KR20160088730A - System and Method for Recommending Bug Fixing Developers based on Multi-Developer Network - Google Patents

Abstract

The present invention relates to a system and a method for recommending a bug patch developer based on a multi-developer network. More specifically, the present invention relates to a bug patch developer recommending technology for reducing a development time and costs of a software. To this end, according to an embodiment of the present invention, the bug patch developer recommending system based on a multi-developer network comprises: a preprocessing unit, an extracting unit, a modeling unit, and a recommending unit.

Description

{System and Method for Recommending Bug Fixing Developers based on Multi-Developer Network}

The present invention relates to a bug-correcting developer recommendation system and method based on a multi-developer network, and more particularly, to a bug-correcting developer recommendation technique for reducing software development time and cost.

The present invention is derived from the research carried out by the Seoul Metropolitan Government and the Seoul Industry Promotion Agency as part of the project to train the mobile business leader. [Project Assignment Number: HM120006, Title: Training of creative mobile software professionals].

Often, an error message is output or a device malfunctions in the process of using various software through a computer such as a PC (personal computer), a smart phone, or a video game machine. This phenomenon is caused by a simple device malfunction, but it is often caused by defects in the hardware or the software itself. This means that the contents of the program constituting the hardware and the software contain erroneous codes. A computer fault or malfunction caused by a program defect is called a bug.

Bugs occur more often in complex software, and bugs that occur are assigned to developers along with bug reports for correction. At this point, the new bug is assigned to the developer by the triager in the bug repository.

However, when many bug reports are not assigned to the appropriate developers, there are cases where they are reallocated, which increases the time and cost of bug correction and increases the development cost of the software.

Korean Unexamined Patent Publication No. 2012-0069388 entitled " Automatic Static Defect Classification and Reporting Automation System and Method "thereof classifies classes according to types of bugs based on data extracted from one or more bug detectors, Bug check data extracted from the bug's precondition and bug detectors are cross-checked to give added points and the bug items according to the given scores are finally adjusted for each class, Among the reported bugs, it is easy to allocate items that are likely to be true alarms, and overall, it suggests techniques that can reduce the false alarm rate.

However, the prior art is merely a bug allocation technique for efficiently reviewing a vast amount of result data extracted from the bug detector and finding the bug of the target software in a short time. In order to correct the allocated bug, It does not mention the technology to allocate efficiently to developers.

That is, in the prior art, many bug reports are not allocated to appropriate developers and are reallocated. Accordingly, the time and cost of bug correction are increased, and the development cost of software is increased. I can not solve it.

Korean Patent Publication No. 2012-0069388 (Publication date: June 28, 2012)

SUMMARY OF THE INVENTION It is an object of the present invention to provide a bug-corrected developer recommendation system and method based on a multi-developer network.

An object of the present invention is to solve the problem that the time and cost of bug correction increases and the development cost of software increases when many bug reports are not allocated to appropriate developers and are reallocated.

The present invention aims at improving the quality of software and reducing the time and cost of development by recommending an appropriate bug correction developer to correct many bugs that occur in complicated software.

According to an aspect of the present invention, there is provided a bug-corrected developer recommendation system based on a multi-developer network, including a preprocessor, an extractor, a modeler, and a recommender.

The pre-processing unit preprocesses the received new bug report, and the extracting unit extracts a bug report similar to the preprocessed new bug report in the bug report stored in the bug storage, and the modeling unit selects the bug report based on the extracted similar bug report The recommendation unit computes a recommendation rank of the candidate developer using the modeled multi-developer network, and based on the calculated recommendation rank, You can recommend a developer to fix a new bug report.

At this time, the extracting unit can extract the similar bug report by using a KLB (Kullback-Leibler divergence) between the preprocessed new bug report represented by a vector and a bug report previously stored in the bug repository, and the modeling unit The activity information may include the number of comments and commits indicating the activity information that the developer participated in the bug correction, and the modeling unit may use the developer who corrected the extracted similar bug report, Some elements of the new bug report can use the activity information of the candidate developer selected based on the developer who corrected the same report.

Meanwhile, the multi-developer network-based bug correction developer recommendation method according to an exemplary embodiment of the present invention includes a step of preprocessing a received new bug report, a bug report similar to the preprocessed new bug report in the bug report stored in the bug storage, Modeling a multi-developer network using activity information of a candidate developer selected based on the extracted similar bug report, and calculating a recommendation rank of the candidate developer using the modeled multi-developer network And recommending a developer to correct the received new bug report based on the calculated recommendation rank.

In this case, the extracting step may extract the similar bug report by using KLB (Kullback-Leibler divergence) between the preprocessed new bug report represented by a vector and the bug report previously stored in the bug repository, The modeling step may use, as the activity information, the number of comments and commits indicating the activity information in which the developer participated in the bug correction, and the modeling step may correct the extracted similar bug report One developer can use the activity information of the candidate developer selected based on the developer who has corrected the same report with some elements of the received new bug report.

The present invention is advantageous in that a bug-corrected developer recommendation system and method based on a multi-developer network can be provided.

The present invention has the effect of solving the problem of increasing the time and cost of bug correction and increasing the development cost of software when many bug reports are not allocated to appropriate developers and reallocated.

The present invention has the effect of improving the quality of software and reducing the development time and cost by recommending an appropriate bug correction developer for correcting many bugs occurring in complicated software.

The present invention has the effect of recommending a more appropriate bug correction developer by using a developer of a bug report having a component such as a new bug as well as developers of past similar bug reports and using the developer's comment information and commit information .

1 is a schematic block diagram of a bug-corrected developer recommendation system based on a multi-developer network according to an exemplary embodiment of the present invention.
FIG. 2 is a schematic diagram illustrating a bug-corrected developer recommendation system based on a multi-developer network according to an exemplary embodiment of the present invention. Referring to FIG.
3 is a diagram illustrating an MDN based bug selection algorithm according to an embodiment of the present invention.
4 is a flowchart illustrating a bug correction developer recommendation method based on a multi-developer network according to an exemplary embodiment of the present invention.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description of the present invention, a detailed description of known functions and configurations incorporated herein will be omitted when it may make the subject matter of the present invention rather unclear. In the following description of the embodiments of the present invention, specific values are only examples.

The present invention relates to a bug-correcting developer recommendation system and method based on a multi-developer network, and more particularly, to a bug-correcting developer recommendation technique for reducing software development time and cost.

In the present invention, a new bug received may be referred to as a 'query', and a new bug report and a bug report previously stored in the bug repository may be referred to as a 'document'.

In addition, the smoothed unigram model (UM) used in the present invention is a smoothed unigram model, which is used to prevent occurrence probability of a probability vector for a specific document from becoming zero, Which means that the unigram model has been smoothed by using the collection model that measures

1 is a schematic block diagram of a bug-corrected developer recommendation system based on a multi-developer network according to an exemplary embodiment of the present invention.

Referring to FIG. 1, a bug-correcting developer recommendation system 100 based on a multi-developer network according to an exemplary embodiment of the present invention includes a preprocessing unit 110, an extracting unit 120, a modeling unit 130, ).

First, a full description of the present invention will be described in detail after a general outline of the present invention has been schematically described with reference to Fig. 2.

FIG. 2 is a schematic diagram illustrating a bug-corrected developer recommendation system based on a multi-developer network according to an exemplary embodiment of the present invention. Referring to FIG.

Referring to FIG. 2, a bug-corrected developer recommendation system based on a multi-developer network according to an exemplary embodiment of the present invention includes a step of, when a new bug report is submitted, Perform pre-processing such as word removal (a). At this time, the word token can be used as a feature in the next step. Next, the system of the present invention retrieves and extracts a bug report similar to the received new bug report from the previously stored bug report using smoothed UM and KL divergence (b). Next, the system models (c) the multi-developer network (MDN) using the activity information of the selected candidate developers (ie, the number of comments and the number of commits). Next, the system computes (d) the recommendation rank of the candidate developers and recommends the appropriate candidate to solve the new bug report. Here is a closer look.

The preprocessing unit 110 preprocesses the received new bug report. That is, upon receiving a new bug report, the preprocessing unit 110 performs a pre-processing including (or including) tokenization, stemming, stop words, etc. based on the received new bug report Natural language processing).

The following is a detailed description. The processing techniques may include the oldest and most well known natural language techniques used by data mining researchers. Strictly speaking, the preprocessing is a preliminary operation for selecting the partial pattern to be noticed at present or extracting and removing the unnecessary information by shaping the partial pattern so that essential information can be easily extracted from the observed pattern, Refers to the process of recognizing and processing a language commonly used by humans, including information retrieval, question and answer, system, automatic translation and interpretation.

Tokenization refers to the technique of separating the text data of a document into tokens after defining the documents to be indexed. In general, tokens are words, phrases, or terms. .

Stemming is a morphological analysis technique that uses tokenized information to treat parts of a word that are common to a word as a root. Removal of stop words is done by using headwords keyword, descriptor), that is, removing unnecessary elements.

That is, the preprocessing unit 110 performs preprocessing to divide the sentence of the received new bug report into word tokens and processes unnecessary roots of the word (for example, 'logging' and 'ogged' ? og "), and remove common idioms (for example, to remove abbreviations such as" He'll "or" Haven? ").

The preprocessing unit 110 may preprocess the received new bug report, vectorize the preprocessed new bug report, and the vectorized model may be expressed as smoothed UM. The description will be described in more detail later.

The extracting unit 120 extracts a bug report similar to the preprocessed new bug report in the bug report stored in the bug storage.

The bug repository contains all the information about bug reports that occurred in the past. For example, in the bug storage, information on developers who have corrected the bug report, bug report information that occurred in the past, bug information included in the bug report, correction history information of developers who corrected the bug report, and the like are stored .

In addition, the bug report stored in the bug repository basically records an error in the program source code file in various fields. The bug report includes a report of the bug report, the current status, the severity and the priority of the current bug report Important fields may be included. At this time, the component of the bug report, the modified developer (Fixer), and the priority field are influenced by the developer in order to fix the defect of the corresponding bug report.

The extracting unit 120 searches for a bug report similar to the preprocessed new bug report in the bug report previously stored in the bug storage, and then extracts the retrieved similar bug report. In order to extract a similar bug report, the extracting unit 120 may use smoothed UM to convert the bug report into a probability vector. Also, in order to calculate the similarity between the new bug report and the previously stored bug report, Kullback-Leibler divergence).

The extraction unit 120 extracts a classical vector space model (VSM) and other statistical models (e.g., LDA, Latent Dirichlet Allocation) used by most researchers in the course of IR (Information Retrieval) By using smoothed UM (smoothed unigram model), which shows better performance, you can search for bug reports similar to the received new bug report.

More specifically, it is as follows. First, the unigram of each document is expressed as a vector (i.e., a probability vector) composed of two or more random variables on the vector space.

However, if a word does not occur in one document, the probability of occurrence of the probability vector for this document becomes zero, and this phenomenon can cause problems in the probability expression. Thus, to solve this problem, the first unigram model makes the unigram model smoothed by using a collection model that measures the probability of occurrence of a particular word in the entire collection of documents.

Equation 1 defined below represents the smoothed UM of the Kth document, the second part of which expresses the collection model. In other words, the second part of Equation (1) is to use a smoothed UM model to avoid the probability of occurrence of zero if a particular word did not occur at all in the bug report, it means.

That is, Equation (1) represents smoothed UM.

는 k 리포트의 가중치 벡터(weight vector)를 의미한다.

는 k 리포트 안에 단어 수를 의미하며, ω_k(n)은 k 리포트 안에 n번째 출현 빈도를 나타낸다. K는 리포트의 총 개수를 의미하며, μ는 수학식 1에서 두 부분의 서로 다른 가중치를 의미한다.In this case,? Denotes a term,

Denotes a weight vector of the k-th report.

Denotes the number of words in the k-report, and ω _k (n) denotes the n-th occurrence frequency in the k-report. K denotes the total number of reports, and μ denotes different weights of the two parts in Equation (1).

Next, the extracting unit 120 can search for a similar bug report using KL (Kullback-Leibler divergence).

More specifically, KL divergence is a function used to compute the difference between two probability distributions, which is used to calculate the similarity between a new bug report (ie, query) and a bug report pre-stored in the bug repository. At this time, the higher the KL divergence, the lower the similarity.

That is, Equation (2) below represents a method of calculating the degree of similarity using KL divergence.

는 쿼리 q에서 용어 ω가 나타나는 확률을 의미하고,

는 버그 리포트 k에서 용어 ω가 나타나는 확률을 의미한다.At this time,

Denotes the probability that the term? Appears in the query q,

Means the probability that the term ω appears in the bug report k.

Accordingly, the extracting unit 120 uses the KLL (Kullback-Leibler divergence) between the pre-processed new bug report (smoothed UM) and the bug report previously stored in the bug repository, The report can be retrieved and extracted.

The modeling unit 130 uses the extracted activity information of the candidate developer based on the similar bug report (in this case, the activity information is activity information that the developer participated in bug correction, which means the number of comments and the number of commits) , And a multi-developer network (MDN) with a weight vector.

That is, when modeling a multi-developer network (MDN) having a weight vector, the modeling unit 130 may include, as the activity information, the number of comments indicating the activity information in which the developer participated in the bug correction, commits can be used.

In addition, the modeling unit 130 may be configured to generate the activity information of the candidate developer selected on the basis of the developer who corrected the extracted similar bug report and the developer who corrected the same report with some elements of the received new bug report , A multi-developer network (MDN) having a weight vector can be modeled. A more detailed explanation is as follows.

In the present invention, a multi-developer network (MDN) technique is proposed to recommend a more suitable bug correction developer.

A multi-developer network refers to an improved social network used to analyze the behavior of developers in one activity (eg, comment activity), while a multi-developer network (MDN) And the source code change process.

When compared to social networks, MDN considers not only the developer's comment activities, but also the communication activities they have done in the course of modifying the source code.

In order to model the MDN, the modeling unit 130 may first extract a candidate developer. At this time, since the software component field is more important than the other fields in the bug report, the modeling unit 130 corrects the similar bug report extracted by the extracting unit 120 when extracting the candidate developer Extracts developers who have fixed bug reports with components, such as received new bug reports, as well as those who have done so.

Then, the modeling unit 130 creates an intersection of two sets of developers as shown in Equation (3). That is, Equation (3) below represents the concept of the candidate developer D.

At this time, D _sim means a set of developers extracted from a similar bug report using a new bug, and D _comp means a set of developers extracted from a bug report having a new bug-like component (ie, component). Accordingly, the modeling unit 130 extracts the candidate developers D from the intersections of the two developer sets.

Next, the modeling unit 130 may extract the number of comments and the number of commits of the candidate developers D to form a list of many developers D. More specifically, The modeling unit 130 can define the MDN as follows. First, the concept will be described in detail, and then the understanding will be provided through the following embodiments.

The MDN of the software product p, a bug report submitted to the bug repository, is represented by a directional graph. This can be expressed as MDN _p = (V, E), where V is a set of developers and E is a set of relationships including comments and commits between developers participating in software product p.

In the two-dimensional space, the names of the developers can be displayed at each node of the MDN, and each edge has a weight vector.

로 정의할 수 있으며, 이하 수학식 4는 가중치 벡터의 정의를 나타낸다.The modeling unit 130 calculates a weight vector

, And Equation (4) represents the definition of the weight vector.

는 MDN의 가장자리를 의미한다.

는 버그 수정 과정에 참여한 참여자(contributors)를 의미하고,

는 배정된 개발자, 즉 담당자(assignee)를 의미한다.From here

Means the edge of the MDN.

Means contributors participating in the bug fix process,

Means the assigned developer, or assignee.

는 담당자(

)한테, 참여자(

)에 의해 변경된 커밋 수를 의미하고,

은 담당자(

)한테, 참여자(

)에 의해 달린(posted) 코멘트 수를 의미한다. 즉, MDN은, V=D이고, 각 가중치 벡터가

인 것을 포함하는, 몇가지 조건을 가지게 된다.

The person in charge (

), Participant (

), ≪ / RTI >< RTI ID = 0.0 >

The person in charge (

), Participant (

The number of comments posted by the user. That is, MDN is V = D and each weight vector is

, And so on.

3 is a diagram illustrating an MDN based bug selection algorithm according to an embodiment of the present invention.

Referring to FIG. 3, an MDN-based bug selection process according to an embodiment of the present invention is as follows. First, in Step 1, the modeling unit 130 extracts the number of comments and the number of commits of each developer. In this case, A, B, C, and D represent developers, and the vectors such as (1, 7), (3, 4) and so on displayed on the developer's edge indicate the number of comments and the number of commits by the developer.

Next, in step 2, the recommendation unit 140 compiles a ranking score of each developer. In this case, the ranking score calculates the recommendation rank of the candidate developers, and can be calculated using Equation (5). That is, Equation (5) below represents a calculation method for calculating a recommendation rank of developers (i.e., a ranking score).

Next, finally, in step 3, the recommender 140 recommends a suitable developer to correct the received new bug report based on the recommendation rank of the candidate developers calculated in the step 2.

That is, the recommendation unit 140 calculates the recommendation rank (i.e., ranking score) of the candidate developers using the multi-developer network (MDN) modeled by the modeling unit 130, You can recommend a developer to correct the received new bug report.

For example, in FIG. 3, Developer D commits 9 times in the bug report (A and 3, B and 3 and C and 3 times in detail), 15 comments for developers (5 for A, 4 times, and 6 times for C).

Accordingly, the developer D has a ranking score of "0.99 (0.33 + 0.33 + 0.33) * (0.33 + 0.27 + 0.40) = 0.99" , And D, respectively. Developer C, on the other hand, scored the second highest ranking among developers with "(0.22 * 0.40) = 0.09", ie, ranking score of 0.09.

That is, the recommendation unit 140 selects the developer D having the highest ranking score among the candidate developers A, B, C, and D based on the ranking score (i.e., the recommended ranking) As a developer to fix a new bug report, we can recommend it as a first rank.

Accordingly, by using the MDN-based bug selection algorithm when a new bug is submitted, the present invention can recommend the most suitable developer to solve a new bug.

Hereinafter, an operational flow diagram of the present invention will be briefly described based on the details described above.

4 is a flowchart illustrating a bug correction developer recommendation method based on a multi-developer network according to an exemplary embodiment of the present invention.

Referring to FIG. 4, in the multi-developer network based bug correction developer recommendation method according to the embodiment of the present invention, the preprocessing unit 110 preprocesses the received new bug report (S410).

At this time, upon receiving a new bug report, the preprocessing unit 110 performs preprocessing including tokenization, stemming, stop words, etc. based on the received new bug report do.

Also, the preprocessing unit 110 may vectorize the preprocessed new bug report, and the vectorized model may be represented by smoothed UM. This has been described in detail above and will be referred to here.

Next, in operation S420, the extraction unit 120 extracts a bug report similar to the preprocessed new bug report from the bug report stored in the bug repository.

At this time, the extracting unit 120 searches for a bug report similar to the preprocessed new bug report in the bug report stored in the bug storage, and then extracts the retrieved similar bug report. In order to extract a similar bug report, the extracting unit 120 may use the smoothed UM to convert the bug report into a probability vector, and further, to calculate the similarity between the new bug report and the previously stored bug report, Kullback-Leibler divergence). This has been described in detail above and will be referred to here.

Next, the modeling unit 130 models the multi-developer network (MDN) having the weight vector by using the activity information of the candidate developer selected on the basis of the extracted similar bug report (S430) .

At this time, when modeling the MDN, the modeling unit 130 may use, as the activity information, the number of comments and the number of commits indicating the activity information on which the developer participated in the bug correction. In addition, the modeling unit 130 may be configured to generate the activity information of the candidate developer selected on the basis of the developer who corrected the extracted similar bug report and the developer who corrected the same report with some elements of the received new bug report The MDN can be modeled. This has been described in detail above and will be referred to here.

Next, according to the present invention, the recommendation unit 140 calculates a recommendation rank (i.e., ranking score) of the candidate developer using the modeled multi-developer network, and based on the calculated recommendation rank, You can recommend a developer to fix a new bug report.

At this time, the recommendation rank of the developers (i.e., the ranking score) can be calculated with reference to the above Equation (5), and the description thereof will be described in detail hereinabove.

The present invention can recommend a more appropriate bug correction developer for correcting a large number of bugs occurring in complex software through a bug correction developer recommendation method based on a multi developer network, And the cost can be reduced.

A bug-correcting developer recommendation method based on a multi-developer network according to an exemplary embodiment of the present invention may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, and the like, alone or in combination. The program instructions recorded on the medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of computer software. Examples of computer-readable media include magnetic media such as hard disks, floppy disks and magnetic tape; optical media such as CD-ROMs and DVDs; magnetic media such as floppy disks; Magneto-optical media, and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

As described above, the present invention has been described with reference to particular embodiments, such as specific elements, and specific embodiments and drawings. However, it should be understood that the present invention is not limited to the above- And various modifications and changes may be made thereto by those skilled in the art to which the present invention pertains.

Accordingly, the spirit of the present invention should not be construed as being limited to the embodiments described, and all of the equivalents or equivalents of the claims, as well as the following claims, belong to the scope of the present invention .

100: Bug fix based on multi-developer network
110: preprocessor 120:
130: Modeling unit 140: Recommendation unit

Claims (9)

A pre-processing unit for pre-processing the received new bug report;
An extracting unit for extracting a bug report similar to the preprocessed new bug report in the bug report stored in the bug storage;
A modeling unit for modeling a multi-developer network using activity information of a candidate developer selected based on the extracted similar bug report; And
A recommendation unit for calculating a recommendation rank of the candidate developer using the modeled multi-developer network, and recommending a developer to correct the received new bug report based on the calculated recommendation rank;
A developer recommendation system that fixes a bug based on a multi - developer network. The method according to claim 1,
The extracting unit
Extracts the similar bug report by using KLB (Kullback-Leibler divergence) between the pre-processed new bug report represented by the vector and the bug report previously stored in the bug repository
Bug Correction Developer recommendation system based on multi developer network. The method according to claim 1,
The modeling unit
As the activity information, the number of comments and commits indicating the activity information in which the developer participated in the bug correction is used
Bug Correction Developer recommendation system based on multi developer network. The method according to claim 1,
The modeling unit
The developer who has corrected the extracted similar bug report and the activity information of the candidate developer selected based on the developer who has corrected the same report by a part of the received new bug report
Bug Correction Developer recommendation system based on multi developer network. Pre-processing the received new bug report;
Extracting a bug report similar to the preprocessed new bug report in a bug report previously stored in the bug repository;
Modeling a multi-developer network using activity information of a candidate developer selected based on the extracted similar bug report; And
Computing a recommendation rank of the candidate developer using the modeled multi-developer network, and recommending a developer to correct the received new bug report based on the calculated recommendation rank;
A developer recommendation method based on a multi-developer network that includes bug fixes. 6. The method of claim 5,
The extracting step
Extracts the similar bug report by using KLB (Kullback-Leibler divergence) between the pre-processed new bug report represented by the vector and the bug report previously stored in the bug repository
Developer recommendation method based on multi - developer network. 6. The method of claim 5,
The modeling step
As the activity information, the number of comments and commits indicating the activity information in which the developer participated in the bug correction is used
Developer recommendation method based on multi - developer network. 6. The method of claim 5,
The modeling step
The developer who has corrected the extracted similar bug report and the activity information of the candidate developer selected based on the developer who has corrected the same report by a part of the received new bug report
Developer recommendation method based on multi - developer network. A computer-readable recording medium having recorded thereon a program for executing the method according to any one of claims 5 to 8.

Images