Big data from online debates to urban design and planning actions: Visualizing large-scale human-centric data for post-participation

Participation context: Hamburg, Germany

The city of Hamburg is utilized as a case study in this research to examine project-based citizen participation. Hamburg, a major city in northern Germany with a population of 1.85 million, is characterized by its cultural and demographic diversity (Statistische Jahrbücher - Statistikamt Nord, 2023). The city is divided into seven boroughs and 104 neighborhoods, each with its own unique identity and set of challenges. This diversity necessitates a contextualized approach to urban design and planning, as one-size-fits-all solutions are often ineffective. Therefore, Hamburg emphasizes democratic participation by encouraging its residents to co-create and share responsibility with its decentralized government. It employs a combination of digital and traditional platforms, integrating online surveys and public forums with town hall meetings to create a comprehensive system for civic engagement. The digital component of this participation model will be further elaborated in the following section.

Participation tool: DIPAS|digital participation system

Established in 2016, the Digital Participation System (DIPAS) serves as Hamburg’s digital platform for incorporating public involvement in urban development projects (DIPAS, 2023). Notable for its design, data collection methods, and variety of data types, such as comments, replies, timestamps, and geolocations, DIPAS aims to democratize urban development by providing residents with a platform to express their views and engage in substantive dialogues. Users can optionally georeference their comments by placing them on specific locations on the map, thereby indicating the spatial focus of their concern or suggestion. It promotes adaptability, transparency, and accessibility to harness collective intelligence in addressing urban challenges (Thoneick, 2021). This is achieved by harmonizing digital and physical domains, where users can engage via conventional web browsers or specialized touch desks in participation workshops, fostering inclusive civic engagement (Lieven, 2019).

To further facilitate this engagement, DIPAS presents users with an interface displaying civic contributions, serving as both an informational artifact and an interactive medium. These visual components enable users to conceptualize urban initiatives within specific community settings, integrating both content and spatial variations. An example of a participation project interface with its discussion can be seen in Figure S2 in Supplemental Material.

While this study focuses on consultative participation processes, as defined by the structure of the DIPAS platform, it is important to acknowledge the increasing relevance of co-creative models in urban planning. Co-creation involves citizens in shaping agendas, generating proposals, and contributing to decision-making, extending beyond feedback collection. Although the methodological requirements differ, the analytical and visualization frameworks that are introduced in this study also support co-creative settings by helping to structure shared insights and make collective narratives accessible.

Participation process: Participants and data collection

The DIPAS platform removes common barriers to participation, such as eligibility criteria and mandatory registration, allowing a broad cross-section of residents, including various age groups, socio-economic backgrounds, and digital literacy levels, to engage in urban governance. The platform operates through specialized web domains created for each civic project, enabling residents to provide input and commentary specific to ongoing initiatives. It supports fully digital, asynchronous participation, allowing individuals to contribute at their own pace.

All datasets used in this study were provided by the Municipality of Hamburg in anonymized form, with no access to personal or identifying information. The DIPAS platform enforces user anonymity as part of its operational protocol, ensuring that all citizen contributions are de-identified prior to release. This study adheres to these ethical standards and analyzes participation data at an aggregated level. Furthermore, the visualizations developed focus on structural patterns and thematic clustering rather than individual-level tracing, reinforcing both data protection and ethical use of public input. This anonymity is crucial in creating a safe space for civic discussion.

Each DIPAS project is initiated and managed by a municipal authority and typically remains open for public input over a period of three to sixweeks. This time frame ensures extensive participation while maintaining a manageable scope for downstream data processing. For this research, we analyzed datasets from 22 distinct DIPAS projects, which together encompass more than 19,000 comments submitted in response to online planning debates. Eight of these projects used an earlier platform architecture (Drupal 7), with data provided as .xlsx files, while the remaining 14 were hosted on an updated system (Drupal 8), which supports automated data retrieval via API.

To contextualize the relevance of the cross-project visualizations presented later in this paper, it is important to note that these 22 projects span a variety of urban design and planning themes, including mobility infrastructure (Elbchaussee), climate-responsive neighborhoods (Klimafreundliches Lokstedt), public realm design, green space interventions, and district-scale planning frameworks (Eimsbüttel). While each project is thematically focused and geographically anchored, several share overlapping districts, complementary timeframes, or similar policy objectives. This makes cross-project analysis both possible and meaningful. For instance, recurring concerns about accessibility, environmental quality, or housing pressure emerge in different project contexts, enabling comparative insight. These relationships justify our use of integrated visualizations later introduced in this paper, which aim to capture systemic issues across planning initiatives and provide post-participants with a broader, evidence-based view of community discourse.

Data preprocessing

The data preprocessing steps include data extraction, translation, linguistic pattern identification, and storage of the preprocessed data in a database. Initially, the relevant dataset for the project is extracted and subjected to translation using Python libraries such as “pandas” for data manipulation and “googletrans” for interfacing with Google Translator services. Titles, comments, and replies are translated from German into English and stored in a data frame.

This translation step was necessary to ensure compatibility with the NLP techniques applied in this study, particularly topic modeling and sentiment analysis, which are optimized for English-language corpora and demonstrate greater accuracy and model stability in English. While translation may introduce minor semantic drift or nuance loss, the analytical focus remains on broader discourse structures and thematic patterns that are generally preserved across translation. In real-world applications, by using the same embedded libraries that supported the original translation, the results and visualizations can be accurately translated back into German, allowing local stakeholders to access and interpret the findings in their native language without loss of structure or analytical value.

Following translation, the Natural Language Toolkit (NLTK) Python library is used for data cleaning. This process involves converting all text to lowercase, tokenizing the text, removing punctuation, special characters, and stopwords, and applying stemming and lemmatization techniques.

After cleaning, terms that appear less frequently than a designated threshold are eliminated. This minimum word frequency threshold (cut-off value) of 10 was applied to remove infrequent terms unlikely to contribute meaningfully to topic formation. Additionally, bigrams and trigrams, combinations of words that frequently co-occur, are generated. The Term Frequency-Inverse Document Frequency (TF-IDF) algorithm from the Gensim library is then applied to filter out words that are overly common and thus non-contributory to the clustering algorithms. In TF-IDF, the topic contribution threshold of 0.03 was set to exclude marginal topics that contributed less than 3% to the dataset, based on initial model tuning and inspection of coherence scores and keyword clusters. These values were selected iteratively to balance model interpretability and granularity. This configuration yielded the most coherent and stable results for downstream visualization and analysis.

To ensure data quality and analytical robustness, a comprehensive filtering and validation process preceded model execution. Incomplete, duplicate, and off-topic entries were removed, and text normalization ensured linguistic consistency across the dataset. A 10-fold cross-validation was conducted to verify topological homogeneity, confirming that topic clusters and distributions remained stable across data partitions. This combination of quality control and validation procedures ensured that the dataset was both thematically relevant and structurally consistent, providing a reliable foundation for subsequent visualization and interpretation.

Data analysis methods

After data preprocessing, advanced analytical techniques are applied to both textual and spatial data, structured into three key areas: topic modeling with BERTopic, sentiment analysis using TextBlob, and geospatial data analysis through interactive mapping. Collectively, these methods provide a comprehensive understanding of citizen participation by capturing the semantic, emotional, and spatial dimensions of public discourse as discussed in S1. Details of the Data Analysis Methods are in the Supplemental Material.

Each method used in this study addresses a distinct aspect of large-scale participation data. Topic modeling (BERTopic) uncovers latent thematic structures, sentiment analysis captures emotional tone, spatial and temporal mapping situates discourse in place and time, and network-based visualizations (e.g., Sunburst and Text-Network Diagrams) reveal structural and interactional patterns (Campello et al., 2015; Devlin et al., 2019). Each method offers a unique perspective, and their combined use forms a multi-method framework that supports interpretation beyond surface-level trends.

While each technique has methodological limitations, such as topic modeling’s tendency to abstract complex narratives into simplified thematic clusters, or sentiment analysis’s sensitivity to contextual ambiguity and its potential to misclassify sarcasm, negation, or culturally specific expressions (Liu, 2012), these shortcomings are mitigated through methodological triangulation. By applying complementary techniques in parallel, the analysis allows for cross-validation of findings, enhancing the reliability and interpretability of results and reducing the risk of biased or incomplete insights. Table 1 outlines the role of each technique within the analytical framework, specifying its function, scale of application, and relevance to interpreting participation data for urban planning purposes.

OPEN IN VIEWER

Table 1.

Analytical techniques used in this study, including their interpretive focus, scale of application, and implications for post-participation analysis in urban design and planning.

Technique	Analytical function	Applicable scale	Planning implication
Topic modeling (BERTopic)	Identifies thematic clusters in citizen input	Dataset/Project/Topic	Identifies dominant issues to guide planning priorities
Sentiment analysis	Assesses emotional tone of public feedback	Topic/Project	Detects dissatisfaction or support; informs tone of response
Geospatial mapping	Links comments to spatial locations	Project	Pinpoints areas of high engagement or localized issues
Network-based visualizations	Visualizes hierarchical depth of discussions	Project/Topic	Reveals complexity of engagement; informs dialogue depth

Data interpretation criteria

The interpretation of results derived from preprocessing and analyses steps often faces challenges due to the complexity of the data and the limitations of tabular presentations in illustrating correlations and hierarchical relationships. To overcome these challenges and provide actionable insights for post-participants, this research proposes the development of a rigorous set of interpretation criteria. These criteria combine the quantitative rigor with the qualitative depth of digital citizen participation data, enabling a more detailed and meaningful analysis.

Table 2 provides a structured overview of data interpretation criteria essential for analyzing digital citizen participation in urban design and planning. These criteria are developed based on observed analytical needs during exploratory interactions with planners and researchers who routinely work with citizen data in post-participation phases. Rather than adopting a predefined set of evaluation metrics, we focused on functionalities that post-participants explicitly sought, such as identifying dominant concerns, locating spatial clusters of sentiment, or assessing engagement complexity. These practice-driven dimensions informed the selection and structuring of the visualization modules introduced in this study.

OPEN IN VIEWER

Table 2.

Data interpretation criteria in digital citizen participation.

Type	Criterion	Description	Significance
Quantitative criteria	Topic number	Measures the number of topics under projects	Helps policymakers understand the diversity of issues under discussion
	Reply number	Indicates engagement levels	Highlights topics or comments that drive public discourse
	Support/Like number	Identifies contributions with significant public support	Reflects the relevance of contributions
	Coherence scores	Assesses the clarity and structure of created topics	Ensures comprehensible and actionable public input
	Sentiment scores	Evaluates the emotional tone of discourse	Offers insights into public sentiment on specific issues
	Frequent words	Identifies predominant themes	Provides an overview of salient topics
	User activity	Quantifies individual participation	Offers insights into the inclusiveness of the platform
Qualitative criteria	Content analysis	Assesses the depth and diversity of perspectives in discourse	Captures detailed and contextual aspects of citizen discourse
	Discussion hierarchies	Reveals patterns and prioritization within discussions	Helps elucidate social dynamics and relationships
	Network analysis	Maps social dynamics and information flow	Provides insights into how participants interact and share information
	Time analysis	Examines temporal engagement patterns	Analyzes the pattern of the engagement over time
	Quality of arguments	Evaluates the logical structure and persuasiveness of arguments	Assesses the strength and validity of contributions
	Diversity of perspectives	Inspects the range of views	Highlights inclusivity and the representation of different viewpoints

Accordingly, quantitative criteria, such as Topic Numbers, Reply Numbers, and Sentiment Scores, offer empirical rigor by providing objective metrics that facilitate systematic analysis of large datasets. These metrics help quantify aspects like issue diversity, engagement levels, and public sentiment, enabling policymakers to gauge community concerns effectively. In contrast, qualitative criteria like Content Analysis, Discussion Hierarchies, and Diversity of Perspectives delve into the nuanced and contextual dimensions of citizen discourse. They capture the depth and diversity of perspectives, reveal underlying patterns, and offer insights into social dynamics and participant interactions. Together, these criteria form a comprehensive framework that balances empirical data with contextual understanding, thereby enhancing informed decision-making in urban transformation processes.

Thematic structures and hierarchical relationships

A thorough understanding of project-specific data is essential in the initial phase of the analytical procedure. Utilizing Topic Modelling (TM) techniques enhances the depth of understanding of dialogues within the citizen participation process. This foundational step facilitates the identification of key issues and trends within public discussions. For example, by constructing the BERTopic model for the “Elbchaussee” dataset from DIPAS, 17 topics were generated based on the highest coherence value of the model. These topics included pertinent keywords and their importance scores, calculated using the Class-based Term Frequency-Inverse Document Frequency (c-TF-IDF) metric, which assesses term relevance within specific topics rather than individual documents (Grootendorst, 2022). Following the BERTopic model, 17 corresponding bar charts were created as visual glossaries, capturing topics and their characteristic keywords or phrases (Figure 1(a)). Insights into the most relevant community issues, such as “city,” “cycling,” and “traffic,” were identified within the “Elbchaussee” dataset. The granularity of this method enhances the precision and specificity of data interpretation, significantly contributing to the understanding of community preferences and concerns.OPEN IN VIEWER

After identifying topics within specific projects, the next analytical phase focused on discerning hierarchical relationships among these topics. Examining topic hierarchies revealed structured interrelations, showing how specific topics are nested under overarching themes and exposing varying levels of granularity within the discourse. In the same “Elbchaussee” dataset, a dendrogram visualizes the multi-level hierarchical relationships among topics (Figure 1(b)). For example, an overarching theme like “Cycling” can be subdivided into more focused categories such as “Experience,” “Bike Routes,” and “Infrastructure.” This hierarchical structure deepens the analysis by situating specific topics within a broader context, providing a contextual understanding of their significance and interconnectedness with wider developmental concerns.

The BERTopic model yielded 17 distinct topics from the Elbchaussee dataset, derived through density-based clustering and coherence optimization, which balances thematic granularity with interpretability. This number is not pre-set but emerges from the model’s internal algorithm, guided by a coherence score that evaluates semantic consistency within each cluster. Each topic consists of a ranked list of keywords produced using class-based TF-IDF, with associated weights indicating their relevance. Based on these keywords and representative comments, thematic labels were qualitatively assigned to aid interpretation (e.g., “Cycling Safety,” “Public Transport Access,” and “Green Space Usage”). While topic modeling is inherently unsupervised, the resulting structure offers a high-level overview of discourse complexity.

For post-participants, the number and composition of topics signal the thematic diversity and fragmentation of public input. A higher number of coherent topics may reflect multidimensional discourse, suggesting that different groups prioritize different concerns. Understanding topic content and overlap enables post-participants to identify emerging tensions, prioritize issues, and tailor interventions accordingly, whether through design refinements, communication strategies, or targeted follow-up.

It is important to note that these topics are derived entirely from BERTopic modeling and are distinct from the project-defined topic categories within the DIPAS platform (e.g., Topic 452 and Topic 19), which are created by the system to structure citizen input during the participation phase. This distinction is maintained throughout the analyses.

Applying hierarchical investigations to citizen participation data has significant implications for post-participation analysis. It supports the development of strategies that address issues at both general and specialized levels. The hierarchical structuring of topics allows post-participants to understand how specific issues fit within broader thematic frameworks, facilitating the creation of detailed and contextually relevant interventions. Additionally, this approach offers a multi-tiered perspective on topics, providing post-participants with contextual insights that support informed and multifaceted urban interventions. This heightened level of analysis is crucial for offering a comprehensive overview of textual data in urban environments, thereby enhancing the capacity to engage more meticulously with the diverse concerns of the citizenry.

Quantitative assessment of semantic similarities

Following the thematic and hierarchical analysis, lexical overlaps among different topics indicated significant semantic similarities. These overlaps necessitated a detailed exploration to discern interrelationships and subtle distinctions between topics. To identify such similarities, topics were graphically represented using an intertopic distance map and a similarity matrix.

In Figure 2(a), the intertopic distance map is created for the “Elbchaussee” dataset from DIPAS. In this multidimensional space, each circle on the map represents a topic, with the circle size indicating its relative importance and overlaps signifying high degrees of thematic similarity. In Figure 2(b), on the other hand, topics were quantitatively assessed for their similarities by computing similarity scores ranging from 0 to 1. The quantitative foundation of the similarity matrix enhances the reliability of interpreting topic interrelations, making it particularly effective for identifying clusters of semantically related topics.OPEN IN VIEWER

While both the dendrogram (Figure 1(b)) and the intertopic distance map (Figure 2(a)) visualize relationships among topics generated through topic modeling, they offer different perspectives and serve complementary purposes. The dendrogram presents a hierarchical structure, clustering topics based on their semantic proximity in a tree-like format. It reveals how specific topics are nested under broader thematic umbrellas, supporting structured exploration of discussion depth and sub-theme organization. In contrast, the intertopic distance map offers a spatial representation of topics in a two-dimensional semantic space, where the proximity between circles indicates overall similarity, and the size of each circle reflects the relative prevalence of that topic in the dataset.

For post-participants, the dendrogram is useful when understanding the layered structure of public discourse, for example, how general themes like mobility branch into more focused concerns such as parking or cycling. The intertopic distance map, on the other hand, enables quick identification of thematic overlaps, redundancies, or topic clusters that may require integrated responses. Taken together, these visualizations help interpret both the structure and the spread of discourse complexity in a given participatory context.

Similarity scores in the presented matrix were calculated by representing each topic as a vector, with each word in the vocabulary assigned a weight based on its relevance to that topic. Cosine similarity was then used to compare these vectors, evaluating the degree of similarity between topics. The similarity scores were organized into a matrix, with each entry indicating the closeness between two topics. This matrix is subsequently visualized using a heatmap, where each cell reflects the level of similarity between the topics being compared.

For instance, in the “Elbchaussee” dataset heatmap (Figure 2(b)), a substantial similarity score of 0.88 exists between “Parking and Pedestrian Flow” (T0) and “Cycling Infrastructure” (T1), suggesting these issues are frequently discussed together and conceptually interconnected. Similarly, “Parking and Pedestrian Flow” (T0) registers a similarity score of 0.83 with “Public Transportation Services” (T6), indicating an intrinsic relationship between various transportation concerns. Such analyses provide insights into efforts, for example, parking spaces should not be isolated from broader concerns such as cycling routes and public transportation networks. Addressing these interconnected issues synergistically helps alleviate congestion and improve overall urban mobility.

Consequently, the dual capabilities of semantic analysis enhance the ability to make substantive recommendations based on detailed understandings of topic relationships. By isolating clusters of closely related topics, post-participants can detect underlying patterns or emerging trends in community sentiment, enabling more targeted and evidence-based policy initiatives.

High-dimensional data clustering

Following the generation of TMs, post-participants obtain topics with their prevalent words; however, the dataset still includes thousands of sentences from comments, replies, and descriptions on the online platform. To effectively trace content through these topics, it is necessary to represent this high-dimensional data using clustering techniques that combine the content with the TM results. The aim is to simplify complex datasets by grouping similar data points, thereby aiding in the identification of patterns and relationships. In this study, t-distributed Stochastic Neighbor Embedding (t-SNE) is employed, which is a machine learning technique that reduces dimensionality by preserving local structures, ensuring that similar data points remain close in the lower-dimensional space (Zhu et al., 2019). t-SNE was selected for its ability to reduce high-dimensional topic distributions into a two-dimensional space while preserving local semantic relationships between comments, thereby enabling post-participants to visually detect thematic clusters, overlaps, and outliers within complex discourse data.

The results of the t-SNE were plotted using the “Elbchaussee” dataset, as presented in Figure 3. In the figure, each point represents a comment or reply, with proximity indicating similar topic distributions. Distinct clusters emerged in the plot, such as Topics 1, 10, and 11. This distinction is due to their unique focuses, with Topic 11 concentrating on community governance issues like “bike” and “dog.” Conversely, some clusters showed proximity to other topics, such as Topic 2, which represents more interconnected arguments (Figure 3(a)). These clusters reflect topic differentiation, while outliers indicate comments that deviate from primary themes. Additionally, interactive features like hover tools were embedded into the interfaces to provide additional context, allowing deeper exploration of specific clusters, as seen in Figure 3(b). In this way, t-SNE serves as a robust tool for high-dimensional data visualization while maintaining local document structures.OPEN IN VIEWER

After identifying topics and presenting them in multidimensional spaces, determining which require further investigation becomes crucial. To achieve this, three metrics are deployed: (i) average topic contributions across documents to display the average contribution of each topic, highlighting prevalent themes (Figure S3i); (ii) documents based on topic contribution percentages, offering a granular view of topic distribution (Figure S3ii); and (iii) Kernel Density Estimate (KDE) plots to reveal the modes and spread of topic contributions, indicating clusters and the thematic composition of the corpus (Figure S3iii). Together, these plots help post-participants interpret the relative focus and diffusion of public concerns.

For instance, topics with high average contributions and dense clustering may reflect persistent concerns that cut across different user groups, while topics with low contribution values might indicate niche but coherent perspectives. These patterns inform the prioritization of engagement strategies: widely discussed and emotionally charged topics (when combined with sentiment scores) may call for immediate attention, whereas low-frequency topics may uncover emerging themes or overlooked voices. Thus, topic contribution analysis functions as both a diagnostic and prioritization tool, enabling informed decision-making by assessing the scope, depth, and balance of community discourse.

Geospatial insights into community engagement

The spatial maps generated in this study represent projects within DIPAS. Each dataset is color-coded, with comments and replies mapped using geospatial coordinates, providing a visual representation of the entire DIPAS dataset or selected sub-datasets based on the post-participants’ preferences (Figure 4). This geospatial approach serves several key objectives. First, it enables post-participants to allocate resources more effectively to areas of high engagement, or “hotspots,” where public interest is concentrated and additional resources may be needed. Second, it identifies regions with unusual engagement levels, such as in the “Eimsbüttel” project, where comments from peripheral areas offer unique community perspectives (Figure 4(a)). These insights inform outreach strategies aimed at improving participation from underrepresented groups.OPEN IN VIEWER

Moreover, this approach facilitates the mapping of conflicts and consensus by displaying overlapping comment areas, thereby signaling regions of disagreement or collective agreement related to project-specific or emerging local issues. Cross-project data integration, such as between the “Klimafreundliches Lokstedt” and “Eimsbüttel” projects (Figure 4(b)), provides broader insights into community perspectives. This spatial analysis highlights areas of interest, conflict, and consensus, ultimately informing more targeted and inclusive urban interventions.

Temporal evolution of participation data

Another important advantage of incorporating geospatial aspects of participation data is that it allows for the examination of the temporal evolution of public discourse, thereby understanding the dynamics of civic engagement over time. To achieve this, the study utilizes the timestamps attached to each comment and reply within the DIPAS dataset and constructs a temporal profile that captures evolving patterns, as well as peaks and troughs in public engagement. This temporal dimension serves multiple analytical and policy objectives. Firstly, it identifies critical moments in public discourse, such as the emergence of new themes or the intensification of conflicts, which may require immediate intervention by post-participants. Secondly, fluctuations in engagement levels can be correlated with external events (e.g., announcements or public meetings), offering post-participants exploratory insights into how civic discourse evolves over time. While causality cannot be confirmed, these patterns can help anticipate emerging issues when interpreted alongside contextual knowledge. Lastly, the system evaluates the continuity of participation through descriptive temporal visualizations; a stable profile, reflected by steady comment distribution over time, signals consistent engagement, which is crucial for long-term projects requiring sustained public involvement.

For example, the temporal progression of the “Elbchaussee” project is illustrated through maps from March 3, April 4, and April 17, 2018 (Figure 5). The steady engagement levels during these months suggest a well-managed participation initiative without significant fluctuations. This temporal analysis, while not highlighting major peaks or drops, demonstrates the stability and constant evolution of public engagement. By tracking such patterns, the analysis helps identify key trends, correlate engagement with external events, and ensure continuity, thereby supporting the development of adaptive and sustained public engagement strategies in urban transformation projects.OPEN IN VIEWER

Emotional resonance in community engagement

Sentiment analysis serves as the third analytical pillar in this study, complementing spatial and temporal analyses by adding an emotional dimension to public discourse. By examining the affective responses within the DIPAS dataset, this study categorizes comments along a spectrum from positive to negative. Aggregating these sentiment scores and mapping them across various projects in a map interface offers several advantages in the post-participation phase. Firstly, it facilitates the examination of cross-community interactions by identifying shared sentiment trends across different projects and neighborhoods. This fosters dialogue and collaboration, guiding collective initiatives to address common concerns. Secondly, sentiment analysis enhances problem identification and resolution, as persistent negative sentiment in specific areas, such as the “Holstenkamp” project (Figure 6(a)), may indicate issues requiring immediate attention and targeted interventions.OPEN IN VIEWER

Additionally, the visual tools developed in this study enable users to access specific comments by hovering over geographic markers, facilitating close textual analysis and spatial interpretation of user input (Figure 6(b)). Combining sentiment scores with spatial and temporal data, along with content tracing features, creates a three-dimensional framework for public sentiment analysis. This framework provides critical support for developing effective, context-sensitive policies. This methodology is essential for understanding the intersections of participatory activities and for guiding effective decision-making in urban transformation projects.

Comparative topological analysis of discussion topics

The primary feature of SDs and TNDs is their ability to represent the topological structures of discussion clusters across DIPAS digital participation projects. These visualizations illustrate high-level hierarchies and assess the depth and engagement within community dialogues. For instance, Figure 7(a) integrates SDs into the DIPAS dataset to compare engagement levels across projects. The “Stadteingang Elbbrücken” project exhibits a complex, branching structure, indicative of a rich and diverse community discourse. In contrast, simpler structures in projects such as “Spielplatz Op’n Hainholt” suggest lower engagement, often attributable to the specialized nature of topics requiring more expertise. Conversely, TNDs reveal intricate discussions at the topic level, as demonstrated in Figure 7(b), where topics like Topic 452 and Topic 12 in the “Magistrale Wandsbek” project display complex structures. Meanwhile, Topic 16 exhibits a simpler structure, suggesting lower engagement. By utilizing TNDs and SDs, widely discussed topics can be distinguished from niche interests, thereby facilitating the evaluation and comparison of discussion narratives across projects for cross-project analysis.OPEN IN VIEWER

Hierarchical and sentiment analysis

While both SDs and TNDs provide insights into discussion topology, SDs are particularly effective for analyzing hierarchical relationships and sentiment trends in digital participation topics. By visually representing the hierarchy of topics, comments, and replies, SDs simplify the tracing of conversation flows. Developed as interactive plots, they allow post-participants to hover over nodes to preview specific comments and replies. This interactivity enables users to zoom into discussions for a detailed view of nested comments and their hierarchical structure. For instance, in the “Stadteingang Elbbruecken” project, “Topic 19” includes multiple comments, with “Comment #3” receiving a series of replies (Figure 8). This functionality aids in tracking conversation threads dynamically across various levels.OPEN IN VIEWER

In addition to representing hierarchy, SDs offer instant visual insights into the emotional tone of discussions through color-coded sentiment scores. Positive scores indicate agreement or support, while negative scores reflect criticism or dissent. Root-level comments are organized by sentiment, allowing post-participants to quickly gauge the emotional tone of key discussions. For instance, in Figure S4 in Supplemental Material, comments are arranged from most negative to most positive within the highest hierarchical circle. Consequently, the integration of hierarchical dynamics with sentiment analysis in interactive SDs provides an effective medium in analyzing digital participation projects within the DIPAS dataset.

Temporal dynamics and comment quality

While SDs emphasize hierarchical and sentiment analyses, TNDs are instrumental in revealing temporal relationships and assessing the substance of individual comments. TNDs achieve this by representing comments and replies as nodes and the time intervals as connections between these nodes. In TNDs, shorter edges indicate faster response times, reflecting higher levels of engagement with the debate topics, whereas longer edges suggest delayed responses, indicating lower urgency. Similarly, node sizes correspond to comment and reply lengths, serving as proxies for comment quality. Larger nodes represent more thoughtful, detailed contributions, as they include more information, while smaller nodes reflect shorter, potentially reactive replies to other comments. However, since comment lengths and time intervals vary from minutes to days and from one-word replies to paragraphs, logarithmic normalization is applied to these parameters, enabling more balanced visual representation across the network. After normalization, the temporal edges preserve relative differences in engagement speed while allowing post-participants to interpret response patterns on a unified scale, thus improving readability and comparative analysis of discourse dynamics. Additionally, TNDs incorporate color-coded sentiment scores to depict the emotional tone of content in an unstructured manner, unlike SDs.

Figure S5 in the Supplemental Material illustrates a TND formed for Participation Topic 19 in the “Stadteingang Elbbruecken” project. As presented, it is possible to trace and integrate temporal intervals, comment quality, and sentiment scores through the TND system. Together, SDs and TNDs constitute powerful tools for analyzing complex and dynamic digital participation narratives. SDs excel in mapping hierarchical relationships and sentiment, while TNDs highlight temporal interactions and comment quality. Combined, these visualizations enhance post-participants’ ability to interpret discussions and make informed decisions in urban transformation projects.