Browsing by Author "Cowan, Donald"

Now showing 1 - 4 of 4

Adaptive Human-Chatbot Interactions: Contextual Factors, Variability Design and Levels of Automation
(University of Waterloo, 2023-12-05) Melo dos Santos, Glaucia; Alencar, Paulo; Berry, Daniel; Cowan, Donald
The landscape of software development is undergoing a significant transformation characterized by various factors. A notable shift is the surging demand for software developers, driven by industries' increasing reliance on software solutions to support their operations. This increased demand is accompanied by an escalation in the complexity of software development projects. In this dynamic environment, modern software systems interact with numerous external systems, interfaces, data sources, and work practices. This complexity requires developers to navigate a complex environment while creating software. Adding to this landscape is the emergence of AI-based conversational systems, a transformative trend that is shaping software development processes. These systems, powered by artificial intelligence and natural language processing, enable human-like interactions through chatbots and virtual assistants. Software developers are increasingly turning to AI-powered chatbots to support their work. These chatbots play diverse roles, ranging from technical query resolution or load testing to providing project management insights and automating routine tasks. By harnessing the capabilities of these AI-driven tools, developers can potentially enhance productivity, access pertinent information swiftly, and optimize their workflows. However, amid these developments, many challenges arise due to the intricate web of contextual factors that influence software development processes, especially when chatbots come into play. These contextual factors act as distinct pieces of a puzzle, each altering how software development functions in the presence of chatbots. Unfortunately, the existing research landscape has a limited understanding of these contextual intricacies, resulting in insufficient design methods to adequately support developers using chatbots. Moreover, addressing the customization of automation levels in these interactions remains unexplored. With the growing complexity of software development, coupled with the emergence of advanced, AI-based conversational systems, the integration of chatbots to support developers in their work has become prominent. There is a pressing need to address the challenges in human-chatbot interactions, particularly in leveraging conversational agents’ advances to tailor interactions to developers’ specific contexts and desired levels of automation. This research explores the design of context-based adaptive interactions between software developers and chatbots. By understanding and integrating the contextual factors that influence software development with chatbots, we aim to gain novel insights into developers’ expectations regarding these interactions and the levels of automation involved and advance the design of human-chatbot adaptive applications. First, I perform a user study to investigate the requirements of conversational agents in software development. I uncovered a vast list of desired requirements and insights from participants, including that they are interested in working with such tools, in various parts of the development lifecycle such as managing their tasks and version control. One of the insights of this study was that contrary to the authors' beliefs, not all developers were interested in automating all possible tasks. This insight led me to the next part of this thesis, which was the investigation of the factors that influence how much automation is desired in systems. I then perform a literature review focused on studies about taxonomies of levels of automation. I aimed to uncover from these studies, the factors that influence systems switching from one level of automation to a different level. I identified these factors and composed a list of 61 factors, which we divided into five categories, system, task, human, environment, and quality. I propose feature model designs to represent these factors and their relationships and instantiate this model with use cases. This research provides a roadmap for the design of adaptive chatbot interactions that align with developers' specific needs and workflows. Empirical studies are conducted to gain insights from developers' experiences and expectations, ultimately driving the design of context-aware chatbot interactions. Additionally, by examining the influence of varying levels of automation on these interactions, I sought to identify factors that shape the variability of automation levels, bridging the gap between human-system interactions and autonomous systems.
Assessing ML classification algorithms and NLP techniques for depression detection: An experimental case study
(Public Library of Science (PLOS), 2025) Lorenzoni, Giuliano; Tavares, Cristina; Nascimento, Nathalia; Alencar, Paulo; Cowan, Donald
Context and background. Depression has affected millions of people worldwide and has become one of the most common mental disorders. Early mental disorder detection can reduce costs for public health agencies and prevent other major comorbidities. Additionally, the shortage of specialized personnel is very concerning since depression diagnosis is highly dependent on expert professionals and is time-consuming. Research problems. Recent research has evidenced that machine learning (ML) and natural language processing (NLP) tools and techniques have significantly benefited the diagnosis of depression. However, there are still several challenges in the assessment of depression detection approaches in which other conditions such as post-traumatic stress disorder (PTSD) are present. These challenges include assessing alternatives in terms of data cleaning and pre-processing techniques, feature selection, and appropriate ML classification algorithms. Purpose of the study. This paper tackles such an assessment based on a case study that compares different ML classifiers, specifically in terms of data cleaning and pre-processing, feature selection, parameter setting, and model choices. Methodology. The experimental case study is based on the Distress Analysis Interview Corpus - Wizard-of-Oz (DAIC-WOZ) dataset, which is designed to support the diagnosis of mental disorders such as depression, anxiety, and PTSD. Major findings. Besides the assessment of alternative techniques, we were able to build models with accuracy levels around 84% with Random Forest and XGBoost models, which is significantly higher than the results from the comparable literature which presented the level of accuracy of 72% from the SVM model. Conclusions. More comprehensive assessments of ML classification algorithms and NLP techniques for depression detection can advance the state of the art in terms of improved experimental settings and performance.
Graph-Based Spatial-Temporal Cluster Evolution: Representation, Analysis, and Implementation
(University of Waterloo, 2023-08-28) da Silva Portugal, Ivens; Alencar, Paulo; Cowan, Donald; Berry, Daniel
Spatial-temporal data are information about real-world entities that exist in a location, the spatial dimension, and during a period of time, the temporal dimension. These real-world entities, such as vehicles, people, or parcels and called spatial-temporal objects, may move, group, and continue the movement together, forming clusters. Although there have been significant research efforts to understand clusters, there is a lack of research that provides methods and software tools to support the representation, analysis, and implementation of graph-based spatial-temporal cluster evolution. Understanding this evolution is critical for dealing with spatial-temporal problems encountered in domains, such as service supply and demand, supply chain management, traffic and travel flows, human mobility, and city planning. This thesis presents an approach to graph-based cluster evolution and its representation, analysis, and implementation. The proposed solution introduces a representation of the structure of a spatial-temporal cluster with the identification of the cluster at several timestamps and linkages, and a representation of 14 spatial-temporal relationships clusters have during their existence. The proposed solution also introduces a graph representation of cluster evolution with nodes acting as clusters and edges as relationships. This solution provides analysis methods for the structure of spatial-temporal clusters that monitor the cluster changes in both location and size over time, and analysis methods for the spatial-temporal cluster relationships the clusters have during existence that calculate the frequency or density of such relationships in specific locations. The solution also provides analysis methods for a graph-based representation of spatial-temporal cluster evolution including integrated results that examine spatial-temporal clusters and their connections, and can provide, for example, aggregated results at a location or time of the day, identify ever-increasing or ever-decreasing regions, growth or decay rates, and measure the similarity between the evolution of two clusters. The approach also provides an implementation of the proposed representation and analysis methods. The effectiveness of the approach is evaluated through four case studies using different spatial-temporal datasets to show the results that can be produced, which include, exploratory analyses and specific analyses on ever-increasing and ever-decreasing regions, similarity values, and the movements the clusters represent. Overall, the proposed approach advances research in the spatial-temporal domain by providing novel representation and analysis methods as well as implementation tools that can improve the understanding about how clusters evolve in space and time. Such results can lead to many advantages such as higher income, reduced costs, and better transportation services, as well as the discovery of trends in cluster movement and improved decision-making processes in city planning.
A Variability-Aware Design Approach to the Data Analysis Modeling Process
(University of Waterloo, 2018-10-25) Tavares, MariaCristina; Alencar, Paulo; Cowan, Donald; Berry, Daniel
The massive amount of current data has led to many different forms of data analysis processes that aim to explore this data to uncover valuable insights such as trends, anomalies and patterns. These processes support decision makers in their analysis of varied and changing data ranging from financial transactions to customer interactions and social network postings. These data analysis processes use a wide variety of methods, including machine learning, in several domains such as business, finance, health and smart cities. Several data analysis processes have been proposed by academia and industry, including CRISP-DM and SEMMA, to describe the phases that data analysis experts go through when solving their problems. Specifically, CRISP-DM has modeling as one of its phases, which involves selecting a modeling technique, generating a test design, building a model, and assessing the model. However, automating these data analysis modeling processes faces numerous challenges, from a software engineering perspective. First, software users expect increased flexibility from the software as to the possible variations in techniques, types of data, and parameter settings. The software is required to accommodate complex usage and deployment variations, which are difficult for non-experts. Second, variability in functionality or quality attributes increases the complexity of these systems and makes them harder to design and implement. There is a lack of a framework design that takes variability into account. Third, the lack of a more comprehensive analysis of variability makes it difficult to evaluate opportunities for automating data analysis modeling. This thesis proposes a variability-aware design approach to the data analysis modeling process. The approach involves: (i) the assessment of the variabilities inherent in CRISP-DM data analysis modeling and the provision of feature models that represent these variabilities; (ii) the definition of a preliminary framework design that captures the identified variabilities; and (iii) evaluation of the framework design in terms of possibilities of automation. Overall, this work presents, to the best of our knowledge, the first approach based on variability assessment to design data modeling process such as CRISP-DM. The approach advances the state of the art by offering a variability-aware design a solution that can enhance system flexibility and a novel software design framework to support data analysis modeling.