Mitchell Joblin

Tong Liu, Yushan Liu, Marcel Hildebrandt et al. · bytedance

Graphs can model real-world, complex systems by representing entities and their interactions in terms of nodes and edges. To better exploit the graph structure, graph neural networks have been developed, which learn entity and edge embeddings for tasks such as node classification and link prediction. These models achieve good performance with respect to accuracy, but the confidence scores associated with the predictions might not be calibrated. That means that the scores might not reflect the ground-truth probabilities of the predicted events, which would be especially important for safety-critical applications. Even though graph neural networks are used for a wide range of tasks, the calibration thereof has not been sufficiently explored yet. We investigate the calibration of graph neural networks for node classification, study the effect of existing post-processing calibration methods, and analyze the influence of model capacity, graph density, and a new loss function on calibration. Further, we propose a topology-aware calibration method that takes the neighboring nodes into account and yields improved calibration compared to baseline methods.

Wolfgang Mauerer, Mitchell Joblin, Damian A. Tamburri et al.

We report on a large-scale empirical study investigating the relevance of socio-technical congruence over key basic software quality metrics, namely, bugs and churn. In particular, we explore whether alignment or misalignment of social communication structures and technical dependencies in large software projects influences software quality. To this end, we have defined a quantitative and operational notion of socio-technical congruence, which we call socio-technical motif congruence (STMC). STMC is a measure of the degree to which developers working on the same file or on two related files, need to communicate. As socio-technical congruence is a complex and multi-faceted phenomenon, the interpretability of the results is one of our main concerns, so we have employed a careful mixed-methods statistical analysis. In particular, we provide analyses with similar techniques as employed by seminal work in the field to ensure comparability of our results with the existing body of work. The major result of our study, based on an analysis of 25 large open-source projects, is that STMC is not related to project quality measures -- software bugs and churn -- in any temporal scenario. That is, we find no statistical relationship between the alignment of developer tasks and developer communications on the one hand, and project outcomes on the other hand. We conclude that, wherefore congruence does matter as literature shows, then its measurable effect lies elsewhere.

Mitchell Joblin, Sven Apel, Claus Hunsen et al.

Knowledge about the roles developers play in a software project is crucial to understanding the project's collaborative dynamics. Developers are often classified according to the dichotomy of core and peripheral roles. Typically, operationalizations based on simple counts of developer activities (e.g., number of commits) are used for this purpose, but there is concern regarding their validity and ability to elicit meaningful insights. To shed light on this issue, we investigate whether commonly used operationalizations of core--peripheral roles produce consistent results, and we validate them with respect to developers' perceptions by surveying 166 developers. Improving over the state of the art, we propose a relational perspective on developer roles, using developer networks to model the organizational structure, and by examining core--peripheral roles in terms of developers' positions and stability within the organizational structure. In a study of 10 substantial open-source projects, we found that the existing and our proposed core--peripheral operationalizations are largely consistent and valid. Furthermore, we demonstrate that a relational perspective can reveal further meaningful insights, such as that core developers exhibit high positional stability, upper positions in the hierarchy, and high levels of coordination with other core developers.

Mitchell Joblin, Sven Apel, Wolfgang Mauerer

Software evolution is a fundamental process that transcends the realm of technical artifacts and permeates the entire organizational structure of a software project. By means of a longitudinal empirical study of 18 large open-source projects, we examine and discuss the evolutionary principles that govern the coordination of developers. By applying a network-analytic approach, we found that the implicit and self-organizing structure of developer coordination is ubiquitously described by non-random organizational principles that defy conventional software-engineering wisdom. In particular, we found that: (a) developers form scale-free networks, in which the majority of coordination requirements arise among an extremely small number of developers, (b) developers tend to accumulate coordination requirements with more and more developers over time, presumably limited by an upper bound, and (c) initially developers are hierarchically arranged, but over time, form a hybrid structure, in which core developers are hierarchically arranged and peripheral developers are not. Our results suggest that the organizational structure of large projects is constrained to evolve towards a state that balances the costs and benefits of developer coordination, and the mechanisms used to achieve this state depend on the project's scale.

Yushan Liu, Yunpu Ma, Marcel Hildebrandt et al.

Conventional static knowledge graphs model entities in relational data as nodes, connected by edges of specific relation types. However, information and knowledge evolve continuously, and temporal dynamics emerge, which are expected to influence future situations. In temporal knowledge graphs, time information is integrated into the graph by equipping each edge with a timestamp or a time range. Embedding-based methods have been introduced for link prediction on temporal knowledge graphs, but they mostly lack explainability and comprehensible reasoning chains. Particularly, they are usually not designed to deal with link forecasting -- event prediction involving future timestamps. We address the task of link forecasting on temporal knowledge graphs and introduce TLogic, an explainable framework that is based on temporal logical rules extracted via temporal random walks. We compare TLogic with state-of-the-art baselines on three benchmark datasets and show better overall performance while our method also provides explanations that preserve time consistency. Furthermore, in contrast to most state-of-the-art embedding-based methods, TLogic works well in the inductive setting where already learned rules are transferred to related datasets with a common vocabulary.

Anna Himmelhuber, Stephan Grimm, Sonja Zillner et al.

Similarly to other connectionist models, Graph Neural Networks (GNNs) lack transparency in their decision-making. A number of sub-symbolic approaches have been developed to provide insights into the GNN decision making process. These are first important steps on the way to explainability, but the generated explanations are often hard to understand for users that are not AI experts. To overcome this problem, we introduce a conceptual approach combining sub-symbolic and symbolic methods for human-centric explanations, that incorporate domain knowledge and causality. We furthermore introduce the notion of fidelity as a metric for evaluating how close the explanation is to the GNN's internal decision making process. The evaluation with a chemical dataset and ontology shows the explanatory value and reliability of our method.

Anna Himmelhuber, Mitchell Joblin, Martin Ringsquandl et al.

Graph neural networks (GNNs) are quickly becoming the standard approach for learning on graph structured data across several domains, but they lack transparency in their decision-making. Several perturbation-based approaches have been developed to provide insights into the decision making process of GNNs. As this is an early research area, the methods and data used to evaluate the generated explanations lack maturity. We explore these existing approaches and identify common pitfalls in three main areas: (1) synthetic data generation process, (2) evaluation metrics, and (3) the final presentation of the explanation. For this purpose, we perform an empirical study to explore these pitfalls along with their unintended consequences and propose remedies to mitigate their effects.

Aneta Koleva, Martin Ringsquandl, Mitchell Joblin et al.

High-quality Web tables are rich sources of information that can be used to populate Knowledge Graphs (KG). The focus of this paper is an evaluation of methods for table-to-class annotation, which is a sub-task of Table Interpretation (TI). We provide a formal definition for table classification as a machine learning task. We propose an experimental setup and we evaluate 5 fundamentally different approaches to find the best method for generating vector table representations. Our findings indicate that although transfer learning methods achieve high F1 score on the table classification task, dedicated table encoding models are a promising direction as they appear to capture richer semantics.

Martin Ringsquandl, Houssem Sellami, Marcel Hildebrandt et al.

The application of graph neural networks (GNNs) to the domain of electrical power grids has high potential impact on smart grid monitoring. Even though there is a natural correspondence of power flow to message-passing in GNNs, their performance on power grids is not well-understood. We argue that there is a gap between GNN research driven by benchmarks which contain graphs that differ from power grids in several important aspects. Additionally, inductive learning of GNNs across multiple power grid topologies has not been explored with real-world data. We address this gap by means of (i) defining power grid graph datasets in inductive settings, (ii) an exploratory analysis of graph properties, and (iii) an empirical study of the concrete learning task of state estimation on real-world power grids. Our results show that GNNs are more robust to noise with up to 400% lower error compared to baselines. Furthermore, due to the unique properties of electrical grids, we do not observe the well known over-smoothing phenomenon of GNNs and find the best performing models to be exceptionally deep with up to 13 layers. This is in stark contrast to existing benchmark datasets where the consensus is that 2 to 3 layer GNNs perform best. Our results demonstrate that a key challenge in this domain is to effectively handle long-range dependence.

Christof Tinnes, Timo Kehrer, Mitchell Joblin et al.

Model transformations play a fundamental role in model-driven software development. They can be used to solve or support central tasks, such as creating models, handling model co-evolution, and model merging. In the past, various (semi-)automatic approaches have been proposed to derive model transformations from meta-models or from examples. These approaches require time-consuming handcrafting or recording of concrete examples, or they are unable to derive complex transformations. We propose a novel unsupervised approach, called Ockham, which is able to learn edit operations from model histories in model repositories. Ockham is based on the idea that meaningful edit operations will be the ones that compress the model differences. We evaluate our approach in two controlled experiments and one real-world case study of a large-scale industrial model-driven architecture project in the railway domain. We find that our approach is able to discover frequent edit operations that have actually been applied. Furthermore, Ockham is able to extract edit operations in an industrial setting that are meaningful to practitioners.

Yushan Liu, Marcel Hildebrandt, Mitchell Joblin et al.

Biomedical knowledge graphs permit an integrative computational approach to reasoning about biological systems. The nature of biological data leads to a graph structure that differs from those typically encountered in benchmarking datasets. To understand the implications this may have on the performance of reasoning algorithms, we conduct an empirical study based on the real-world task of drug repurposing. We formulate this task as a link prediction problem where both compounds and diseases correspond to entities in a knowledge graph. To overcome apparent weaknesses of existing algorithms, we propose a new method, PoLo, that combines policy-guided walks based on reinforcement learning with logical rules. These rules are integrated into the algorithm by using a novel reward function. We apply our method to Hetionet, which integrates biomedical information from 29 prominent bioinformatics databases. Our experiments show that our approach outperforms several state-of-the-art methods for link prediction while providing interpretability.

Yushan Liu, Marcel Hildebrandt, Mitchell Joblin et al.

The graph structure of biomedical data differs from those in typical knowledge graph benchmark tasks. A particular property of biomedical data is the presence of long-range dependencies, which can be captured by patterns described as logical rules. We propose a novel method that combines these rules with a neural multi-hop reasoning approach that uses reinforcement learning. We conduct an empirical study based on the real-world task of drug repurposing by formulating this task as a link prediction problem. We apply our method to the biomedical knowledge graph Hetionet and show that our approach outperforms several baseline methods.

Marcel Hildebrandt, Jorge Andres Quintero Serna, Yunpu Ma et al.

We propose a novel method for fact-checking on knowledge graphs based on debate dynamics. The underlying idea is to frame the task of triple classification as a debate game between two reinforcement learning agents which extract arguments -- paths in the knowledge graph -- with the goal to justify the fact being true (thesis) or the fact being false (antithesis), respectively. Based on these arguments, a binary classifier, referred to as the judge, decides whether the fact is true or false. The two agents can be considered as sparse feature extractors that present interpretable evidence for either the thesis or the antithesis. In contrast to black-box methods, the arguments enable the user to gain an understanding for the decision of the judge. Moreover, our method allows for interactive reasoning on knowledge graphs where the users can raise additional arguments or evaluate the debate taking common sense reasoning and external information into account. Such interactive systems can increase the acceptance of various AI applications based on knowledge graphs and can further lead to higher efficiency, robustness, and fairness.

Marcel Hildebrandt, Jorge Andres Quintero Serna, Yunpu Ma et al.

We propose a novel method for automatic reasoning on knowledge graphs based on debate dynamics. The main idea is to frame the task of triple classification as a debate game between two reinforcement learning agents which extract arguments -- paths in the knowledge graph -- with the goal to promote the fact being true (thesis) or the fact being false (antithesis), respectively. Based on these arguments, a binary classifier, called the judge, decides whether the fact is true or false. The two agents can be considered as sparse, adversarial feature generators that present interpretable evidence for either the thesis or the antithesis. In contrast to other black-box methods, the arguments allow users to get an understanding of the decision of the judge. Since the focus of this work is to create an explainable method that maintains a competitive predictive accuracy, we benchmark our method on the triple classification and link prediction task. Thereby, we find that our method outperforms several baselines on the benchmark datasets FB15k-237, WN18RR, and Hetionet. We also conduct a survey and find that the extracted arguments are informative for users.