The literature indicates that, in general, TAs lack teaching experience, and many of them need appropriate pedagogical training, particularly because they need more knowhow on feedback resources, as well as in providing students with adaptive feedback. Given the mentioned perspective on feedback in introductory programming, one important question is: how can TAs assist in providing adaptive feedback that meets students' needs? To address this issue, guidelines can be provided to clarify what effective feedback is for novice student programmers. However, there is still a lack of guiding elements that constitute adaptive feedback. Thus, to mitigate these problems, we propose a taxonomy for providing adaptive feedback, covering different approaches and feedback resources in problem-solving by novice programmers. The body of knowledge represented in the proposed taxonomy is helpful to be presented to TAs, who should use it to guide their pedagogical decision-making process in relation to adaptive feedback in terms of synthesis and analysis. To our knowledge, a taxonomy for this purpose has yet to be published in the computing education literature. In summary, we intended to investigate the following research questions: (1) What can be adapted in the feedback provided to the beginning programming student during problem-solving? (2) What can be adapted from the feedback provided to the novice programmer during problem-solving? As a method for designing the proposed taxonomy, we followed the Design Science Research (DSR) methodology using its main stages: problem identification, elaboration, testing, and evaluation. As a result, we develop a novel comprehensive taxonomy of adaptive feedback that captures the many elements that one teacher might consider when helping a given student. The taxonomy includes various approaches and features to adaptive feedback, expressed in categories, subcategories, and contextual elements. The main contribution of this paper is to provide a taxonomy that can help TAs use a relevant pedagogical body of knowledge on how to provide adaptive feedback for programming by novices.

In this article, we investigate the influence of variables related to gamification strategies promoted by an online learning environment on the “gaming the system” behavior presented by elementary school students in mathematics. To this end, we performed our research in four main stages: exploratory database analysis, database annotation, detector development, and analysis of the results. We developed a data-driven automatic detector involving the training of single and joint machine learning algorithms to classify “gaming the system” behavior and obtain models with the best accuracy, including exploring, evaluating, and comparing different algorithms. The detector's development involved training machine learning algorithms as a single category, including decision tree, K-NN, neural network, and Support Vector Machine and ensemble categories, such as Random Forest, Gradient Boosting Machine, Adaptive Boosting, and XGBoost. In addition, three human tutors annotated the database with log data of problem-solving from the math domain for elementary school students, who evaluated each student's problem-solving attempt in game and non-game classes. In total, 12,514071 rows of data were labeled in two rounds of labeling, with Cohen's Kappa of 0.82 for inter-rater reliability for differentiating between gaming and non-gaming behavior. Our findings suggest that gamification strategies used by online learning environments can provoke “gaming the system” behavior in elementary school students. At the same time, we provide evidence from the literature and exploratory data analysis that the incentive to seek performance over learning has motivated students to manifest non-learning behaviors. We hope these insights will contribute to improving the planning and development of a gamification strategy in an online learning environment.

In this paper, we present a machine learning approach to classify erasures in children's manuscript essays. We aim at highlighting challenges faced by children through their learning process, specifically when it comes to their subjective ability of converting thoughts into written text. The image processing software receives an entire scanned essay as input, then it segments each word. Afterwards, a classifier decides whether each word is "clean" or if it corresponds to an "erasure". We evaluate several Machine Learning (ML) techniques, including Support Vector Machine (SVM), boosted and bagged decision trees, k-nearest neighbor, Naive Bayes, discriminant analysis,logistic regression, and deep neural networks. To this end, manuscript essays containing erasures were scanned, each word(with and without erasure) was segmented, and were separated into two classes, providing knowledge to the system for the identification of new samples. Due to the difference between image dimensions, it was necessary to use features capable of describing the image independently of it corresponding to a letter or a word, or character dimension used by different children during writing. To solve this problem, we identify strokes directions, usually consisting of parallel lines. The obtained results show that the image descriptors when used alone achieve maximum accuracy limits of the order of 90%, but combined with the others, they exceed 98% accuracy. Compared to other used classifiers, SVM with Gaussian kernel produced higher accuracy.

Code smells indicate poor implementation choices that may hinder program comprehension and maintenance. Their informal definition allows developers to follow different heuristics to detect smells in their projects. Machine learning has been used to customize smell detection according to the developer's perception. However, such customization is not guided (i.e. constrained) to consider alternative heuristics used by developers when detecting smells. As a result, their customization might not be efficient, requiring a considerable effort to reach high effectiveness. In fact, there is no empirical knowledge yet about the efficiency of such unguided approaches for supporting developer-sensitive smell detection. This paper presents Histrategy, a guided customization technique to improve the efficiency on smell detection. Histrategy considers a limited set of detection strategies, produced from different detection heuristics, as input of a customization process. The output of the customization process consists of a detection strategy tailored to each developer. The technique was evaluated in an experimental study with 48 developers and four types of code smells. The results showed that Histrategy is able to outperform six widely adopted machine learning algorithms - used in unguided approaches - both in effectiveness and efficiency. It was also confirmed that most developers benefit from using alternative heuristics to: (i) build their tailored detection strategies, and (ii) achieve efficient smell detection.

In this paper, we propose a Web-based interactive learning environment with a knowledge-based assistant for helping undergraduate students in solving problems on designing digital circuits, focusing the relationship between Boolean algebra and Digital Logic. Particularly, we have addressed problem solving situations that essentially consist of mapping a logic expression into another equivalent, for example, the task of simplifying digital circuits. To do that, we have considered to operate on two symbolic linked representations: Boolean Algebra and Logic, connecting these two representations to a visual one with logic gates. This integration comes from the importance to involve students in three equivalent representations, permitting them to make connections between them. The learning environment contains a knowledge-based assistant that works on a collection of scenarios with several associated problems, where the students are placed in a problem-solving situation. To each problem, there are pedagogical resources needed to provide the students with feedback and knowledge support based on their performance and identified needs. Furthermore, there is an adaptive mechanism that is responsible for how problems should be sequenced or ordered, and even we have designed a switcher mechanism, which permits the students to change from one to another representation. We evaluate our approach by using scenarios and conclude that the system is feasible.

This paper presents a framework for building mechanisms for automatic analysis of introductory programming exercises. This framework consists of a structure that provides the combination of different types of analysis mechanisms, providing the configuration of various families of analyzers. One of these analyzers is based on a tree inclusion mechanism that verifies whether a particular structure (i.e. combination of language constructs) is contained in a solution. For evaluating our proposal, we conduct a study with 232 submissions from two introductory programming classes. The results indicate the following groups of students: one group that failed on syntactic and functional correctness stages; eight groups that submitted solutions with similar structure. Finally, the results provide evidences that is possible to identify different templates of correct solutions.

Developers continuously maintain software systems to adapt to new requirements and to fix bugs. Due to the complexity of maintenance tasks and the time-to-market, developers make poor implementation choices, also known as code smells. Studies indicate that code smells hinder comprehensibility, and possibly increase change- and fault-proneness. Therefore, they must be identified to enable the application of corrections. The challenge is that the inaccurate definitions of code smells make developers disagree whether a piece of code is a smell or not, consequently, making difficult creation of a universal detection solution able to recognize smells in different software projects. Several works have been proposed to identify code smells but they still report inaccurate results and use techniques that do not present to developers a comprehensive explanation how these results have been obtained. In this experimental report we study the effectiveness of the Decision Tree algorithm to recognize code smells. For this, it was applied in a dataset containing 4 open source projects and the results were compared with the manual oracle, with existing detection approaches and with other machine learning algorithms. The results showed that the approach was able to effectively learn rules for the detection of the code smells studied. The results were even better when genetic algorithms are used to pre-select the metrics to use.

In this paper, we propose an agent-based ITS for learning propositional logic using two linked viewpoints and representations to address the problem on checking the validation of a given argument, articulating syntactic and semantic inference approaches. One motivation for this integration comes from the importance to involve students in two complementary viewpoints, permitting students to make connections between the two viewpoints involving different strategies, used in problem solving situations. Basically, the system works by placing students in a problem-solving situation and providing needed knowledge support based on his performance. We discuss the system and show its architecture and functionality, as well as its implementation details. Through scenarios with adequate coverage, we evaluate the value of the proposed approach and conclude that it is feasible and useful.

The current research is situated on context of teacher support in Virtual Learning Environments (VLE), specifically related to decision-making process. The knowledge about student performance is fundamental to solve pedagogical challenges and to define different ways in teaching and learning. Following this trend, VLE provides extra information by reports, graphs and interface's alert. However, there is demand of statistical tools to support the teacher decisions, because the learning analysis is limited to frequency analysis. This paper discusses an analytical approach to deal with e-learning data. Our focus is to identify groups of learners based on their answers. Therefore, this paper pointed out some main objectives: understand profiles of answers in order to guide a student to future learning activities, and identify which criteria, in each group, is the most relevant for the tutor's help. Several techniques are useful for e-learning issues. However, we focus in Educational Data Mining (EDM) methodology. Our paper selected data of an English e-learning course from PSLC repository for case study in validation step. Preprocessing techniques of EDM were applied on selected dataset. Initially, we removed incomplete, noisy and inconsistent data of sample. After the preprocessing, we executed two steps: clustering and prediction analysis. Firstly, we executed the clustering process, because we needed to identify student's group based on their answers. After understanding the groups, we predicted behaviors of students on each cluster, which were defined in the last step. In our research, it defines that the prediction will use a regression methodology, and the clustering will execute K-means algorithm. The study identified five student's groups based on their answer, such as Expert, Good, Regular, Bad and Criticism answers. Consequently, the prediction analysis defined that the score of tutor's help ("Avg Assistance Score") is the most interesting factor for our investigation. The approach executed the Stepwise Backward Regression, which is a semi-automated process of building a model by successively adding or removing variables based solely on the t-statistics of their estimated coefficients. Thus, other result is that the presence of the variables "Incorrect" and "Correct First Attempts" belongs to three regression models obtained by the approach. This work's findings presents knowledge about answers profiles of VLE students in two main perspectives. First, it analyzes the usage of Open Learning Data to characterize behavioral profiles of answers using multivariate analysis techniques. Second, our analysis contributes to expand present knowledge about how student performance changes the teacher decisions in VLE. This approach tends to be a useful tool in analytical process when the VLE system does not provide statistical tools.

Essentially social query consists in sharing a question on a social network and waiting for responses. The usual strategy of sharing a public question is inefficient because it does not ensure an answer, either quality or agility. We assume that directing the question to a specific user is the best way to get a right and quick answer. However, choosing this person could be a hard task and by choosing a wrong one we could receive an incorrect answer or the person might just ignore the question. In this paper, we proposed, implemented and evaluated a model based on a multi-criteria approach to find the most suitable user to receive a question on Twitter. We rank expert candidates using the Weight Product Model. One of the main differentials of our model is the use of multi-criteria and that it was built for a social network context. In the evaluation, using as metrics the Normalized Discounted Cumulative Gain and the Spearman Rank Correlation Coefficient, our technique achieved promising results in predicting the usefulness of real responders answer.

This paper presents a virtual environment which enables people interested in acquiring and sharing knowledge about Brazilian popular music to congregate in communities. The environment includes a music virtual library and a cooperative learning system based on the harmony trees theory, an innovative harmonic teaching method which has been used successfully by its author for several years.

This paper focus on results related to the design of a model of a Computer-based Interactive Learning Environment oriented to the Petri net domain, named PN-M. This environment adopt a a multi-agent approach based on a view about the domain knowledge structure as defined in the MATHEMA interactive learning environment. PN-M is focused on problem solving. Hence in this paper we introduce the domain of the problem, that is the design of interlocking Petri models for track vehicle systems.