Visual Process Representations (VPR)

• Visual Process Representations (VPR) are automated, narrative visualizations that convert detailed log data into clear workflows mapped to KM processes such as Access, Store, Share, and Apply.
• They integrate sequential pattern mining with human-centered design, employing iterative prototyping and co-design workshops to transform expert teacher actions into engaging, accessible visual formats.
• Empirical evaluations reveal that enriched pictorial formats improve task performance and engagement while balancing narrative richness against potential cognitive overload.

Visual Process Representations (VPR) are automatically generated, intuitive visualisations of expert teachers’ workflows. In the reported formulation, they transform fine-grained log data—clicks, keystrokes, and navigation—into a sequence of Panels grouped into Knowledge Management (KM) “Sections” labelled Access, Store, Share, and Apply. By combining Sequential Pattern Mining (SPM), KM processes, and storytelling techniques, VPRs are intended to make tacit expert know-how explicitly accessible to novice teachers while reducing the burden of manual workflow documentation in university settings with high staff turnover (Fernández-Nieto et al., 6 Aug 2025).

1. Definition, scope, and stated goals

VPRs are defined as a design approach for converting expert log data into clear visualisations of workflows. The target domain described in the source is teacher work, especially workflows recorded across systems such as LMS platforms, Zoom, and repositories. The underlying problem is framed in KM terms: expertise loss disrupts teaching, documenting teachers’ workflows is time-intensive, and manual capture diverts experts from core responsibilities (Fernández-Nieto et al., 6 Aug 2025).

The primary goals specified for VPR in teaching workflows are threefold. First, they aim to capture procedural know-how without imposing extra work on experts. Second, they aim to share and transfer that know-how to novices in an engaging, easy-to-follow format. Third, they aim to support task accuracy, reduce trial-and-error, and mitigate knowledge loss due to staff turnover. Within this formulation, VPR is not merely a visualization layer; it is a mechanism for codifying operational know-how into artefacts that can be stored, shared, and reused (Fernández-Nieto et al., 6 Aug 2025).

The paper places particular emphasis on narrative structure. Panels, sections, temporal sequence, and comic-inspired flow are not peripheral stylistic choices but part of the representation’s intended intelligibility. This suggests that VPR is positioned at the intersection of process mining, KM, and visual storytelling rather than as a conventional process diagram or static checklist.

2. Sequential Pattern Mining as the computational basis

The computational motivation for VPR is grounded in SPM. The stated rationale is that SPM handles temporal, sequential data at scale, and that institutional log data capture fine-grained actions of experts. On this basis, SPM is used to uncover recurrent subsequences corresponding to expert KM processes. The paper uses an off-the-shelf SPM/process-variation analysis via pm4py to identify frequent action subsequences, which are then mapped to KM subprocesses (Fernández-Nieto et al., 6 Aug 2025).

The integrated overview provides standard SPM notation while noting that the paper does not present explicit formulas. Let $D = \{s_1,\dots,s_N\}$ be the set of recorded event sequences. A sequential pattern $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$ is a subsequence of $s_i$ if it appears in order. The standard definitions given are:

$$\mathrm{supp}(P) = \frac{|\{\, s \in D : P \subseteq s \,\}|}{|D|}$$

$$\mathrm{conf}(P \Rightarrow Q) = \frac{\mathrm{supp}(P \biguplus Q)}{\mathrm{supp}(P)}$$

$$\mathrm{lift}(P \Rightarrow Q) = \frac{\mathrm{conf}(P \Rightarrow Q)}{\mathrm{supp}(Q)}$$

No bespoke LaTeX pseudocode is provided in the text. The algorithmic pipeline instead consists of identifying frequent action subsequences and mapping them to KM subprocesses. In practical terms, the resulting structure is not presented as raw mined patterns; it is transformed into a human-consumable sequence of panels and sections. A plausible implication is that the representation problem is treated as equally important as the mining problem: extracting expert regularities from logs is necessary, but insufficient unless the output is intelligible to novice educators.

3. Human-centred design methodology

The design methodology is described as a three-phase service-design framework, sometimes described as a four-step process when iterative prototyping is counted separately. The framework is attributed to Evenson and Dubberly and integrates KM theory with storytelling (Fernández-Nieto et al., 6 Aug 2025).

Phase 1 is the exploratory stage, described as Observation and Reflection. Co-design workshops with 13 teachers over two years yielded four Design Requirements: DR1 flexible granularity, DR2 fine-grained notes/images, DR3 link novices to experts, and DR4 timely access to relevant knowledge. From these, the authors derived three Design Goals. DG1 is narrative simplicity and goal alignment, including goal-orientation and decluttering. DG2 is intuitive navigation and comic-inspired flow through panels, sections, and temporal sequence. DG3 is adaptable visualisation options, specifically text versus pictorial presentation, with or without context.

Phase 2 is the generative stage, described as Modelling and Prototyping. Here, log data are modelled into process steps, then into KM subprocesses, and finally into the VPR structure. Two iterations are reported. Iteration 1, a low-fidelity prototype, mapped steps to Panels and Sections and explored text versus pictorial styles, with and without screenshots. Iteration 2, a high-fidelity prototype, refined the layout and added a scrollytelling overview, clickable URLs, and zoomable screenshots (Fernández-Nieto et al., 6 Aug 2025).

Phase 3 is the evaluative stage, implemented as a between-subjects user study. Quantitative measures included accuracy, time, memorability, engagement, and usability, complemented by qualitative feedback. The structure of this methodology indicates that VPR is conceived as a human-centred KM artefact rather than as a purely algorithmic visualization output. The representation is iteratively shaped around requirements concerning granularity, context, navigation, and practical accessibility for novice teachers.

4. Representational structure and the four variants

All four VPR variants present the exact same KM-mapped workflow content, with Sections corresponding to Access, Store, Share, and Apply, but they differ in visual affordances and contextual depth (Fernández-Nieto et al., 6 Aug 2025).

Variant	Form	Context
P1	Plain numbered list grouped under KM Section headers	No pictorial elements, no screenshots
P2	Comic-inspired panels and icons; sequential art; colour to highlight current step	No screenshots
P3	Same numbered list as P1	Thumbnail screenshots and brief annotations
P4	Panels and icons as in P2	Embedded screenshots, zoomable

P1 is a textual list of steps. It uses a plain numbered list grouped under KM Section headers and includes neither pictorial elements nor screenshots. P2 is a pictorial-style representation built from comic-inspired panels and icons, using sequential art and colour to highlight the current step, again without screenshots. P3 preserves the numbered-list structure of P1 but enriches it with thumbnail screenshots and brief annotations. P4 combines the pictorial elements of P2 with embedded, zoomable screenshots, making it the most visually rich and the variant with the highest information density (Fernández-Nieto et al., 6 Aug 2025).

The distinctions among these variants operationalize DG3, the requirement for adaptable visualisation options. They also instantiate a controlled comparison between visual style and contextual augmentation. A common assumption in workflow support is that richer context necessarily improves performance. The reported evaluation does not fully support that assumption: screenshots can aid navigation in some cases, but heavy context can also overload novices.

5. Empirical evaluation design

The empirical study is a between-subjects comparison involving $N = 160$ teachers recruited on Prolific, with 40 participants per prototype. Two tasks were used: Task 1, a marking-correction workflow with 18 multiple-choice questions, and Task 2, online poll creation with 16 multiple-choice questions. The procedure had three activities: demographics and familiarisation; task performance with Part A completed using the prototype and Part B completed after the prototype was removed to measure memorability; and an engagement survey, a usability survey, and open-ended suggestions (Fernández-Nieto et al., 6 Aug 2025).

The performance metrics are explicitly defined. Accuracy is the number of correct answers per task. Completion time is measured per task. Memorability is Part B accuracy after the prototype has been removed. Usability and engagement were measured on Likert scales from 1 to 5 covering clarity, attractiveness, fun, curiosity, intent to reuse, completeness, and ease (Fernández-Nieto et al., 6 Aug 2025).

The statistical analysis uses regression models with binary prototype indicators and no additional covariates, Bonferroni correction for multiple pairwise comparisons, Cohen’s $d$ for effect size, and correlation analyses for time versus score. Participants answering in less than 30 seconds per question were excluded, following Douglas et al. 2023. This design places emphasis on both immediate task support and retained procedural understanding, while also examining subjective reception of the visual forms.

6. Findings, limitations, and interpretive issues

The quantitative results indicate that all prototypes were effective, with pictorial styles slightly outperforming textual styles on task accuracy. For combined Part A and Part B accuracy, P2 led across both tasks, with average scores of 18/18 for Task 1 and 16/16 for Task 2. For Task 1 Part A pairwise comparisons, none of the reported differences reached $p < .05$ after Bonferroni correction. The excerpted comparisons are P1 vs P2: Coef $0.707$, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$0, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$1, Bonf-p $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$2, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$3; P1 vs P3: Coef $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$4, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$5, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$6, Bonf-p $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$7, $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$8; and P2 vs P4: Coef $$P = \langle e_1 \rightarrow e_2 \rightarrow \dots \rightarrow e_k \rangle$$9, $s_i$0, $s_i$1, Bonf-p $s_i$2, $s_i$3 (Fernández-Nieto et al., 6 Aug 2025).

Completion time results are more mixed. For Task 1 Part A, P2 was fastest at approximately 350 seconds, while P3 and P4 were slowest at approximately 450 seconds. For Task 2 Part A, P4 was faster than P2, suggesting that screenshots aided navigation. However, no pairwise completion-time comparisons survived Bonferroni correction. Memorability results show that P2 allowed quickest recall with stable high scores, whereas contextual prototypes P3 and P4 regained some Part A gains, but heavy context could overload novices (Fernández-Nieto et al., 6 Aug 2025).

Engagement and usability results favor the more visual variants. For the sample engagement item “I found this visualisation fun,” the ordering is P4 $s_i$4 P2 $s_i$5 P3 $s_i$6 P1, with P4 vs P1 reported as $s_i$7, $s_i$8, and P2 vs P1 as $s_i$9, $$\mathrm{supp}(P) = \frac{|\{\, s \in D : P \subseteq s \,\}|}{|D|}$$0. P2 and P4 were rated highest on completeness, clarity, and ease of following, whereas P1 scored lowest overall. The qualitative feedback also identifies limits of enrichment: 29 participants reported the visualisation as too complex or overwhelming, especially for P2, P3, and P4; 20 suggested simplifying colours or icons or changing screenshot use; and 9 found it hard to distinguish sections from steps (Fernández-Nieto et al., 6 Aug 2025).

These findings support a more discriminating interpretation than a simple “more visuals are better” claim. Pictorial simplification appears to help comprehension and engagement, while contextual augmentation may improve certain navigation tasks but can increase information density. This suggests that the central design trade-off in VPR lies between narrative richness and cognitive load.

7. Knowledge-management significance and implementation trajectory

Within the source’s discussion, VPRs are positioned as KM artefacts for high-turnover educational settings. Automated capture through a browser extension combined with SPM is presented as requiring minimal extra effort from experts. The resulting VPRs are described as living codified artefacts that can be stored, shared, and updated in a KM repository. Their panel-and-section structure maps directly to the KM subprocesses Access, Store, Share, and Apply, thereby reinforcing KM theory in practice (Fernández-Nieto et al., 6 Aug 2025).

The practical recommendations are explicit. Institutions are advised to instrument teaching systems with lightweight log capture, apply off-the-shelf SPM or process-variation tools such as pm4py, map frequent action subsequences to the four-fold KM framework, and group them as Sections. The source further recommends generating two parallel VPR formats: a simple comic in the P2 style for rapid comprehension and an annotated list in the P3 style for users who prefer linear textual guidance. It also recommends toggles for contextual screenshots, an overview navigation bar or scrollytelling element that highlights the current Section and step, iterative refinement of colour-coding, terminology, and panel layout, and embedding VPRs within onboarding workflows and KM repositories so that new teachers have immediate access to current procedural know-how (Fernández-Nieto et al., 6 Aug 2025).

In this formulation, VPR addresses two linked problems: the capture of expert procedural behavior from operational traces, and the communication of that behavior in a form that novice practitioners can follow. The paper’s results indicate improved task performance, usability, and engagement, particularly with enriched visuals, while also showing that process memorability and task-time improvements were limited. The broader implication is not that a single visual form is universally optimal, but that workflow codification benefits from adjustable representational depth, especially when the audience consists of novices acquiring institution-specific know-how (Fernández-Nieto et al., 6 Aug 2025).