Video summary

HN ứng dụng AI trong NCKH sáng ngày 2/8 phần 2

Main summary

Key takeaways

Educational

Main ideas & lessons conveyed

1) Class workflow: group assignments + structuring research manuscripts

  • The instructor organizes students into 5 numbered groups and assigns them “lesson” tasks.

  • Students present briefly, and the instructor then guides them to categorize:

    • Research gaps (e.g., population gap / knowledge gap / methodology gap)
    • Research objectives
    • Whether the framing fits an intro structure, such as: problem definition → concept → context → burden → existing knowledge/gaps → objectives

  • Key emphasis: there isn’t time to solve everything, so students must prioritize the most important, testable elements.

2) Research gap identification & intro construction (template-like logic)

Students discuss how a study addresses different types of gaps.

  • Population/knowledge gap
    • Example pattern: prior work compares cities; the new study compares districts within a city.
  • Methodology gap
    • Example pattern: prior work may not quantify an effect; the new work quantifies an outcome (example discussed: urban heat island effect vs mortality).

An effective intro should include components such as:

  • Definition of the concept
  • Context
  • Burden (exposure and outcomes like deaths)
  • Problem statement
  • What’s already known
  • Missing pieces (gaps)
  • Research objectives

3) Systematic review workflow using AI (a step-by-step, reproducible approach)

The instructor demonstrates an AI-assisted pipeline for systematic reviews.

AI is used to:

  • Generate a search strategy
  • Iteratively filter results

It outputs:

  • Search strategy strings for different databases (e.g., Scopus, Web of Science)
  • Inclusion/exclusion criteria
  • A table/log showing attempts, filters, and resulting hit counts (e.g., ~54 results)
  • Extracted fields per included record (including “click” types and extraction components)

Critical cautions:

  • You must still understand the domain to judge whether criteria are correct.
  • AI output must be reproducible (rebuilding the same search string should yield the same results).
  • Connectivity matters (e.g., forgetting to connect to the Pub/AI database).

Detailed AI systematic review methodology (as described/demonstrated)

  1. Run AI initial search planning
    • Generate a table/log containing:
      • color coding (search criteria types)
      • search strategy
      • search logs/attempts
      • number of filters applied
      • resulting number of records
  2. Review/adjust database query inputs
    • Modify criteria if important concepts are missing (e.g., add “traditional medicine” if needed).
  3. Apply inclusion/exclusion criteria
    • Must explicitly cover: subject matter, concept, context, design, plus:
      • time
      • language
      • document type (e.g., whether full text is possible)
  4. Construct and reuse search strings
    • AI provides search strings for databases (example: Scopus/Web of Science).
  5. Execute retrieval
    • Ensure proper connection/auth (e.g., “forgot to connect” → connect → run again).
  6. Screening & extraction preparation
    • AI proposes screening/extraction fields and outputs.
    • Emphasis: systematic reviews still require human verification/screening, ideally by more than one person.

4) Methodology writing principles (internal/external validity + reproducibility + evidence)

The instructor provides core principles for writing the Methodology section.

  • Principle A: Enough detail to assess validity
    • Internal validity (intrinsic logic): reasonable logic about causal/measurement correctness.
    • Measurement specificity, including examples:
      • device brand/model
      • posture (lying/sitting/standing)
      • timing (morning vs afternoon)
      • number of measurements and averaging
      • calibration and procedure standardization
    • External validity:
      • describe target population and sample selection criteria
      • include sample size and sampling method so others can judge appropriateness
  • Principle B: Step-by-step reproducibility
    • Write methodology like an instruction sequence (step 1 → step 2 → step 3…), so another researcher can recreate the process and obtain similar outcomes.
  • Principle C: Evidence + citations
    • Back protocols/analysis methods with references:
      • cite measurement protocols
      • cite statistical methods (e.g., references for DAG methods)

Transparency components that increase trust:

  • Provide code (analysis/reproducibility code)
  • Provide supplementary material, checklists
  • Data-sharing may be restricted, but code transparency is strongly encouraged

Detailed “Methodology section” structure guidance (as stated)

  • Methodology should contain required headings (a “10 headings” set is mentioned, though not all are enumerated in the subtitles).
  • The instructor repeatedly stresses including:
    • study design and study setting/time frame
    • variable definitions
    • sample size and sampling method
    • data analysis plan
    • ethics considerations
    • reproducible step-by-step procedures
  • Variable definition should include at least ~5 items:
    1. variable name
    2. variable value type/classification (e.g., quantitative/qualitative/nominal)
    3. measurement/data collection method
    4. tools/instruments used
    5. procedure context/administration details (e.g., validated translated questionnaire in Vietnam)

5) Publication strategy: story, novelty, and combining study types

The instructor emphasizes:

  • Successful articles tell an engaging “story” with practical relevance.
  • Use innovative framing even if the data isn’t “extraordinary.”
  • Combine formats (examples mentioned):
    • literature review + case report (“report-metalis” concept)
    • in clinical research, select a small subset of special samples when full datasets/tests are too expensive
  • “Forest vs trees”:
    • Avoid becoming only hyper-specialized; build T-shaped expertise:
      • deep domain specialization (vertical)
      • broad related-field knowledge (horizontal), including AI/biotech/etc.

6) Abstract writing rules (stands alone + structured content)

Main requirements:

  • The abstract must stand independently (title/abstract should determine whether readers continue).
  • Don’t refer vaguely to “Table 2” without explaining its meaning.
  • Write the abstract after completing the paper for coherence.
  • Abstract structure should include:
    • background
    • research gaps
    • research objectives
    • methods (subjects, sample size, main outcomes)
    • results with specific numeric/statistical statements
    • conclusions + keywords
  • A suggested guideline: a structured 8-sentence framework (including a word allocation rule—about 80–100 words for a specific resource section within a ~300-word abstract).

7) Title/headline strategy (3 types + writing constraints)

Three main headline types:

  1. Descriptive title
    • Describes the relationship/topic and may include BICO/4W1H-style elements.
  2. “Manifesto” / main result title
    • States the main finding directly.
  3. Question title
    • Asks an engaging question.

Title constraints:

  • Prefer few words (roughly 10–18 words mentioned)
  • Avoid abbreviations (except widely known ones)
  • Avoid vague terms and overly technical jargon
  • Use strong important words early
  • Use context (city/place) if it matters; may omit if too narrow (e.g., very small district)
  • For articles, a colon may be used to showcase strengths (method strength or nationwide scope)
  • For theses, safety/neutral phrasing is preferred

8) Building AI “systems” / personal assistants for research writing

The instructor demonstrates an approach to build AI as a structured assistant.

Workflow concept:

  • Write a system prompt (the “system”) specifying:
    • role(s) (e.g., international research writing assistant)
    • rules:
      • differentiate data vs interpretation
      • ensure consistency
      • adhere to standards
      • ensure completeness
    • process outputs (outline, section-by-section writing, peer-review edits, etc.)
  • Optionally upload resources (skills document/book/materials)

It can then:

  • Generate structured drafts using the expected paper format
  • Perform peer-review-style edits when a user uploads a manuscript
  • Stay consistent across tasks

Contrast highlighted:

  • Building a more automated “skill system” (in the demo)
  • Using cheaper paid assistants (e.g., generic ChatGPT-like services) with system prompts

Detailed “system prompt / assistant” methodology (as described)

  • Create a “system” specifying:
    • differentiate data vs interpretation
    • ensure consistency
    • adhere to writing/reporting standards
    • enforce stepwise/procedural completeness
  • Provide required user inputs:
    • topic
    • research objectives
    • relevant literature/resources
    • (for peer review) upload manuscript + needed context
  • Output expectations:
    • outline per paper sections and subsections
    • step-by-step writing structure for intro/methods/results/discussion
    • checklists/standards alignment
    • revised edit suggestions

9) Practical classroom logistics + copyright notes

  • Journal submission logistics include providing:
    • manuscript
    • tables/figures
    • supplement files
    • code/checklists
  • Copyright restrictions noted:
    • recorded lecture files cannot be shared
    • slides/exercise/lecture files may be shared with permission/rights (and proper credit)
  • Students are encouraged to ask questions and follow up via email.

Speakers / sources featured (explicitly mentioned)

  • The course instructor / “teacher” (name not clearly stated)
  • Ms. Linh (student participant; mentioned multiple times)
  • The five student groups (Groups 1–5; unnamed individuals)
  • Professor Nguyen Van Tuan (referenced as an example for a descriptive title; not otherwise speaking)
  • Ho Chi Minh City and Can Tho (contextual examples)
  • Scopus (database)
  • Web of Science (database)
  • PRISMA (mentioned as a systematic review standard)
  • CONSORT (mentioned for RCT guidelines)
  • Corel / Corel-like tool (mentioned for qualitative research; exact product unclear)
  • GPT / ChatGPT-like systems and related AI tools (general references; e.g., “Gemini” mentioned)
  • Pub / PubMed (referred to as “pub network” / connection point)
  • US copyright law (general legal source, not a person)

No other clearly identifiable named speakers are distinguishable from the subtitle text.

Original video