Literature Review Evidence Management: From Search Results to Synthesis

A literature review is not a pile of PDFs. It is a chain of decisions: what to search, what to keep, what to read first, what to extract, how to compare studies, and how to turn those comparisons into an argument. When the chain breaks, deadlines become chaotic and your rubric suffers. This guide connects deduplication, screening, reading priority, extraction, evidence matrices, argument maps, and reference-library structure into one navigable workflow. It also shows where WisPaper tools fit at specific steps and ends with a practical submission decision.

Before adding more papers, identify the bottleneck. Most review problems fall into one of six stages. Use this table to diagnose your situation and jump to the relevant section.

If you are entering at the search stage, begin with the next section. If you already have a library, skip to the section on reference-library structure and work backward.

Cochrane guidance recommends planning searches from eligibility criteria, using multiple sources, combining free-text terms with subject headings, documenting the process, deduplicating records, screening in stages, linking multiple reports of one study, and rerunning searches before publication. For a course paper or first manuscript, this translates into a simple rule: decide what counts as relevant before you search, not after.

Write a one-paragraph eligibility statement. It should name the population, intervention or phenomenon, study designs you will accept, date range, and language limits. Then translate that statement into search terms. PubMed supports field tags, phrase searching, filters, Advanced Search, saved searches, Similar Articles, citation lookup, and links to full text when available; its search details reveal how terms were translated and combined. Google Scholar documents searching by author and title, date sorting and date limits, Cited by and Related articles, citation export, library access links, email alerts, and ways to locate alternative versions of a work. Use these features to make your search reproducible, not to accumulate endless results.

When you need a scholarly search interface that keeps results in one place, WisPaper search provides a starting point for finding academic papers. For a public explanation of finding papers, see the WisPaper Scholar Search entry point. If you are working in Chinese, the Chinese Scholar Search page offers the same entry point in that language.

Deduplication is the cheapest way to reduce screening load. Cochrane recommends deduplicating records as part of the search and selection process. In practice, export your results into a reference manager and remove exact duplicates by DOI or title. Then look for near-duplicates: conference papers that became journal articles, preprints that were later published, and multiple reports of the same study. Cochrane also recommends linking multiple reports of one study, so keep a note when two records describe the same underlying project.

A simple deduplication log helps. Record the date, databases searched, number of records before and after deduplication, and the reason for any manual removals. This log becomes part of your methods section and makes your process transparent, which NIH guidance treats as core to rigorous research practice.

Cochrane recommends screening in stages. A workable student version is three passes:

1. Title and abstract screen. Remove records that clearly fail your eligibility criteria. Keep a reason code for each exclusion.
2. Full-text screen. Read the methods and results sections of remaining records. Exclude those that do not meet design, population, or outcome criteria.
3. Final inclusion check. Confirm that each included record has the information you need for extraction. If not, mark it as incomplete and decide whether to request it through your library.

Use a screening spreadsheet with columns for record ID, decision, reason, and notes. This is not bureaucracy; it is the evidence trail your supervisor or marker may ask for. If you are working with a supervisor, share the spreadsheet early so feedback arrives before you have read everything.

You cannot read everything closely. Rank included records by two dimensions: relevance to your question and study type. A systematic review or meta-analysis may give you the landscape quickly. A primary study that directly tests your question deserves close reading. A tangential study can be read for background only.

A reading-priority matrix helps:

When you need to ask questions based on papers you have already saved or added from search results, Library QA supports that step. It lets you query your library rather than re-reading every PDF from the beginning.

Extraction turns reading into usable evidence. Build a template before you read your first full text. A minimal template includes:

• Citation and study ID
• Research question or objective
• Population and setting
• Design and sample size
• Intervention or phenomenon
• Main outcomes and measures
• Key findings relevant to your review
• Limitations stated by the authors
• Notes on relevance to your argument

Keep one row per study in a spreadsheet or table. If a study reports multiple outcomes, add columns rather than separate rows, so the study remains the unit of analysis. This matters when you later link multiple reports of one study, as Cochrane recommends.

For longer research tasks that involve literature review, inspiration, or reproduction use cases, Scholar Agent provides a workspace for those tasks. If your extraction reveals a paper you need to reproduce or plan around, PaperClaw supports research-paper reproduction planning from an uploaded PDF and reproduction instructions.

An evidence matrix is an extraction table with a comparative purpose. It should let you see patterns across studies at a glance. Columns typically include study, design, sample, context, key finding, effect direction, and limitations. Rows are studies. If you have many studies, group them by theme or design.

The matrix does three jobs. First, it exposes gaps: if no study addresses a sub-question, that is a finding. Second, it prevents overgeneralization: you can see whether a claim rests on one study or several. Third, it gives you the raw material for synthesis. NIH guidance treats careful assessment of prior knowledge, strengths, weaknesses, and knowledge gaps as part of rigorous practice; the matrix is where that assessment becomes visible.

When you need to save or upload papers and organize them in the WisPaper library, My Library provides that structure. Keep your matrix separate from the library; the library stores papers, while the matrix stores your judgments about them.

An argument map connects your claims to evidence and gaps. Start with your central claim, then list the sub-claims that support it. For each sub-claim, note which studies support it, which qualify it, and which contradict it. This is where the evidence matrix pays off: you can move rows into argument positions.

A simple argument map has three layers:

1. Claim layer: what you will argue.
2. Evidence layer: which studies support or qualify each claim.
3. Gap layer: what remains unresolved and why it matters.

If your map shows that a sub-claim has only one supporting study, narrow the claim or mark it as tentative. If two studies conflict, do not average them; describe the conflict and explain possible reasons such as differences in population or measure.

A reference library should help you find the right paper at the right moment. Organize it by project, then by theme, then by status. Status categories might include: to screen, included, extracted, cited, and background. Use consistent naming and tags. If you use a reference manager, keep the same structure there.

For a new project, Research Projects provides a place to create and manage the project. For citation screening, TrueCite checks BibTeX against real academic databases to flag hallucinated references. Run this check before submission, especially if you have assembled citations from multiple sources.

If you are starting from scratch with a deadline, follow the full chain: plan search, deduplicate, screen, prioritize, extract, matrix, argument map, library structure. If you already have a library but no synthesis, start at extraction and build the matrix. If you have a matrix but no argument, start at the argument map. If you are revising after supervisor feedback, start at the diagnostic table and fix the weakest link first.

Consider a concrete example. A student has two weeks to submit a course paper. The rubric rewards a clear research question, critical synthesis, and proper citation. The student searches two databases, exports results, deduplicates, and screens titles and abstracts. They prioritize ten studies for full reading, extract them into a matrix, and map two claims and one gap. They organize the library by theme and status, then run a citation check. With three days left, they draft from the argument map rather than from the PDFs.

Before you submit, make one decision: is your evidence chain complete enough for the claim you are making? Check four things. First, every major claim in your draft has at least one supporting study in your matrix. Second, every included study has an extraction row. Third, your search and screening log is complete enough to describe in your methods. Fourth, your citations have been checked. If any of these is missing, fix that link before polishing prose. If all four are present, submit with a clear conscience and a documented process.