International Open Access Journal Platform

1 Introduction

The importance of research papers for universities and research institutions is undeniable, and a large body of research exists on how to write them. Genre analysis represents a dominant framework for research article examination, revealing the rhetorical structure and core linguistic features of each textual section. Research on the genre of research papers, both domestically and internationally, has primarily been conducted from two perspectives: cross-linguistic and cross-cultural comparisons (Ahmad & Amira, 2012; Samraj, 2002; Niu, 2018) and cross-disciplinary comparisons (Samraj, 2008; Xu et al., 2007). Since Swales (1990) put forward and refined the notion of “move” for textual structural analysis and proposed his three-move Create a Research Space (CARS) model for introductions, scholars have adopted this framework across diverse languages and academic fields (Fredrickson & Swales, 1994; Ahmad, 1997), continuously revising and refining it. Among these studies, some have analyzed entire research papers using a move-step structure (Nwogu, 1997; Posteguillo, 1999), while others have focused their move analysis on specific sections of research papers (e.g., Introduction, Methods, Results, Discussion, and Conclusion) (Lim, 2006; Peacock, 2002).

Introductions constitute the most widely examined section of research articles (Swales, 2004). Located at the beginning of the paper, the introduction serves as a “roadmap” (Samraj, 2008), providing readers with the perspective and framework for interpreting the subsequent content and guiding them through the detailed information in the paper. A high-quality introduction reflects the author’s core academic abilities, including literature review, problem articulation, logical organization, concise expression, and mastery of academic conventions. Therefore, developing the ability to write effective introductions is a key skill in academic writing training.

2 Literature review

2.1 Genre analysis

Genre analysis originated in literary studies but was systematically applied to language teaching and academic contexts through the pioneering work of linguists such as John Swales, eventually developing into a key area focused on “how to organize specific types of texts for specific purposes”. Texts within an identical genre share consistent patterns with regard to communicative goals, textual organization, stylistic characteristics, and intended audiences (Swales, 1990). For the genre of the research paper, Most empirical research articles (RAs) adhere to a stable, predictable structure termed the IMRD framework and its variants: Introduction (I), Methods (M), Results (R), and Discussion (D).

2.2 The Move-Step structure of introductions

Swales (1990) proposed the “Create a Research Space” (CARS) model for analyzing academic article introductions, decomposing the text into different “moves” and “steps”. He defined a move as a functional textual unit organized to serve a specific, shared communicative purpose. Unlike a paragraph, whose boundary is determined by grammatical form, a move is distinguished by its communicative function. A move can be as short as a phrase or as long as a sentence or a paragraph. Within a move, there can be several smaller units called steps, which represent the specific rhetorical strategies used to achieve the overall function of the move. Their identification relies on analyzing the writing conventions of a specific discourse community (ibid.). An introduction’s rhetorical framework consists of three core moves: Establishing a Territory, Establishing a Niche, and Occupying the Niche. Subsequent research has revised and refined this framework (e.g., Samraj, 2008; Maswana et al., 2015; Niu, 2018).

Building on studies of the complete move structure of introductions, many papers have conducted more specific, micro-level analyses focusing on a single move within the introduction. Alanazi and Alqarni (2022) analyzed the first sentence of RA introductions in the fields of linguistics and translation studies. Swales and Najjar (1987) compared research papers in educational psychology and physics, focusing specifically on Move 3 (Occupying the Niche).

While there have been many studies on the first move (Establishing a Territory) of introductions, their findings differ significantly due to variations in terminology, among other factors. Swales (1990, 2004), in his first move “Establishing a Territory”, proposed three steps, including Step 1A: Claiming Centrality (e.g., “Recently, there has been a spate of interest in...” or “Knowledge of X has great importance for...”). The core function of this step is to demonstrate the importance or significance of the current research topic, or to show that it belongs to an active research area, thereby engaging readers and reviewers. Step 1B: Making Topic Generalizations involves stating consensus, common phenomena, or facts within the field (e.g., “Plumage coloration is known to influence mate selection in mallards”). Step 1C: Reviewing Items of Previous Research. In his 2004 revised framework, Swales observed that scholars struggle to differentiate between “Claiming Centrality” and “Making Topic Generalizations”. Accordingly, he combined the two steps under the unified label “Topic Generalizations of Increasing Specificity”.

Maswana et al. (2015), in their study of RA introductions across five engineering fields, found that the first move, “Presenting the background information,” consisted of two steps: “Reference to established knowledge in the field” and “Reference to main research problems”. The first step corresponds to Swales’ (1990) “Making topic generalizations”. The second step – reference to main research problems – was found to be a common feature in computer science introductions, as illustrated in the example below.

Move 1, Step 1: The external electroluminescence (EL) quantum efficiency (QEEL) of a polymer light-emitting diode (PLED) can be affected by the following four factors: ...Move 1, Step 2: Therefore, the dominating factor for achieving high efficiency for a given polymer is the balance and confinement of electrons and holes. Unfortunately, most conjugated polymers have unbalanced charge-transport properties as the hole mobility is much larger than the electron mobility. (EL, Ref. #2)

Table 1 Move Structures for Introduction Backgrounds Proposed by Previous Researchers

Notable discrepancies exist between the two analytical frameworks proposed by Swales (1990) and Maswana et al. (2015) for the background move within research article introductions (see Table 1). First, the review of previous research is incorporated into different moves. Second, while “Making topic generalizations” is equivalent to “Reference to established knowledge in the field”, the two researchers proposed another two entirely different steps: “Claiming Centrality” (Swales, 1990) and “Reference to main research problems” (Maswana et al., 2015). These micro-level studies of introductions were based on papers in Applied Linguistics (Swales, 1990) and five engineering fields (Structural, Environmental, Electrical, Chemical, and Computer Science) (Maswana et al., 2015). What, then, is the situation regarding the background move in introductions specifically within the field of Computer Science? This paper analyzes this question from both quantitative and qualitative perspectives. The specific research questions are:

1)What steps constitute the background move in the introductions of computer science RAs?

2)What are the specific writing strategies for these steps?

3 Research method

Twenty research articles were selected from top international computer science journals, including ACM Computing Surveys (CSUR) and the Journal of the ACM (JACM). All selected articles were coded sequentially as Paper 1 (P1) through Paper 20 (P20). Second, each paper’s introduction was analyzed using the analytical frameworks proposed by Swales (1990) and Maswana et al. (2015) to identify the background move and its steps. The primary analytical focus falls on the opening paragraph, where the background move generally resides. To identify the rhetorical function of the concluding sentence in the opening paragraph, the second paragraph is also analyzed as an auxiliary reference.

4 Results and discussion

Across the 20 analyzed research articles, all three steps appear at high frequencies (Table 2). Specifically, 80% began their introductions with a claim of centrality (Step 1). This suggests that researchers in the computer science field habitually begin their background move by stating the importance or centrality of their research area. This finding differs from Alanazi and Alqarni (2022), who found that authors in high-impact journals preferred to open their introductions with sentences fulfilling Move 2 or Move 3 functions. This inconsistency may stem from the fact that Alanazi and Alqarni’s corpus draws on linguistics and translation studies, which reveals cross-disciplinary variations in move configurations. Furthermore, while just over half (60%) of the introductions further move on to the specific research topic (Step 2), the vast majority (95%) raise a major problem within the field (Step 3). Based on the problem raised, the author smoothly transitions into a review of previous literature, summarizing how others have attempted to solve the problem. During this review, a research gap is identified, leading ultimately to the author’s own study. In this small corpus, the majority of Step 3 (76%) were found at the end of the first paragraph, with the remainder (24%) located in the second paragraph. This indicates that researchers tend to include Step 3 as a component of the first background move (Samraj, 2002; Maswana et al., 2015). These three steps function as the background or territory establishment within the first paragraph of the introduction.

Table 2 Frequency of steps in the background move of computer science introductions

In this study, the framework proposed by Maswana et al. (2015) was adopted, as it was originally developed for engineering disciplines, including computer science. Maswana et al. (2015) categorize the literature review as an independent Move 2, which differs from Swales (1990), who locates the literature review under Step 1C of Move 1. For this reason, Swales’ alternative classification was not applied in the coding process.

Based on the findings of this study, the background move in the introduction of computer science research papers consists of three steps, as shown in Table 3 below:

Table 3 Composition of the Background Move in Computer Science Research Article Introductions

Step 1: Claiming centrality

Most claims in the Claiming Centrality step address practical real-world issues (see Examples 1-3). These sentences use highly affirmative verbs that establish a firm position, directly declaring the key status of a specific technology, method, or task. Examples include “cornerstone” (Example 1), “state of the art approaches” (Example 2), and “fundamental task” and “a wide range of applications” (Example 3). Only a minority relates to issues within the research community itself (see Examples 4-5), such as “increasing attention in the research community” (Example 4) and “researchers have shown considerable interest” (Example 5). The preference for claiming real-world importance, rather than research-centric centrality, is likely because computer science is a typical engineering discipline where research is usually application-oriented and aims to solve specific, real-world problems. Such claims of centrality are often supported by citations to literature, as the statement “this field is important” is not a fact but an argument that requires justification. This finding is similar to that of Alanazi and Alqarni
(2022).

Example 1:

Retrieval-Augmented Generation (RAG) has established itself as the cornerstone for grounding Large Language Models (LLMs) in external, verifiable knowledge. (P4)

Example 2:

Recurrent neural networks, long short-term memory and gated recurrent neural networks in particular, have been firmly established as state of the art approaches in sequence modeling and transduction problems such as language modeling and machine translation. (P5)

Example 3:

Surface reconstruction of 3D scenes is a fundamental task in 3D vision, with a wide range of applications in downstream tasks such as AR/VR and embodied AI. (P11)

Example 4:

Semi-supervised video object segmentation (VOS) aims to segment and track the target object throughout a video sequence, given its mask in the first frame as a prompt. This task has received increasing attention in the research community because of its broad applicability in human–robot interaction, video editing, autonomous driving, and annotation assistance, etc. (P3)

Example 5:

Recently, researchers have shown considerable interest in pre-trained language models such as BERT for Natural Language Processing (NLP) tasks due to their promising performance. (P8)

Step 2: Making topic generalizations

After claiming centrality for the broader field, authors often narrow their focus to a specific topic or problem. In the introductions of the 20 papers, three main strategies for topic generalization emerged.

1) In a few cases, the author directly states the research topic (e.g., “agents” in Example 6), elaborates on its fundamental characteristics (“These agents should be...”), and subsequently raises an existing problem (“Tackling such a complex interactive task demands agents...”):

Example 6:

The advancement of artificial general intelligence is largely dependent on the development of agents that are proficient in complex interactive reasoning tasks. These agents should be capable of exhibiting problem-solving abilities akin to humans within dynamic, open-world environments. For example, the ScienceWorld benchmark features a task where an agent must determine the electrical conductivity of an unknown object. In a simulated environment, the agent must navigate to appropriate rooms, locate and acquire essential items, such as batteries and light bulbs, build a circuit, perform an experiment, and interpret the results. Tackling such a complex interactive task demands agents to exhibit long-horizon planning, long-term memorization, subgoal decomposition, spatial reasoning, exception handling, and commonsense knowledge capabilities. (P20)

2) The most common strategy for topic generalization involves the sequential introduction of two key technical terms: after identifying a broad field of centrality, the author typically shifts to a more specific topic, which often represents the solution or technical approach relevant to the broader field. The centrality of this specific topic may also be briefly explained. In the following example, the broad field “generative modeling and natural language processing” leads to the specific topic “image edition tools”, whose centrality is then noted. The problem with this specific topic is raised at the beginning of the second paragraph, followed by a literature review summarizing prior solutions.

Example 7:

Recent progress in generative modeling and natural language processing enables easy creation and manipulation of photo-realistic images, such as with DALL·E 2 or Stable Diffusion. They have given birth to many image editing tools like Control Net Instruct-Pix2Pix, and others, that are becoming mainstream creative tools for artists, designers, and the general public. (P6)

3) Another strategy involves progressive focusing of the research topic, typically requiring three successively narrower terms to establish the final focus, as illustrated by the progression Time-series data → time-series forecasting → deep learning approaches in the example below. This broad-to-narrow topic focusing process reflects the funnel-shaped argumentation structure characteristic of introductions.

Example 8:

Time-series data is ubiquitous in several domains such as retail, finance, manufacturing, healthcare,and natural sciences. In many of these domains, time-series forecasting, i.e., predicting time-series into the future, is a critical problem – for example, in applications like retail forecasting, climate and weather predictions, and traffic forecasting. In the last decade, deep learning approaches [SFGJ20; OCCB19; SYD19] have become popular in forecasting, often outperforming statistical approaches like ARIMA [BJ68]. (P12)

Step 3: Raising a problem within the field

Step 3 is operationally defined as any statement identifying unresolved limitations, research gaps, or persistent challenges within a given field. This includes methodological limitations (e.g., Example 10), data imperfections (e.g., Example 11), and conceptual gaps (e.g., Example 9). All coding tasks were completed solely by the researcher. The absence of inter-coder reliability testing is recognized as a limitation of this study. This problem is one recognized not only by the author but also by many other researchers in the field. Consequently, this step is typically followed by a literature review revealing solutions to the problem and then by the gap the author has identified, i.e., the problem the current paper aims to solve. Based on the location of the problem statement within the introduction, two main patterns emerge.

1) Most frequently, Step 3 appears at the end of the first paragraph, with the second paragraph containing the literature review discussing its solutions (Example 9). In Example 9, the first sentence introduces the core topic. The second sentence provides three pieces of evidence supporting the topic’s importance. Finally, the author shifts tone to raise the problem using phrases like “There is a conceptual gap” and “remains a challenge”.

Example 9:

Autonomous systems increasingly rely on complex learned models to supply predictions that are the basis for decision-making. Self-driving cars rely on deep neural networks to plan paths around nearby pedestrians, robotic manipulators leverage learned grasp models to plan high-throughput pick-and-place maneuvers in factories, and AI-enabled trading agents optimize the financial future of in-vestors. There is a conceptual gap between prediction and decision-making, and it remains a challenge to ensure that systems make good decisions despite imperfect predictions. (P6)

2) Occasionally, when the literature review is brief, the author may incorporate it within the same first paragraph (Example 10). The phrase “However, a limitation of such models” introduces the field problem, while “Previous work” briefly summarizes prior research responses. In such cases, the author typically presents their own research approach in the second paragraph.

Example 10:

...However, a limitation of such models for inference in the real world is the issue of grounding: while training LLMs on massive textual data may lead to representations that relate to our physical world, connecting those representations to real-world visual and physical sensor modalities is essential to solving a wider range of grounded real-world problems in computer vision and robotics. Previous work interfaces the output of LLMs with learned robotic policies and affordance functions to make decisions, ... (P10)

In this small corpus, two introductions exhibited a cycling or recurrence of Step 3 (Example 11). The first “However” points out the problem: point cloud data is often incomplete, necessitating restoration (Therefore, it is an essential task). The second “However” points out a flaw in existing restoration methods. These two “howevers” create a progressive argument, moving from “the problem exists” to “current solutions are inadequate”, thereby allowing the author to transition smoothly to their proposed technical solution in the second paragraph. This cycling of a move-step pattern (Crookes, 1986) demonstrates the author engaging in a “layer-by-layer unpacking” and “funnel focusing” of a complex issue, rather than a flat description, ultimately leading the reader towards the very specific technical challenge or research gap that the paper intends to address.

Example 11:

However, due to occlusion, light reflection, transparency of surface material, and limitations of sensor resolution and viewing angle, it will cause a loss of geometric and semantic information, resulting in incomplete point clouds. Therefore, it is an essential task to repair incomplete point clouds for further applications. Since the 3D point cloud is unstructured and unordered, the majority of the deep-learning-based methods for dealing with 3D data transform point cloud to collections of images (e.g., views) or regularly voxel-based representations of 3D data. However, multi-view and voxel-based representation leads to unnecessarily voluminous and limits the output resolution of voxels . (P13)

In summary, based on the small corpus established for this study, the background move of RA introductions in computer science synthesizes the findings of previous researchers. Generally, it comprises three steps: Claiming Centrality (80%), Making Topic Generalizations (60%), and/or Raising a Problem Within the Field (95%). For the first step, the majority of claims pertain to real-world problems. For topic generalization, three main strategies were identified: directly stating the research topic; using a second, more specific key term as the focused topic; or establishing the topic after a progression through three successively narrower terms. Regarding the final step of raising a problem within the field, depending on its location within the text, it most frequently appears at the end of the first paragraph; occasionally, when the literature review is brief, the author may place it in the middle of the first paragraph, prior to the literature review.

5 Conclusion

Based on the frameworks for the background move in RA introductions proposed by earlier researchers (Swales, 1990; Maswana et al., 2015), this study analyzes the background move of 20 papers in the discipline of computer science. It proposes a unique move-step framework for this discipline and identifies specific writing strategies for each step. Because the corpus contains merely 20 research articles, the findings possess limited generalizability. Therefore, the observations made here should be considered exploratory and preliminary rather than a precise characterization of the overall patterns in this discipline. Nevertheless, this stepwise structural framework assists early-career researchers, including undergraduate and postgraduate students, in completing the preliminary stage of manuscript writing: drafting the introductory background segment of research articles.

References

[1] Ahamad, M. I., & Amira, M. Y. (2012). A genre analysis of Islamic academia research article introductions. Procedia-Social and Behavioral Sciences, 66, 157-168.

[2] Ahmad, U. K. (1997). Scientific research articles in Malay [Doctoral dissertation]. The University of Michigan.

[3] Alanazi, M., & Alqarni, M. (2022). The very first sentence in research article introductions: A rhetoric comparative approach. Heliyon, 8(8), Article e10241.

[4] Crookes, G. (1986). Towards a validated analysis of scientific text structure. Applied Linguistics, 7, 57-70.

[5] Fredrickson, K. M., & Swales, J. M. (1994). Competition and discourse community: Introductions from Nysvenska Studier. In B. L. Gunnarsson, P. Linell, & B. Nordberg (Eds.), Text and talk in professional contexts (p. 9-22). ASLA.

[6] Lim, J. M. H. (2006). Method sections of management research articles: A pedagogically motivated qualitative study. English for Specific Purposes, 25(3), 282-309.

[7] Maswana, S., Kanamaru, T., & Tajino, A. (2015). Move analysis of research articles across five engineering fields: What they share and what they do not. Ampersand, 2, 1-11.

[8] Niu, G. L. (2018). A genre analysis of Chinese and international research article introductions based on comparable corpora. Journal of Xi’an International Studies University, 26(1), 25-30.

[9] Nwogu, K. N. (1997). The medical research paper: Structure and functions. English for Specific Purposes, 16, 119-138.

[10] Peacock, M. (2002). Communicative moves in the discussion section. System, 30(4), 479-497.

[11] Posteguillo, S. (1999). The schematic structure of computer science research articles. English for Specific Purposes, 18, 139-160.

[12] Samraj, B. (2002). Introductions in research articles: Variations across disciplines. English for Specific Purposes, 21(1), 1-17.

[13] Samraj, B. (2008). A discourse analysis of master’s theses across disciplines with a focus on introductions. Journal of English for Academic Purposes, 7, 55-67.

[14] Swales, J. M. (1990). Genre analysis: English in academic and research settings. Cambridge University Press.

[15] Swales, J. M. (2004). Research genres: Explorations and applications. Cambridge University Press.

[16] Swales, J. M. , & Najjar, H. (1987). The writing of research article introductions. Written Communication, 4(2), 175-191.

[17] Xu, Y. Z., Guo, L. H., & Xu, T. (2007). Genre teaching indispensable for academic writing: An investigation into the introduction writing of English major MA theses. Chinese Foreign Language, 4, 47-51+60.