In software development, there are numerous reasons why a development organization changes. These include cost reduction and retiring developers. Often the quality of software developed by multiple organizations is low, leading to problems such as difficulty in the evolution of source code. To detect factors indicating low evolvability of software, we focus two maintainability characteristics of software development: defect proneness and refactoring tendencies of the source code. We investigate the changes in the relationship of software metrics and the number of defects. Then we compare the results and editing tendencies for refactoring and evolving software in experiments using OpenOffice.org, which is developed by multiple companies, and a tool developed by outsourcing. Defects tend to concentrate on files that depend on many other files. Additionally, the code review confirms that refactoring before and after a development organization change is important for future developments.
Online community networks provide a variety of interaction mechanisms, many of which focus on the confirmation of functional roles or the validation of user experience results. This study is concerned with the increasing emphasis on the publicity of community media among colleges and universities. Therefore, this study adopts the fan pages of university institutions as the object of research, in order to investigate the relationships between different types of information formats (links, photos, and videos) and the emotions and behaviors of their fans. Using stimulus-organism-response theory, a model of the relationship between information format, emotional response, and behavioral response is tested. Data were collected from the twenty largest university fan pages in the USA in this study. A total of 13,006 posts and user comments are collected, compared, and analyzed. Findings validate an interactive correlation between user behavior and information format. Notably, university fan pages with larger fan bases are more capable of applying information formats to prompt user interaction and users' willingness to comment and share, consequently expanding the organic reach of their posts. These results reiterate that users' emotions significantly affect their behaviors and that information serves as a key function in user experiences. Effectively triggering user emotions has become the primary means for community administrators to produce viral posts.
An innovative engineer who can address a social challenge using big data processing, AI and cloud computing technologies with the generation of new business and value is required from industry. enPiT is an education project to develop the advanced IT engineer based on practical education in cooperation between industry and academia promoted by Ministry of Education, Culture, Sports, Science and Technology (MEXT) of Japan. In this paper, we introduce how we designed PBL centered curriculum named AiBiC Spiral under the framework of enPiT education project, and analyze educational effect of our program based on work products and the questionnaire result targeting students who took and completed to the program of 2017.
This paper proposes extended association rule mining that can deal with correlation functions. The extended association rule is expressed in the form of: A ⇒ Correl(X, Y ) where Correl(X, Y ) is a correlation function with two variables X and Y . By this extension, data analysts can discover the condition A that lead to low (or high) correlation between two given variables from a large dataset. In order to show the efficacy of the proposed method, a case study is performed on an industry dataset of software developments, assuming the scenario of discovering a condition, where software development effort is predictable (or unpredictable) from the size of the project, i.e. there exists a significantly high (or low) correlation between size and effort. Since such a condition cannot be obtained by conventional association rule mining, we confirm the efficiency of the proposed extended association rule mining.
In real-world applications such as fraud detection, target class values have an unequal size. This problem is called class-imbalanced data. Many strategies have been proposed to deal with this situation. Most of them focused on changing the data characteristics. For example, adjusting class distribution is one of the most popular approaches for this matter. In this work, we proposed TOP (TwO-levels of Positive resampling framework), an alternative framework to resolve such a problem. Our technique exploits DBSCAN mechanism and other resampling algorithms in order to maximize classification performance. It is able to dynamically draw two boundaries that represent similarity level between consideration positive and other instances. Many possible resampling techniques such as undersampling or over-sampling are allowed to perform inside those areas. We benchmarked TOP with three types of resampling techniques including over-sampling, down-sampling, and hybrid sampling by training eleven machine learning algorithms on fifteen datasets. As a result, our technique outperformed other techniques in several evaluation metrics.
According to the spread of the Twitter service, spoofing accounts of famous people (i.e., a malicious user makes an account to pretend famous people) are increasing. To prevent such spoofing, Twitter, Inc. provides verified accounts that are guaranteed as genuine accounts of themselves by Twitter, Inc. However, not all famous people use the verified accounts. Instead of that, famous people sometimes embed the link of their twitter accounts on their official website. But their Twitter accounts are not always embedded on that. So, it is not always easy to distinguish whether the accounts are genuine or spoofing. To identify spoofing Twitter accounts, we propose a new discrimination method focusing on the following functions. The basic idea of the method is based on the PageRank citation ranking. An important assumption of our method is that famous people accounts often have many followers, but not have many follows. In the experiment, we applied our method to 27 genuine accounts and 37 spoofing accounts. As a result, 96% of genuine accounts and 75% of spoofing accounts are identified correctly.
Development of social media has brought a huge change to social communities in several aspects. They offer a place where social media users can post information, express opinions, and share interests. However, some information and opinions may cause a negative impact on the person mentioned in the post and that person can become a target of defamation. In Thailand, although defaming someone on social media is illegal, most social media users are not aware of it. To raise awareness of this issue, this paper proposes the classification of defamatory text in Thai language. Several approaches to text classification are used to analyze textual comments to political news and articles on Facebook, including word n-grams, character ngrams, specific terms, grammatical dependency structure, and sentiment polarity. The experiment is conducted using two machine learning methods with several combination of the approaches. The result shows that SVM performed better than Naïve Bayes, and word n-grams and character n-grams are more efficient than other approaches with F score of 0.64 and accuracy of 0.74. In addition, dependency structure, specific terms, and sentiment polarity perform quite well with precision of 0.65 and accuracy of 0.66, but with lower recall rate of 0.35. We discuss linguistic variations in Thai language which affect the performance of the methods.
k-dimensional (k-d) trees are one of the most important data structures in the fields of data engineering and so-called Big Data. In this paper we propose a scheme parallelizing the construction of a k-d tree. Since efficient presorting is required for constructing a balanced k-d tree, we also developed a parallelized heapsort algorithm. The proposed scheme is 3.59 times faster than sequential construction of a k-d tree.
In this paper, focusing on the problem that estimation accuracy of software development effort greatly varies among software projects, we propose a predictability classification method for software projects before conducting effort estimation. In the proposed method, given a project to be estimated, we first evaluate whether the effort can be accurately estimated or not by identifying the project as "predictable" or "unpredictable". In case of predictable projects, we conduct the effort estimation. Otherwise, estimation is avoided. As a result of an experiment to assess the effectiveness of the proposed method using six industry datasets, (i) the mean square residual and residual variance are shown to be suitable measures for recognition of predictability; and (ii) the average absolute error is significantly reduced in five datasets, by avoiding the estimation when a project belongs to the unpredictable class, which proves the effectiveness of the proposed method. By using the proposed method, practitioners become aware of cases when they can rely on the estimation and when they cannot.
The purpose of this paper is to offer a better under- standing of the rapid evolution of technology, with respect to the Energy Informatics field, Machine Learning and Deep Learning. Several data analysis and cleaning techniques as well as forecasting models have been proven to be highly effective in many research areas and will be presented this work. Moreover, challenges, of different nature, are encountered in the research process itself as well as afterwards, related to data availability, technology gaps or ethics. Scientific research employs specialists in various fields of science, needs an infrastructure and a legal organizational framework for carrying out the activity. There is a huge need for data scientists and data analyzing and processing techniques due to the rapid modernization of the world and the increasing amount of generated data. In this work, Energy Load Forecasting models and challenges will be highlighted.
Open Source Software (OSS) does not work without contributions from the community. In particular, long-term contributors (LTCs) (e.g., committer), defined as contributors who spend at least one year on OSS projects, play a crucial role in a project success because they would have permission to add (commit) code changes to a project’s version control system, and to become a mentor for a beginner in OSS projects. However, contributors often leave a project before becoming a LTC because most contributors are volunteers. If contributors are motivated in their work in OSS projects, they might not leave the projects. In this study, we examine the phenomena involved in becoming a LTC in terms of motivation to continue in OSS projects. In particular, our target motivation is to understand what is involved in long-term contribution with other expert contributors. We study classifier to identify a LTC who will contribute patch submissions for more than one year based on collaboration in terms of the code review process. In detail, we analyze what review feedbacks encourage a contributor to continue with OSS project. Using a Qt project dataset, we build a prediction model to identify a LTC. We find that not only contributor’s activities, but also a reviewer feedbacks, useful in identifying LTCs.
In open-source software (OSS) ecosystems, software popularity is valuable information to developers because they continually want to know whether their software is attracting and gaining acceptance. Meanwhile, software popularity is also an important indicator to suggest if a software is beating its competitors in an OSS ecosystem. Accordingly, it is important to understand the popularity growth of packages (i.e., how fast packages become popular). In this paper, we conduct an exploratory study on packages in the node package manager (npm) to understand: (1) the characteristics of popularity growth, and (2) the factors that could affect popularity growth. We propose a method to model popularity growth as a curve and find that popularity growth mathematically follows three models - accelerated growth model (i.e., quadratic model), steady growth model (i.e., linear model), and decelerated growth model (i.e., square root model). The results show that 51.56% of the studied packages depict steady growth model, followed by accelerated growth model and decelerated growth model, 40.02% and 7.20% respectively. Furthermore, we reveal that factors including age, dependents, new features and functionalities have impacts on popularity growth. Our study shows potential tips for helping practitioners on developing and evolving packages in a competitive OSS ecosystem.
There were five problems of the discriminant analysis. The first problem (Problem1) was the defect of a number of misclassifications (NM) that caused many problems. We found the relation of NM and the coefficients of linear discriminant functions (LDFs) firstly. Moreover, we proposed the minimum NM (MNM) instead of NM. MNM is an important statistic in the discriminant analysis. Problem2 was that there was no study about the discrimination of linearly separable data (LSD). Although Vapnik proposed a hard-margin Support Vector Machine (H-SVM) that defined LSD clearly, nobody investigated the LSD-discrimination. Let us MNMk is MNM of the k-variable model, and MNM(k+1) is MNM of (k+1)-variable model to add one variable to the k-variable model, MNMk is larger than equal MNM(k+1) because the k-variable model is a subset of (k+1)-variable model (MNM monotonic decrease). In addition to this fact, if MNMk = 0, all models including the k-variable are LSD. We call LSD as Matryoshka in cancer gene analysis. If the full model with p variables is big Matryoshka, it contains many smaller Matryoshkas nested within it due to the MNM monotonic decrease. We showed the error rates of statistical discriminant functions based on the variance-covariance matrices are very high for LSD. It was impossible to judge pass / fail using the exam score correctly. If there are two testlets such as T1 and T2 and the passing point is 50 points, f = T1 + T2 - 50 is an obvious LDF, and it's MNM = 0. However, the error rate of Fisher's LDF exceeds 30% because the exam data does not satisfy Fisher's assumption and many successful candidates near the discriminant hyperplane are misclassified. Problem3 is the defect of the generalized inverse matrix technique. Problem4 is that discriminant analysis is not the traditional inferential statistics because there are no standard errors of error rates and discriminant coefficients. We developed four Optimal LDFs (OLDFs) and the 100-fold cross validation for a small sample (Method1). We proposed the simple and powerful model selection method such as the best model that has the minimum mean of error rates among all possible models of the validation samples (M2). We showed the M2 of Revised IP-OLDF (RIP) based on MNM criterion was better than other LDFs by six different types of ordinary data. On October 28, 2015, we obtained six microarrays and realized we forgot Problem5. Many statisticians could not succeed in cancer gene analysis by microarrays. They pointed out three difficulties or excuses such as 1) small n, large p data, 2) NP-hard, 3) Difficulty in separating signal and noise. However, no studies could find microarrays were LSD. We developed the Matryoshka Feature Selection Method (Method2) and could decompose microarrays into plural small Matryoshkas (SMs) and the noise subspace. That is, Method2 could easily solve three difficulties. Because all MNMs of SMs are zero and those are small samples, we can analyze these SMs by standard statistical methods very easy. However, only logistic regression can discriminate all SM accurate as same as RIP and H-SVM. Other standard statistical methods cannot accurately separate the two classes such as cancer and healthy subjects. After many trials, we create new data using RIP discriminant score of all SM instead of genes. With this breakthrough advance, we can propose an oncogene diagnosis. In this paper, we introduce the reason why Problem5 was not solved from 1970.
Social network services (SNS) are more and more popular. People are increasingly accustomed to express their opinions on SNS in two ways: (1) product reviews in online shopping sites and (2) posts, comments, tweets, and chats in social network sites. SNS text classification is challenging due to various natural language phenomena, such as spelling mistakes and variations, polysemy, contextual ambiguity and semantic variations. In this paper, we propose a novel deep learning architecture called Hybrid two Convolutional Neural Networks and Bidirectional LSTM (H2CBi) which combines the strength of both Convolutional Neural Networks (CNNs) and Bidirectional LSTM (BLSTM). We use two CNNs for extracting different positive/negative or bullied/no bullied features from SNS data. We use BLSTM to produce a sentence-level representation by maintaining the order of words and get the ability to learn sequential correlations for the sentence which is the negative sentence without having any negative word. We used two kinds of SNS data in this paper: product review data (Amazon, Movie Review and Yelp) and (2) social network sites data (Twitter I, Twitter II, Facebook and FormSpring.me). Some of our H2CBi models, namely 2WH2CBi, WFH2CBi and WGH2CBi have better performance than baseline models in six out of the seven SNS datasets in terms of accuracy and F measure.
An on-line mining of a data stream consisting of huge transactions is still a quite difficult computation problem because of an explosion of frequent itemsets. In this paper, we study a problem for mining the closed itemsets greater than or equal to size K. We give an on-line approximation algorithm, which is based on incremental intersection of transactions and rejection of closed itemsets less than size K. A closed itemset greater than or equal to size K has an intermediate property between a maximal itemset and an ordinary closed itemset. We use such a framework in order to partially solve an explosion problem in closed itemset mining. We study some theoretical properties, such as an incompleteness for frequent closed itemsets and a completeness of frequent itemsets of the on-line approximation algorithm. We also show preliminary experimental results for evaluating the performance of the proposed on-line algorithm.
Software maintenance is an important activity on the software lifecycle. In this study, we try to establish a benchmark of software maintenance cost. To establish the benchmark, factors affecting work efficiency and unit cost should be clarified, using a dataset collected from various organizations (cross-company dataset). We used dataset includes 837 data points collected by Economic Research Association from 2006 to 2016, and analyzed factors affecting work efficiency of software maintenance and unit cost of engineers. In the analysis, we defined two types of work efficiency. Also, as unit cost of engineers, we defined two types of unit cost. As attributes related to work efficiency, we analyzed address to process improvement, business sector, and required availability rate. As attributes related to unit cost, we analyzed address to process improvement, business sector, and social impact of faults. First, we showed correlation ratio of attributes to work efficiency and unit cost. Then, we analyzed each attributes using boxplots. As a result, business sectors related to both work efficiency and unit cost. The boxplots are useful to estimate software maintenance cost roughly.
To support successful quality managements of open source software (OSS) projects, this paper proposes to measure the balance of developers’ contributions to a source file as an entropy. Through an analysis of data collected from 10 popular OSS projects, the following trends are reported: a source file is more fault-prone as the developers’ contributions to the file are more imbalanced (lower entropy), and the proposed metric can be useful for predicting fault-prone programs.
In manufacture, steel and other metals are mainly cut and shaped during the fabrication process by computer numerical control (CNC) machines. To keep high productivity and efficiency of the fabrication process, engineers need to monitor the real-time process of CNC machines, and the lifetime management of machine tools. In a real manufacturing process, breakage of machine tools usually happens without any indication, this problem seriously affects the fabrication process for many years. Previous studies suggested many different approaches for monitoring and detecting the breakage of machine tools. However, there still exists a big gap between academic experiments and the complex real fabrication processes such as the high demands of real-time detections, the difficulty in data acquisition and transmission. In this work, we use the spindle current approach to detect the breakage of machine tools, which has the high performance of real-time monitoring, low cost, and easy to install. We analyze the features of the current of a milling machine spindle through tools wearing processes, and then we predict the status of tool breakage by a convolutional neural network(CNN). In addition, we use a BP neural network to understand the reliability of the CNN. The results show that our CNN approach can detect tool breakage with an accuracy of 93%, while the best performance of BP is 80%.
We propose a probabilistic key predistribution scheme for wireless sensor networks, where keying materials are distributed to sensor nodes for secure communication. We use a two-tier approach in which there are two types of nodes: regular nodes and agent nodes. Agent nodes are more capable than regular nodes. Our node deployment model is zone-based such that the nodes that may end up with closer positions on ground are grouped together. The keying material of nodes that belong to different zones is non-overlapping. However, it is still possible for nodes that belong to different zones to communicate with each other via agent nodes when needed. We give a comparative analysis of our scheme through simulations and show that our scheme provides good connectivity figures at reasonable communication cost. Most importantly, simulation results show that our scheme is highly resilient to node captures.
