Seminar Group 2

Academic Writing (AW)

Andy Weeger

Neu-Ulm University of Applied Sciences

November 23, 2025

Agenda

• Structure
• Research questions
• Literature and gaps
• Precision and clarity
• Language and style (Anja Zenk)
• Summary

We’ll use real examples from your cohort to illustrate each point.

Structure

The 5-paragraph formula

Every introduction needs:

1. Hook - Why this topic matters now
2. Background - What we know from literature
3. Tension - What’s missing or unresolved
4. Resolution - Your approach to address it
5. Contribution - Expected value of your work

Key principle: Each element gets its own paragraph(s)

Example #1

Artificial Intelligence for Adaptive Microscale Systems (Noha El Nagar)

The structure is clear, but the research question is missing and the link to the literature is weak.

Paragraph 1 (Hook)

“AI has become one of the defining technologies… Bringing AI capabilities to the microscale represents an emerging and largely unexplored research area.” But: focus on the relevance not the gap here

Paragraph 2 (Background)

Systematic review of microrobotics and smart materials research

Paragraph 3 (Tension)

“However, most existing microsystems remain limited in functionality because they depend on external control signals, lack local sensing and decision-making abilities, and cannot adapt to changes in the complex environment.” But: gap is not covered by literature

Paragraph 4 - 6 (Resolution)

“Artificial intelligence offers the theoretical and computational foundation to overcome these challenges… This thesis focuses on designing adaptive microscale systems…” But: RQ is missing

Paragraph 7 (Contribution)

“The expected contribution of this research is twofold. First,… an area that has so far received little attention …” But the contributions are not related to literature

Example #2

Building a Pro-democratic AI Influencer (Tobias Gebhardt)

Interesting topic, but structural disorder (resolution before tension)

Paragraph 1 (Hook)

“It is more important than ever to understand how AI systems influence civic behavior… Cambridge Analytica… concern of many scholars that AI is a threat to democracy…” strong hook

Paragraph 2 (Background)

“AI can be used in the form of virtual or AI influencers… defined as digital characters… need to overcome certain barriers… often achieve high perceived credibility…” decent synthesis

Paragraph 3 (Resolution)

“To combat the potential misuse… this thesis aims to mitigate the damage… This leads to the following research question Q1…” The RQ and aim are presented before the research gap/tension is established

Paragraph 4 (Tension)

“While prior research has examined AI influencers primarily in marketing contexts… there is a lack of connecting both fields… limits understanding of how a pro-democratic AI influencer…” The gap should come before the RQ to justify it

Paragraph 5 (Resolution)

“Therefore, I am proposing Design Science Research… tested in a 2x2 study design… thesis is grounded in Source Credibility Theory…” too lengthy for an introduction; contributions are not explicit

Example 3

Driving Equitable Healthcare in Precision Oncology for Underrepresented Groups (Joy-Angel Adoboe)

Although one can follow it despite deviations from the formula, the ‘real hook’ kicks in late.

Paragraph 1 (Hook)

“In 1951, Henrietta Lacks, a young African-American woman was treated… her cervical cells were retrieved without her permission… Repetitive similar historical events contribute to hesitancy…” the historical anecdote (1951) feels slightly disconnected from the modern AI focus; what is the problem?

Paragraph 2 (Background)

“AI can be defined as… defined by Schwartzberg et al. (2017)…” Core concepts introduced, but weak synthesis of the BoK

Paragraph 3 (Background & Tension)

“As highlighted by Hulsen et al. (2019)… However, research by Landry et al. (2018), show that genomic databases… lack representative data… Bias can be defined ‘as a systematic error’ (Ferrara, 2024)…” Opens with background; heavy reliance on a single source for multiple definitions and examples

Paragraph 4 (Hook)

“The absence of significant data for some less represented groups can pose a challenge in deriving the most appropriate treatment which can lead to bias…” Problematizes the practical problems with biases - motivates the research

Paragraphs 5 & 6 (Resolution)

“Therefore, this research seeks to answer the question on how data and algorithmic bias can be mitigated… using quantitative methods… and secondary data…” RQ is missing; experimental design claim contradicts secondary data use

Paragraph 7 (Contribution)

“The findings of this study will help support healthcare institutions to make more precise decisions… and face less bias with AI tools…” Purely practical contribution; lacks specific theoretical contribution to the BoK

Example #4

Understanding Cognitive Load Across Work Models (Viktoria Kraus)

Structural deviation, difficult to follow the thread

Paragraphs 1 & 2 (Hook)

“Before the COVID-19 pandemic… share increased sharply… the central question of this thesis is: How do employees describe and experience cognitive load…?” Problem is not yet clear, but RQ is stated already, breaking the tension-building arc

Paragraphs 3, 4, 5, 7

“Teleworking has expanded rapidly… Research highlights that well-being outcomes depend on factors such as telework intensity… Cognitive load theory (CLT), a major contribution of John Sweller… addresses how instructional design can support learning…” Attempt to a synthesis (remote work); lengthy introduction of the theory without strong why; intersection between remote work and cognitive load remains somewhat unclear

Paragraph 8 (Tension)

“Despite the growing prevalence… there remains a significant gap in understanding how employees themselves experience… Existing research has primarily focused on objective performance metrics…” The gap is stated but the claim is not backed by references

Paragraph 10 (Contribution)

“Through this approach, the thesis seeks to… contribute to the theoretical expansion of Cognitive Load Theory…” “Theoretical expansion” is vague; does not specify how the theory will be expanded; not related to literature

Takeaways

Always use an outline first.

Have one clear message per paragraph.

If a paragraph contains multiple “Additionally…” or “Moreover…” statements, it probably needs splitting.

Connect the paragraph logically.

Research questions

Good research questions

As discussed in class, a good research question should be:

• Explicit - it must be clearly stated
• Clear and focused - the question should clearly state what you need to do
• Motivated - it is totally clear why you are asking it (aligned with the hook and gap)
• Not too easy to answer - easy to locate in the text
• Researchable - you must have access to data required to answer the question.

Example #1

Knowledge Distillation for Edge Deployment of Large Language Models (LLMs) (Kakara Lalitha Sai Priya)

Specifically, the study will analyze configurations involving teacher-student layer mapping, temperature scaling, and loss function variations to evaluate their impact on model accuracy, latency, and energy consumption.

• Explicit: resarch question is missing (“only aims”)
• Clear and focused: scope is very well defined: independent variables (mapping, temperature, loss function) vs. dependent variables (accuracy, latency, energy)
• Motivated: tension between LLM capabilities and edge device constraints (latency, energy) clearly motivates the study; but rather weak linkages to literature
• Not too easy: complex experimentation needed
• Researchable: Maybe not: one needs a dataset containing both the raw project data and the original human prediction

Example #2

Can AI outperform humans in predicting project completion time in Agile software development (Eldar Gaifullin)

Can an LLM, trained on unstructured textual and historical project data, outperform a human PM in predicting software development delays?

• Explicit: clearly stated (along with a secondary question about parameters).
• Clear and focused: sets up a specific competitive analysis: LLM vs. Human PM.
• Motivated: estimates are historically inaccurate; maybe bc we are ignoring half the data; improving them has high business value (but is LLM the right tech?)
• Not too easy: comparing performance is complex.
• Researchable: maybe not: one needs a dataset containing both the raw project data and the original human prediction

Example #3

Can Artificial Intelligence Truly Understand Human Emotions? A Study on Emotional Intelligence in Chatbots (Merrybell Babu)

To what extent can artificial intelligence, particularly chatbots, understand and replicate human emotional intelligence?

• Explicit: clearly stated.
• Clear and focused: using the word “Understand” in AI research is dangerous (philosophical debate); you cannot measure a machine’s “understanding”, only its “output”.
• Motivated: the gap between simulated empathy and understanding is interesting, but …
• Not too easy: answering if a machine “truly understands” is philosophically impossible (for you).
• Researchable: the method focuses on user perception, but the RQ asks about the machine’s capability to understand.

The RQ could be motivated by a gap in user experience design (and maybe its consequences), not machine consciousness.

Literature and gaps

From bibliography to narrative

Don’t just list - synthesize!

Weak: “Smith (2020) found X. Jones (2021) found Y. Brown (2022) found Z.”

Strong: “Early research established X (Smith 2020), which was extended to include Y (Jones 2021). However, recent work challenges these assumptions by showing Z (Brown 2022).”

Example #1

Generative AI and Organizational Privacy: Modeling Human Risk Factors in Data Leakage Incident (Syed Shah)

Good synthesis and transition to the gap:

Gupta et al. (2023) argue that generative AI models are a ‘double-edged sword’… Byreddy (2024) reinforces this perspective… Feuerriegel et al. (2024) further highlight… Collectively, these studies suggest that…

Then clearly states the gap:

Despite growing scholarly attention, the phenomenon of human-driven data leakage through AI prompts remains insufficiently understood.

And explains why it matters:

This represents a critical gap… vital for developing effective prevention strategies…

Example #2

Reinforcement Learning for Adaptive Cyber Defense (Sudhandra Babu)

Strong narrative arc:

Early research concentrated on anomaly detection… Later work began applying RL to dynamic scenarios… In parallel, scholars demonstrated multi-agent reinforcement learning… These contributions collectively illustrate the shift from reactive to adaptive security

However, only a few studies were cited, so the link to the literature needs to be strengthened.

Note: Uses temporal markers (“early,” “later,” “in parallel”) to show evolution of ideas

Example #3

The Impact of AI in Crowdsourcing Workforce (Jofel Diaz Ortega)

Vague gap:

The collateral effects which can emerge on crowd workers by this coming evolution have not been deeply studied.

• What effects specifically?
• Why does this matter?
• Who is affected?

Takeaways

Your sources are the product of thorough literary research.

Weave these into your narrative — build an argument supported by studies.

For every gap statement, make clear:

1. What’s missing? Be specific about missing knowledge
2. Why does it matter? Real consequences
3. Who cares? Stakeholders affected

Language and clarity

Citation discipline

Weak:

It is widely believed that AI causes bias

Strong:

Research demonstrates that biased training data leads to discriminatory outputs (Kordzadeh & Ghasemaghaei, 2022)

Every factual claim needs a citation

Vague vs. specific references

Weak:

Studies show that…

Researchers have found…

It has been argued…

Strong:

Dietvorst et al. (2015) found that …

Recent meta-analysis reveals … (Wang et al., 2024)

Contrary to earlier assumptions, Liu (2023) demonstrates …

Academic confidence

Too tentative:

It might be possible that perhaps organizations could potentially benefit…

Too bold:

This research will definitively solve…

Appropriate:

This research aims to… Evidence suggests that… The findings are expected to contribute…

Other observations

1. Claims without citations (e.g., Laura Garcia Santos)
2. Overly long sentences (5+ lines; e.g., Smarika Rijal)
3. Undefined technical terms (multiple students)
4. Mixed terminology (use different terms for same concept; multiple students)

Define terms on first use, maintain consistent terminology, cite all claims

Summary

Key takeaways

Structure

• Follow the 5-paragraph formula: Hook — Background — Tension — Resolution — Contribution
• One paragraph = one main idea
• Use transitions between paragraphs

Literature

• Take literature work seriously
• Weave citations into your narrative naturally
• Don’t just list studies—synthesize them
• Every factual claim needs support

Research question

• State it explicitly as a question
• Set it up—explain why we need to ask it

Clarity

• Use specific, concrete language
• Avoid vague terms like “challenges” or “issues” without specifying
• Define core concepts
• Use the “reverse outline” test: Can you summarize each paragraph in one sentence?

Final advices

• Literature is key: gain a solid understanding of the literature. Read, read, read.
• Structure is scaffolding: use it until it becomes natural.
• Make your RQ unmissable: bold it, frame it, highlight it.
• Claims need to be backed: every factual claim has a citation.
• Gaps need specificity: more research needed says nothing.
• Literature tells a story: show evolution and debate.
• Clarity beats complexity: clear writing reflects clear thinking.

Great introductions aren’t written, they’re rewritten!

Q&A