A Writing Framework for Literature Review Papers in SCI Journals

“We haven’t reached communism yet, so we still have to be practical.” —— China Produced a Mao Zedong (documentary)

Keep your feet on the ground and look up at the stars; the thinking of great people is always worth learning from.

Preface

As the earlier post “A Writing Framework for Original Research Papers in SCI Journals“ showed, in an original research paper the title, abstract, introduction, discussion and conclusion are fairly fixed and can be written from a template. Whether you are writing a research paper or a review, if you have no idea where to start, or you don’t know the field and many technical terms are unclear, find excellent review papers related to your topic and read them. Make sure you understand every theory and every term, and read the review carefully to the end; you will then usually have a first idea of how to write about the topic.

Literature review papers have even fewer variables but are basically the same: the title, abstract, introduction, discussion and conclusion are fairly fixed and can follow a template, while the middle part (the specific quantitative and qualitative analysis) is where we can use creativity and imagination on top of the template.

Before we begin, one thing needs to be clear: since all reviews are variations on the same theme, getting into a good journal means writing a better review than others. More broadly, whether you write a review paper or the review part of a research paper’s introduction, the core competitive advantage comes from the analysis methods. The outcome of reviewing a research topic is simply: what has been done, what hasn’t been done, and what can still be done.

Suppose you use newer or more advanced mathematical statistics, bibliometric analysis or computational models to reach conclusions, and then analyse those statistical conclusions further to reach deeper ones (directions already studied, directions not studied, directions that could still be studied, directions that can’t be pursued, the pros and cons of each direction, and why they are worth pursuing). After all, academia is like nesting dolls. Remember: a review that merely lists papers without deeper conclusions of its own is meaningless and won’t be accepted.

The purposes of a review are:

(1) To show readers clearly how the topic has developed

(2) To show readers the existing methods and principles of the topic and the progress of each method

(3) To show readers the trends, directions, gaps and possibilities of the topic

(4) To analyse existing research using mathematical, modelling, bibliometric and other statistical methods

As for the number of references: an English review should cite at least 100 papers, and a Chinese review at least 80.

As in the previous post, we divide the paper into parts and analyse the review framework from beginning to end.

Structure of a Literature Review Paper in an SCI Journal

The structure of a review paper is very simple:

Abstract, Introduction, Methods, Quantitative Bibliometric Analysis, Qualitative Subtopic Analysis, Mathematical Statistical Analysis, Discussion, Conclusion

As with research papers, the writing order is usually middle first, then the two ends:

1. Methods

2. Quantitative Bibliometric Analysis

3. Qualitative Subtopic Analysis

4. Mathematical Statistical Analysis

5. Comparative Analysis

6. Discussion

7. Conclusion

8. Introduction

9. Abstract

We will mainly take apart and analyse two example papers:

(1) A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications [1] This is a typical review whose analysis consists only of several subtopic analyses.

(2) Comprehensive analysis of the relationship between thermal comfort and building control research - A data-driven literature review [2] This review combines bibliometric analysis with other quantitative statistical methods.

Again, we deconstruct the framework in reading order.

Title

As in “A Writing Framework for Original Research Papers in SCI Journals“, the title is the first thing reviewers see and the first criterion for deciding whether to review or reject the paper.

A review title can be built by stacking template parts, or written freely. But it must be specific to a research topic, not something general. For example, “Research progress on load-bearing behaviour of precast beams in prefabricated buildings“ won’t do. We need to narrow the title to the specific content and topic, for example: “Research progress on the load-bearing behaviour of (horizontal/vertical or type of) precast beams in prefabricated buildings (under XX conditions) (under monotonic/cyclic or other loading)”.

Template	Title
Free-style	Exploring an unknown territory: “Sleeping Beauties” in the nursing research literature [3]
Straight to the point	Structure and patterns of cross-national Big Data research collaborations. [4]
Topic + analysis method + free-style	The Circular Economy Umbrella: Trends and Gaps on Integrating Pathways. [5]
Topic + literature review	Reverse logistics for the end-of-life of and end-of-use products in the pharmaceutical industry: A systematic literature review. [6]
Research question + method (including methods within methods)	Is autoimmunology a discipline of its own? A big data-based bibliometric and scientometric analyses. [7]
Topic + innovative method	Climate change and viticulture-a quantitative analysis of a highly dynamic research field. [8]
…	…

If you have no good ideas for your paper, in the end it is because you haven’t read enough. Journal papers are not that rigid: you can use metaphors and other imaginative devices in the title. Of course, think twice and weigh the pros and cons.

Also, a review doesn’t have to analyse only papers. Patents can be used in a review too, and you can even write a review of patents alone. For example:

Patent citations analysis and its value in research evaluation: A review and a new approach to map technology-relevant research. [9]

So new methods and new ideas matter a lot. A good topic alone can make a very good paper, research or review alike: if nobody has studied it and you thought of it, it is a good topic (as long as it stays realistic). Here is an example:

ChatGPT or Bard: Who is a better Certified Ethical Hacker? [10]

That is where a good topic comes from: accessible tools, plus some novelty, plus a question nobody has studied.

Abstract

Taking “Comprehensive analysis of the relationship between thermal comfort and building control research - A data-driven literature review” as an example, let’s look at how its abstract is built:

1. Background
2. Statement of the problem
3. The approach to the problem
4. The review method
5. Overview of the review procedure
6. Results, conclusions and directions of the review

Buildings are responsible for about 30–40% of global energy demand. At the same time, we humans spend almost our entire life, up to 80–90% of the time, inside of buildings. Reducing energy demand through optimal operation is the subject of building control research, while human satisfaction in buildings is studied in the thermal comfort community. Thus, balancing the two is necessary for a sustainable and comfortable building stock. We review both research fields and their relationship using a data-driven approach. Based on specific search terms, all relevant abstracts from the Web Of Science database are downloaded and analyzed using the text mining software VOSviewer. We visualize the scientific landscapes of historic and recent trends, and analyze the citation network to investigate the interaction between thermal comfort and building control research. We find that building control focuses predominantly on energy savings rather than incorporating results from thermal comfort, especially when it comes to occupant satisfaction. We identify potential research directions in terms of bridging the two fields.

1. Opening background

“Buildings are responsible for about 30–40% of global energy demand. At the same time, we humans spend almost our entire life, up to 80–90% of the time, inside of buildings.”

It sets out the technical background and uses percentages to draw the reader in.

2. Statement of the problem

“Reducing energy demand through optimal operation is the subject of building control research, while human satisfaction in buildings is studied in the thermal comfort community.”

3. The approach to the problem

“Thus, balancing the two is necessary for a sustainable and comfortable building stock.”

4. The review method

“We review both research fields and their relationship using a data-driven approach. Based on specific search terms, all relevant abstracts from the Web Of Science database are downloaded and analyzed using the text mining software VOSviewer.”

5. Overview of the review procedure

“We visualize the scientific landscapes of historic and recent trends, and analyze the citation network to investigate the interaction between thermal comfort and building control research.”

6. Results, conclusions and directions

“We find that building control focuses predominantly on energy savings rather than incorporating results from thermal comfort, especially when it comes to occupant satisfaction. We identify potential research directions in terms of bridging the two fields.”

Introduction

The introduction of a review paper is similar to that of a research paper. The flow is roughly:

1. Background: tell a story and raise the problem: in …, problem “Q0” exists.

   Topic background and importance: the introduction first stresses the importance of problem Q0 and the need for ….

2. History of the problem: cite the literature and review how problem Q0 has developed.

3. Review and analysis of existing traditional solutions: review the literature, then analyse the strengths and weaknesses of each method.

   Introduce the traditional methods and review them: because of limitations in … and specific …, such as low … leading to … (give concrete engineering or research cases), the traditional … has problem Q1 / is limited by ….

   Then review each method that addresses Q1 and analyse its strengths and weaknesses.

   Finally point out that, despite this progress, … problems remain.

   And explain the mechanism behind the problem: why can’t it be solved yet?

4. Analysis of existing reviews: review earlier scholars’ reviews and analyse them.

   Describe and analyse previous reviews. If our review covers two fields, analyse the existing reviews in each field separately, and also the existing reviews that combine both. That is:

   (1) Review existing reviews in field A, then point out their shortcomings

   (2) Review existing reviews in field B, then point out their shortcomings

   (3) Review existing reviews combining fields A and B, then point out their shortcomings

5. Purpose and structure of this review:

   Highlight the strengths and weaknesses of earlier reviews, then state the direction of this review, and finally describe the overall structure of the paper.

   Taking “Comprehensive analysis of the relationship between thermal comfort and building control research - A data-driven literature review” as an example (paraphrased):

   “Previous reviews analysed each research field independently or focused on specific examples. However, as highlighted above, energy-efficient operation cannot be achieved without considering human comfort, which is itself a complex topic. The aim of this review is therefore to give a comprehensive overview by (1) analysing historical developments and recent trends, (2) investigating how the two fields interact through the citation network, and finally (3) identifying gaps in the literature on thermal comfort and building control.

   The rest of the paper is organized as follows. Section 2 describes the method of the data-driven literature survey. In Section 3 we first analyse the publications quantitatively, then describe historical developments and recent trends using scientific landscapes and citation networks. We discuss our findings in Section 4 and conclude in Section 5.”

Methodology

Describe the review methodology in detail, including how relevant literature was selected, the keywords used and the screening process. This gives readers transparency and makes the study reproducible.

1. First state which methods are used and how, the general order in which they are applied and the structure of the study, with a research framework diagram.

2. State the principles of data collection, including the criteria for selecting literature, the sources, the keywords and the screening process.

3. In several subsections, explain each analysis method used and how it is applied to the data in this paper.

Quantitative Bibliometric Analysis

Bibliometric analysis reveals the research trends, development dynamics, hot topics and academic influence of a field or topic. After each part of the bibliometric analysis, include a chart of the quantitative results and discuss the visualized data. Common parts of a bibliometric analysis are:

1. Publication output analysis

   Counting publications over a given period shows how active research is, and whether it is growing or declining. It shows how the field changes over time.

2. Journal and publication analysis

   Evaluate the publication volume, impact factor and citations of different journals to find which have the highest reputation and influence.

3. Co-authorship network analysis

   Studying collaborations between authors builds a co-authorship network. This reveals patterns of collaboration and the structure of research groups, and shows the breadth and depth of collaboration. Network analysis of collaboration and citation networks reveals the relationships between authors, institutions or countries and the flow of knowledge in a field. Common network metrics include degree, centrality, density and community structure.

4. Citation analysis

   Citation analysis studies how often and in what patterns papers are cited, which helps assess research impact and the quality of the literature. The most cited papers usually represent foundational or breakthrough work in the field.

5. Keyword and topic trend analysis

   Analysing the keywords and topics in the literature identifies hot issues and trends, and helps predict future research directions.

6. Funding and output correlation analysis

   Analyse the relationship between research funding and output to see how much funding affects research activity.

7. Geographic and institutional distribution analysis

   Study the output and collaboration of specific regions or institutions to reveal geographic concentration and patterns of exchange between institutions.

8. Highly cited researchers and teams analysis

   Identify researchers and teams that are highly active or especially influential in one or several fields.

Here are some common statistical methods that help researchers understand the trends, networks and influence of the literature:

1. Descriptive statistics

   Descriptive statistics describe basic features of the literature set, such as the number of papers, publication years and citation counts. Common measures include the mean, standard deviation, median, frequencies and percentages.

2. Network analysis

   Network analysis studies collaboration and citation networks and reveals relationships between authors, institutions or countries and the flow of knowledge in a field. Common metrics include degree, centrality, density and community structure.

   (1) Degree: the number of edges directly connected to a node. In a collaboration network, a higher degree means the author, institution or country collaborates directly with more others. Analysis: compute the degree of each node to identify the most active researchers or most central institutions.

   (2) Centrality: measures how central a node is in the network. Common measures are closeness centrality, betweenness centrality and eigenvector centrality. Use graph software or libraries (such as NetworkX in Python) to compute the centrality of each node and identify the most influential researchers or core institutions.

   • Closeness centrality: based on the average shortest-path length from a node to all other nodes. The lower the value, the more central the node.
   • Betweenness centrality: measured by the number of shortest paths that pass through a node. The more shortest paths pass through it, the higher its betweenness.
   • Eigenvector centrality: based on the importance of a node’s neighbours; a node connected to several high-centrality nodes also has high eigenvector centrality.

   (3) Density: the ratio of actual edges to possible edges. Higher density means closer collaboration among nodes. Analysis: compute density to assess how tightly connected the collaboration is overall; high density may indicate a close-knit research community.

   (4) Community structure: nodes naturally fall into groups, with denser connections within groups than between them. Analysis: use community detection algorithms (such as the Louvain method or Girvan-Newman algorithm) to identify communities, revealing subgroups or research areas in the collaboration.

3. Cluster analysis

   Cluster analysis groups papers by topic or keyword, helping to identify the main research themes or trends in a field.

   Common techniques include k-means, hierarchical clustering and density-based methods (such as DBSCAN).

   (1) K-means clustering

   • Concept: a partitioning method that divides data into K sets, minimizing the sum of squared distances between each point and its nearest mean (cluster centre).
   • Steps:
     Choose the number of clusters (K): this may need prior knowledge or methods such as the Elbow Method.
     Initialize centres: randomly choose K data points as initial centres.
     Assign points: assign each point to the nearest centre.
     Recompute centres: update each centre to the mean position of the points in its cluster.
     Iterate: repeat assignment and update until the centres no longer change significantly.
     Analyse results: interpret which research theme each cluster may represent.

   (2) Hierarchical clustering

   • Concept: organizes data by building a cluster tree (dendrogram), where each node represents a cluster. It can be bottom-up (agglomerative) or top-down (divisive).
   • Steps:
     Build a similarity matrix: compute similarities between all papers (usually based on keywords, abstracts, etc.).
     Merge or split: in the agglomerative approach each paper starts as its own cluster and the most similar clusters are merged step by step; in the divisive approach all papers start in one cluster that is split step by step.
     Build the dendrogram: record the clustering process as a hierarchical tree.
     Prune: cut the tree at a suitable level.
     Analyse and interpret: read the dendrogram to identify themes or trends.

   (3) Density-based clustering (such as DBSCAN)

   • Concept: DBSCAN (Density-Based Spatial Clustering of Applications with Noise) relies on the density of data points, can form clusters of any shape, and handles noise and outliers.

   • Steps

     Set parameters: choose the radius (ε) and minimum number of points (MinPts), which define the density of a region.

     Classify points: label points as core, border or noise.

     Form clusters: start from a core point, add all directly density-reachable points, and recursively add reachable core points and their neighbours to form a cluster.

     Handle noise: identify and handle points that belong to no cluster.

     Analyse results: interpret each cluster and identify core themes and trends.

So how are these methods applied in bibliometrics? Here are the steps:

(1) Data preparation: collect metadata of relevant papers from databases (such as Web of Science, Scopus, PubMed), including titles, abstracts, keywords, authors, citation counts and publication years.

(2) Feature selection: decide which features to cluster on; common ones are keywords, citation patterns, author links or research topics.

(3) Choose a clustering algorithm:

• K-means: suits large data sets and requires the number of clusters in advance; good when the number of research areas is known.
• Hierarchical clustering: needs no preset number of clusters and shows results as a dendrogram; good for exploratory analysis of hierarchical structure.
• DBSCAN: suits large data sets with irregular cluster shapes or noise; needs no preset number of clusters.

(4) Evaluate and interpret results: assess cluster quality, for example with the Silhouette Coefficient for compactness and separation. Interpret which topic or trend each cluster represents, possibly based on its representative papers or keywords.

(5) Visualization: use tools (such as OriginLab, Gephi, Tableau or Python’s Matplotlib and Seaborn) to visualize the clusters, for example keyword co-occurrence networks or cluster scatter plots. Use the clusters to identify new trends, core research groups or potential research gaps.

4. Time series analysis

   Time series analysis suits studying how publication and citation counts change over time. Autoregressive (AR), moving average (MA) or autoregressive moving average (ARMA) models can be used.

5. Citation analysis

   Citation analysis evaluates the citation counts of papers or authors and identifies highly cited papers and core authors, using indicators such as the h-index, g-index and citation half-life.

6. Factor analysis

   Factor analysis studies latent relationships between variables, for example the underlying factors affecting how often papers are cited. It reduces dimensionality and helps understand the structure behind the data.

7. Correlation analysis

   Correlation analysis assesses how strongly two or more variables are related, for example the number of papers and the intensity of a science policy or global sustainability policy. Common methods are the Pearson correlation coefficient and Spearman’s rank correlation coefficient.

8. Multiple regression analysis

   Multiple regression predicts or explains the relationship between one variable (such as citation count) and several predictors (such as paper length, journal impact factor and number of authors).

9. Principal component analysis (PCA)

   PCA explores and visualizes the structure of high-dimensional data. It suits large literature sets and helps identify the most important patterns and structures in the data.

These methods can be used alone or together to give comprehensive insight into a body of literature. The right choice depends on the research question, the data type and the software available.

Qualitative Subtopic Analysis

This is probably the longest part of the whole paper.

Purpose: to explore each subtopic of the field in depth and to understand and explain data and phenomena qualitatively.

The framework for this part is flexible; roughly:

1. Introduce the subtopic: give its background, why it matters and the related research questions.

2. Analyse the subtopics in subsections: each subsection analyses one subtopic, and each paragraph within it focuses on one research point or argument, supported by citations and data. Don’t just summarize; analyse critically, discuss different views, and identify differences and connections between studies.

3. Visualize numerical performance results of specific aspects of the methods: put them in tables or charts and analyse them.

Mathematical and Statistical Analysis

This part summarizes the data results of the main papers covered in the qualitative analysis, then processes their data further yourself with mathematical analysis to extend their results.

Choose statistical methods that fit the research question and data type. Common ones include descriptive statistics, inferential statistics, regression analysis, analysis of variance (ANOVA) and correlation analysis. Consider the data distribution, sample size and purpose of the study when choosing.

Statistical results need visualization, for example bar charts, box plots and scatter plots. Visualization helps explain the results and is an effective way to show trends and patterns. Interpret the results and explain what they mean for the study. Discuss how far the findings support or refute the hypotheses, analyse relationships between variables and possible causes and mechanisms. Acknowledge possible limitations such as sample bias, measurement error and external factors, and discuss how they might affect the results. Based on the results, propose possible future directions or improvements, for example a larger sample or different methods to verify the findings.

Comparative Analysis

1. Choose what to compare: select the research results to compare, and which data results. Usually these are studies that differ clearly in methodology, experimental design, techniques or conclusions. Define the dimensions of comparison, such as theoretical basis, research method, data collection technique, analysis method and interpretation of results.
2. Describe the comparison method: explain the method and criteria, for example whether statistical methods are used to compare data sets or a qualitative comparison based on expert opinion. If applicable, describe the data processing and analysis tools used, such as statistical software and visualization tools.
3. Present the comparison: use tables, charts or text. Tables and charts show differences between studies more intuitively. The comparison can include numerical comparisons (means, percentages, standard deviations, etc.) and qualitative comparisons (applicability, efficiency, reliability of methods, etc.).
4. Analyse the findings: explain what the comparison means and discuss consistencies and differences between studies. Point out the strengths and limitations of each method or result, and how these differences may affect the field.

Discussion & Conclusion

In my view, because the middle of a review paper is already long, the discussion and conclusion should be written together. The goal is to briefly review the whole paper, integrate the findings, stress the academic contribution and propose future research directions.

This is similar to the discussion and conclusion in “A Writing Framework for Original Research Papers in SCI Journals“, with some differences:

1. Summary of the review
   • Literature summary: summarize the key findings and themes of the reviewed literature, and analyse the research trends and the main theories and methods in the field.
   • Main conclusions: outline the important insights and common problems identified, and their impact on related fields.
2. Evaluation of performance and methods
   • Theoretical evaluation: assess the efficiency and effectiveness of common methods in existing research and their limitations for specific problems.
   • Method comparison: analyse and compare different researchers’ methods and results, pointing out their strengths and weaknesses.
3. Advantages of methods and theories
   • Method advantages: “Compared with method …, method … simplifies data processing and reduces operational complexity.”
   • Efficiency and simplification: stress how new theories or methods improve research efficiency and simplify the steps, and their potential value in practice.
4. Limitations and future research
   • Acknowledge limitations: honestly discuss the limitations of the review, such as incomplete coverage of the literature or biases in some methods.
   • Suggestions for future research: “Future research should focus on improving the breadth and depth of data collection and on developing more accurate analysis models to make the results more generally applicable.”

References

1. Avci, O., Abdeljaber, O., Kiranyaz, S., Hussein, M., Gabbouj, M., & Inman, D. J. (2021). A review of vibration-based damage detection in civil structures: From traditional methods to Machine Learning and Deep Learning applications. In Mechanical Systems and Signal Processing (Vol. 147, p. 107077). Elsevier BV. https://doi.org/10.1016/j.ymssp.2020.107077
2. Park, J. Y., & Nagy, Z. (2018). Comprehensive analysis of the relationship between thermal comfort and building control research - A data-driven literature review. In Renewable and Sustainable Energy Reviews (Vol. 82, pp. 2664–2679). Elsevier BV. https://doi.org/10.1016/j.rser.2017.09.102
3. Kokol, P., Blažun Vošner, H., & Vermeulen, J. (2017). Exploring an Unknown Territory. In Nursing Research (Vol. 66, Issue 5, pp. 359–367). Ovid Technologies (Wolters Kluwer Health). https://doi.org/10.1097/nnr.0000000000000238
4. Hu, J., & Zhang, Y. (2017). Structure and patterns of cross-national Big Data research collaborations. In Journal of Documentation (Vol. 73, Issue 6, pp. 1119–1136). Emerald. https://doi.org/10.1108/jd-12-2016-0146
5. Homrich, A. S., Galvão, G., Abadia, L. G., & Carvalho, M. M. (2018). The circular economy umbrella: Trends and gaps on integrating pathways. In Journal of Cleaner Production (Vol. 175, pp. 525–543). Elsevier BV. https://doi.org/10.1016/j.jclepro.2017.11.064
6. Campos, E. A. R. de, Paula, I. C. de, Pagani, R. N., & Guarnieri, P. (2017). Reverse logistics for the end-of-life and end-of-use products in the pharmaceutical industry: a systematic literature review. In Supply Chain Management: An International Journal (Vol. 22, Issue 4, pp. 375–392). Emerald. https://doi.org/10.1108/scm-01-2017-0040
7. Watad, A., Bragazzi, N. L., Adawi, M., Amital, H., Kivity, S., Mahroum, N., Blank, M., & Shoenfeld, Y. (2017). Is autoimmunology a discipline of its own? A big data-based bibliometric and scientometric analyses. In Autoimmunity (Vol. 50, Issue 4, pp. 269–274). Informa UK Limited. https://doi.org/10.1080/08916934.2017.1305361
8. Marx, W., Haunschild, R., & Bornmann, L. (2017). Climate change and viticulture-a quantitative analysis of a highly dynamic research field. Vitis, 56(1). https://doi.org/10.5073/vitis.2017.56.35-43
9. Van Raan, A. F. J. (2017). Patent Citations Analysis and Its Value in Research Evaluation: A Review and a New Approach to Map Technology-relevant Research. In Journal of Data and Information Science (Vol. 2, Issue 1, pp. 13–50). Walter de Gruyter GmbH. https://doi.org/10.1515/jdis-2017-0002
10. Raman, R., Calyam, P., & Achuthan, K. (2024). ChatGPT or Bard: Who is a better Certified Ethical Hacker? In Computers & Security (Vol. 140, p. 103804). Elsevier BV. https://doi.org/10.1016/j.cose.2024.103804