Context: Coordination in large-scale software development is critical yet difficult, as it faces the problem of dependency management and resolution. In this work, we focus on managing requirement dependencies that in Agile software development (ASD) come in the form of user stories. Objective: This work studies decisions of large-scale Agile teams regarding identification of dependencies between user stories. Our goal is to explain detection of dependencies through users’ behavior in large-scale, distributed projects. Method: We perform empirical evaluation on a large real-world dataset from an Agile software organization, provider of a leading software for Agile project management. We mine the usage data of the Agile Lifecycle Management (ALM) tool to extract large-scale development project data for more than 70 teams running over a five-year period. Results: Our results demonstrate that dependencies among user stories are not frequently observed (the problem affects around 10% of user stories), however, their implications on large-scale ASD are considerable. Dependencies have impact on software releases and increase work coordination complexity for members of different teams. Conclusion: Requirement dependencies undermine Agile teams’ autonomy and are difficult to manage at scale. We conclude that leveraging ALM monitoring data to automatically detect dependencies could help Agile teams address work coordination needs and manage risks related to dependencies in a timely manner.

Abstract People express their opinions and views in different and often ambiguous ways, hence the meaning of their words is often not explicitly stated and frequently depends on the context. Therefore, it is difficult for machines to process and understand the information conveyed in human languages. This work addresses the problem of sentiment analysis (SA). We propose a simple yet comprehensive method which uses contextual embeddings and a self-attention mechanism to detect and classify sentiment. We perform experiments on reviews from different domains, as well as on languages from three different language families, including morphologically rich Polish and German. We show that our approach is on a par with state-of-the-art models or even outperforms them in several cases. Our work also demonstrates the superiority of models leveraging contextual embeddings. In sum, in this paper we make a step towards building a universal, multilingual sentiment classifier.

Continual learning (CL) aims to enable information systems to learn from a continuous data stream across time.However, it is difficult for existing deep learning architectures to learn a new task without largely forgetting previously acquired knowledge.Furthermore, CL is particularly challenging for language learning, as natural language is ambiguous: it is discrete, compositional, and its meaning is context-dependent.In this work, we look at the problem of CL through the lens of various NLP tasks.Our survey discusses major challenges in CL and current methods applied in neural network models.We also provide a critical review of the existing CL evaluation methods and datasets in NLP.Finally, we present our outlook on future research directions.

Context: Incomplete or incorrect detection of requirement dependencies has proven to result in reduced release quality and substantial rework. Additionally, the extraction of dependencies is challenging since requirements are mostly documented in natural language, which makes it a cognitively difficult task. Moreover, with ever-changing and new requirements, a manual analysis process must be repeated, which imposes extra hardship even for domain experts. Objective: The three main objectives of this research are: 1) Proposing a new dependency extraction method using a variant of Active Learning (AL). 2) Evaluating this AL and Ontology-based Retrieval (OBR) as baseline methods for dependency extraction on the two industrial data sets. 3) Analyzing the value gained from integrating these diverse approaches to form two hybrid methods. Method: Building on the general AL, ensemble and semi-supervised machine learning, a variant of AL was developed, which was further integrated with OBR to form two hybrid methods (Hybrid1, Hybrid2) for extracting three types of dependencies (requires, refines, other): Hybrid1 used OBR as a substitute for human expert; Hybrid2 used dependencies extracted through the OBR as an additional input for training set in AL. Results: For two industrial case studies, AL extracted more dependencies than OBR. Hybrid1 showed improvement for both data sets. For one of them, F1 score increased to 82.6% compared to the AL baseline score of 49.9%. Hybrid2 increased the accuracy by 25% to the level of 75.8% compared to the AL baseline accuracy. OBR also complemented the AL approach by reducing 50% of the human effort.