according to the present invention.

Referring to FIG. 3, the tailoring related information management unit 110 stores information about a target process, a tailoring parameter, and a learning material generation rule from an operator in the database 150 (S300). Information about the target process includes a job ID, a job name, and an operator, and the tailoring parameters include external elements of the organization, internal elements, and features of the system to be developed. On the other hand, the learning data generation rule consists of an environment constraint rule, which is a condition for eliminating contradictory situations occurring between the tailoring parameters defined by the organization, and a task extraction rule defined by knowledge related to the use of a specific task in a specific situation.

Next, the tailoring learning material generation unit 120 generates all situations that may occur in the tailoring parameter by combining the tailoring parameter values (S310). At this time, the tailoring learning material generation unit 120 removes a situation that will not occur by applying environmental constraint rules among the generated situation, and generates the learning material by setting unnecessary tasks to 0 by applying a task generation rule within this range. do. In addition, the tailoring learning material generation unit 120 checks the existence of the actual tailoring experience material generated in the existing organization (S320), and if there is a new tailoring experience material is added to the learning material of the database (S360).

The neural network generation / learning unit 130 generates a neural network by setting the number of units including an input unit, an output unit, and a hidden unit, an activation function of each unit, and a unit connection state of the neural network, A neural network generated by a method of reducing an error between a desired result value and a neural network prediction value (that is, by repeatedly performing a predetermined process so that the error becomes smaller than a predetermined threshold or the error becomes 0) is trained (S330). . When the learning of the neural network is completed, the weight in the neural network is adjusted so that the result value is closer to the desired target value in the learning material. At this time, the neural network generation / learning unit 130 inputs a tailoring parameter value representing the project situation. Check (S340). If the tailoring parameter value is input, each tailoring parameter value is propagated to the output layer (S370).

The task suitability determination unit 140 first checks whether a tailoring parameter value indicating a project status is input (S340). If the tailoring parameter value is input, each tailoring parameter value is propagated to the output layer (S370). Next, the task suitability determination unit 140 converts the result generated by the propagation to the output layer into a percentage to determine the suitability of the task based on criteria desired by the organization (S380). Next, the task suitability determination unit 140 checks whether an error of the process tailoring measurement value performed by the neural network is detected (S390). If there is an error, the tailoring learning material management unit 120 generates feedback learning material (S400). When the feedback learning material is generated, step S330 is performed again. If the project environment that you want to perform process tailoring has to be added or changed because there are no existing tailoring parameters, the process tailoring criteria have changed and you can see it as a new process tailoring task and stop the current process tailoring task and create a new process tailoring task. Restart with.

The invention can also be embodied as computer readable code on a computer readable recording medium. The computer-readable recording medium includes all kinds of recording devices in which data that can be read by a computer system is stored. Examples of computer-readable recording media include ROM, RAM, CD-ROM, magnetic tape, floppy disk, optical data storage, and the like, and may also be implemented in the form of a carrier wave (for example, transmission over the Internet). Include. The computer readable recording medium can also be distributed over network coupled computer systems so that the computer readable code is stored and executed in a distributed fashion.

Although the preferred embodiments of the present invention have been shown and described above, the present invention is not limited to the specific preferred embodiments described above, and the present invention belongs to the present invention without departing from the gist of the present invention as claimed in the claims. Various modifications can be made by those skilled in the art, and such changes are within the scope of the claims.

According to the software process tailoring system and method using the neural network according to the present invention, by learning the neural network knowledge and experience related to the process tailoring task, it is possible to predict the process tailoring for a new situation, thereby experience and knowledge of the process tailoring task Compared to rule-based systems, which can only be used in situations that already exist, they can efficiently reuse experience and knowledge and further automate tasks. In addition, automating process tailoring tasks compared to existing systems can reduce the cost and execution time of tasks.

Claims (13)

A database unit for storing tailoring related information including information on a target process, tailoring parameters, and learning material generation rules; A tailoring related information manager configured to store the tailoring related information input from the outside in the database unit; A tailored learning material management unit generating the learning material by applying the learning material generation rule to the combined values generated by combining the values of the tailoring parameter; The neural network is generated by setting the number of units including the input unit, the output unit, and the hidden unit, the activation function of each unit, and the connection state of the unit, and the target value derived from the performance of the process and the neural network derived by the neural network. A neural network generation / learning unit for learning the neural network by the learning material such that an error between the predicted values is smaller than a predetermined threshold value; And And a task suitability determining unit for converting the result of the process performed by the neural network into a percentage to determine the suitability of the task based on a criterion desired by the organization. The method of claim 1, The task extraction rule comprises an environment constraint rule which is a condition for removing contradictions occurring between tailoring parameters defined by an organization and a task extraction rule defined by knowledge related to the use of a specific task in a specific situation. Software process tailoring system using neural network. The method of claim 1, The neural network is a multi-layer forward neural network using a back propagation algorithm and having at least one intermediate layer between an input layer and an output layer. The method of claim 3, wherein The tailoring learning material management unit sets a task selection value required in a project situation represented by the tailoring parameter value and the corresponding tailoring parameter values as input values and target output values, respectively, and sets an ordered pair composed of the input value and the target output value. Software process tailoring system using a neural network, characterized in that provided to the input layer of the neural network. The method according to claim 3 or 4, Each unit constituting the neural network multiplies the input value connected to the multiplier by the weight of the connection, and adds the sum result as an input of an activation function set for each unit to generate an output value and propagate it to another unit. Software process tailoring system using a neural network, characterized in that. The method of claim 5, The neural network generation / learning unit may determine a difference between a target output value that the neural network outputs and an output value that the neural network outputs with respect to an input value input to the neural network.

Calculating the error of each output unit, and modifying the connection weight of each unit constituting the neural network to reduce the calculated error. Storing tailoring related information including information on a target process, tailoring parameters, and learning material generation rules, in a predetermined database unit; Generating the learning material by applying the learning material generation rule to the combined values generated by combining the values of the tailoring parameter; Generating a neural network by setting the number of units including an input unit, an output unit, and a hidden unit, an activation function of each unit, and a connection state of the units; Training the neural network by the learning data such that an error between a target value derived as a result of the process and a predicted value derived by the neural network is smaller than a predetermined threshold value; And And converting the result of the process performed by the neural network into a percentage to determine a fitness of the task based on a criterion desired by the organization. The method of claim 7, wherein The task extraction rule comprises an environment constraint rule which is a condition for removing contradictions occurring between tailoring parameters defined by an organization and a task extraction rule defined by knowledge related to the use of a specific task in a specific situation. Software process tailoring method using neural network. The method of claim 7, wherein The neural network is a multi-layer forward neural network using a back propagation algorithm, wherein at least one intermediate layer exists between an input layer and an output layer. The method of claim 9, The tailoring learning material generation step, Setting a task selection value required in a project situation represented by the tailoring parameter value and corresponding tailoring parameter values as an input value and a target output value, respectively; And And generating a learning material provided to the input layer of the neural network by forming an ordered pair using the input value and the target output value. The method according to claim 9 or 10, The neural network learning step, Multiplying an input value connected to each unit constituting the neural network by a weight of a connection and then summing; And And generating the output value as the input of the activation function set for each unit, and propagating the output to another unit. The method of claim 11, The neural network learning step, The difference between a target output value that the neural network outputs and an output value that the neural network outputs with respect to an input value input to the neural network

Calculating an error of each output unit; And And reducing the calculated error by modifying the connection weight of each unit constituting the neural network. Storing tailoring related information including information on a target process, tailoring parameters, and learning material generation rules, in a predetermined database unit; Generating the learning material by applying the learning material generation rule to the combined values generated by combining the values of the tailoring parameter; Generating a neural network by setting the number of units including an input unit, an output unit, and a hidden unit, an activation function of each unit, and a connection state of the units; Training the neural network by the learning data such that an error between a target value derived as a result of performing the process and a predicted value derived by the neural network is smaller than a predetermined threshold value; And And converting the result of the process performed by the neural network into a percentage to determine the suitability of the task based on a criterion desired by the organization. The computer recording a program for executing the software process tailoring method using the neural network on a computer Recordable media that can be read by

Images