Understanding the Basics in Structuring a Research Paper
In academia, research is the core component that influences new discoveries and shapes the content of that field for future use. Research can be of many different types and be structured in multiple forms based on the area that it deals with as well as the ideas presented in the paper. However, this article will be touching upon the specifics on how to prepare for and write a well-organized research paper.
Initially, any research paper needs to address the following 6 major sections.
- Abstract
- Introduction
- Methodology
- Evaluation
- Conclusion
- References
Having the above 6 sections as the backbone, the research can extend and branch out such that further details can be furnished as well. Initially, let us consider the abstract.
Abstract
The abstract can be pictured as a summarized version of the paper. The abstract needs to briefly deliver the overall contents of the paper to the reader. Hence, the main components that build up an abstract can be viewed upon as the:
- problem or the motivation for the research work.
- proposed solution.
- results and findings.
- interpretation and conclusion of findings.
Each component listed above should be addressed in 2 to 3 sentences since the typical word limit for an abstract is usually 250–300 words. In general, there is no need to specifically provide statistical details or numbers in the abstracts with occasional exceptions.
The abstract is also accompanied by a list of keywords. Keywords are terms or short phrases that indicate the main concepts discussed by the research paper. The conventional number of keywords that can be used to label a research work is 3–5.
The following is an excerpt from a sample abstract that represents the concept of the paper.
(Taken from the paper titled “Active design reviews: principles and practices” [1])
Although many new software design techniques have emerged in the past 15 years, there have been few changes to the procedures for reviewing the designs produced using these techniques. This paper describes an improved technique, based on the following ideas, for reviewing designs. The efforts of each reviewer should be focussed on those aspects of the design that suit his experience and expertise. The characteristics of the reviewers needed should be explicitly specified before reviewers are selected. Reviewers should be asked to make positive assertions about the design rather than simply allowed to point out defects. The designers pose questions to the reviewers, rather than vice versa. These questions are posed on a set of questionnaires that requires careful study of some aspect of the design. Interaction between designers and reviewers occurs in small meetings involving 2–4 people rather than meetings of large groups. Illustrations of these ideas drawn from the application of active design reviews to the Naval Research Laboratory’s Software Cost Reduction Project are included.
Introduction
A reader who finds the abstract to be interesting will proceed further to read the introduction of a paper. Since the ideas presented in the abstract are brief and succinct, the introduction needs to provide a comprehensive detailing of the entire research work. However, the ‘introduction’ does not necessarily need to indulge in all the technical details of the paper. Instead, a proper introduction will contain an extended abstract of the paper that introduces the identified issue that serves as the motivation for the paper, the proposed solution suggested by the researchers (in brief), the outcome of the solution implemented or tested (whether it was positive/effective, or negative/ineffective) and any other interesting conclusions, observations derived from the tests (no need to specify everything. Just a few would do), and finally briefly describe the structure of the paper such as how the following sections are structured and what will be discussed, specifying the contributions rendered by the research work.
Another interesting way to structure the introduction is to break it down into subsections as well. For an introduction that spans too long, structuring it in a way that elucidates each subsection in detail will improve the readability. Thus, the introduction can be branched into the following subsections.
a. Problem statement
The problem statement will outline the identified issue or problem that motivated the research work presented in the paper.
b. Related work
This section can be specified as an implicit subsection of the introduction or removed separately and placed as a major section soon after the introduction or before the conclusion if the related work section plays a major role in motivating the research flow and/or is extensive.
Adjoining the related work, the proposed solution can be contrasted against the previously existing methods that have been explained under the related work and explained in brief.
c. Contribution
With a mild overlook of the proposed solution presented, the contribution needs to specify the major and general use cases or contributions of the proposed solution in addressing the problem stated under the problem statement. If the work tends to address a single major contribution, it can directly be stated in a sentence or two. Moreover, if the proposed solution adds further value with more contributions, it can be listed as an ordered or unordered list based on the priority of each contribution rendered.
A sample introduction that elucidates the main contents that we discussed is shown below.
(Taken from the paper titled “Analyzing Mathematics Curriculum Materials in Sweden and Finland: Developing an Analytical Tool” [2])
Curriculum materials such as commercially produced textbooks and teacher’s guides have a strong presence in mathematics education in large parts of the world. These materials are typically a major resource for teachers’ planning and practice (Stein et al., 2007; Jablonka & Johansson 2010). One of the tasks of curriculum materials is to bring different discourses together (Jaworski, 2009). On the one hand, there is an academic conceptualization from which the intended curriculum derives and, on the other hand, the socio-cultural settings where teaching and learning occur: the enacted curriculum. Writers of curriculum materials may, therefore, interpret the intended curriculum and adjust to the socio-cultural settings to function as a bridge between the two different discourses. From this perspective, curricular materials serve as an important tool for teachers in both enabling and constraining their thoughts and actions (Stein et al., 2007). Further, curriculum materials are not only important resources for teachers in designing teaching (Stylianides, 2007), but also for teacher learning (Doerr & Chandler-Olcott, 2009). For instance, Remillard (2000) and Davis and Krajcik (2005) emphasize that curriculum materials could productively contribute to teachers’ professional development if they encompass elaborated attention to the process of enacting the curriculum. Therefore, potentially, well-designed curriculum materials could create opportunities for teacher learning.
There exists no role model for how to design such materials, since teachers’ use of, and learning from, curriculum materials are related to their experience, knowledge and the particular classroom situation. This study aims to contribute to the knowledge about teacher’s guides and their potential for various kinds of teacher learning in two neighboring countries with quite similar school systems but different teaching styles: Sweden and Finland. One rationale for undertaking a comparative approach is that through a process of investigating similarities and differences in various countries’ curricular materials we reveal some taken-for-granted and hidden aspects (cf., e.g., Andrews, 2009) of teachers’ work in classrooms. Such findings, we believe, could contribute to the international research discourse on aspects of curriculum materials and their influence on teaching and teacher learning.
Teachers in Finland use the textbook and teacher’s guides extensively (Joutsenlahti & Vainionpää, 2010). There are indications that many Finnish teachers are satisfied with the way their teacher’s guides are built, and that they consider them to be very helpful in differentiating the teaching (Heinonen, 2005). Further, Finnish teachers state that the guides provide help and ideas for new ways to teach and simultaneously ensure that the children learn what they are supposed to learn according to the state curriculum (L. Pehkonen, 2004). The Swedish teachers also use the textbook to a very large extent, but seldom use teacher’s guides (Jablonka & Johansson, 2010). Teachers in Finnish classrooms often lead whole-class instructions whereby all pupils are engaged in the same mathematical area (e.g., E. Pehkonen et al., 2007). In Sweden, on the contrary, it is common to conduct teaching as “speed individualization” and as personalized teaching (e.g., Jablonka & Johansson, 2010) where pupils work, in the same classroom, with different mathematical areas. Due to these differences, it is interesting to compare the curriculum materials used in the two countries. In this paper, we aim to answer to the following questions: What similarities and differences in curriculum materials exist within and between two countries, Sweden and Finland, with respect to their potential to contribute to various kinds of teacher learning? How is it possible to develop and amend an analytical tool for examining the potential for teacher learning in curriculum materials?
Another paper that structures the introduction using the subsections technique can be referred to in [3].
Methodology
The methodology officially redirects the entire tone of the paper into the technical mode. This is where you formally present the proposed solution, the prerequisites, clarifications on any special terminology used, architectures, flow models, pipelines, and so on. The detailed, general idea of how exactly the experiments were designed needs to go under this section. Some additional details that could furnish the methodology are shown below.
a. Solution architecture
The solution architecture needs to elucidate the flow of the proposed solution in a concise manner. The flow can either be explained from end-to-end or illustratively presented in the form of a flow model, but then again needs to be elaborated as well. The end-to-end model is basically comprised of various steps and hence, each step has to be further elucidated.
b. Equations
If the proposed solution deals with any mathematical equations or chemical formulas, they need to be stated and each component and specific representative symbols need to be mentioned.
c. Pre-computations
Though not all research solutions may involve pre-computations, the general idea is to state such pre-computed work or requirements, needed for the successful execution of the solution model, under this section as well. In other words, the pre-computations can be viewed as the pre-requisites that were implemented. Some of these pre-requisites may seem obvious and negligible, nevertheless, it is always advisable to state them all in order to replicate the proposed solution without any errors.
d. Use-cases
The use case specifies the applications or instances where the proposed solution can be implemented in solving real-life problems. Stating the use cases will support the firm grounding of your research work and add value to the proposed solution.
An excerpt of a sample methodology section is presented here below.
(Taken from the paper titled “Detecting Context Dependent Messages in a Conversational Environment” [4])
Suppose that we have a data set δ= {(mᵢ, yᵢ)} where mᵢ is a message composed of a sequence of words (wₘᵢ,1, …, wₘᵢ, nᵢ) and yᵢ is an indicator whose value reflects whether mᵢ is context dependent or not. Our goal is to learn a function g(.)ϵ {-1,1}using δ, thus for any new message m, g(.) predicts m a context-dependent message if g(m) = 1. To this end, we need to answer two questions: 1) how to construct δ; 2) how to perform learning using δ. For the first question, we can crawl conversation data from social media like Twitter and ask human labelers to annotate the messages in the data. The problem is that human annotation is expensive and time-consuming and therefore we cannot obtain a large scale data set for learning. To solve the problem, we automatically learn some weak supervision signals using responses of messages in social conversation data and take the signals as {yᵢ} in δ. For the second question, one straightforward way is first extracting shallow features such as bag-of-words and syntax from messages and then employing off-the-shelf machine learning tools to learn a model. The problem is that shallow features are not effective enough on representing semantics in short conversation messages, which will be seen in our experiments. We propose using a Long Short Term Memory (LSTM) architecture to learn a model from δ. The advantage of our approach is that it can avoid explicit feature extraction and large scale human annotations, and carry out feature learning and model learning in a unified framework.
Evaluation
Nearing the end, this is the section where the experiment details, test results, and discussion goes. The solution model proposed under methodology will remain a conceptual model until it is implemented and tested under the evaluation. Rephrasing this statement, the evaluation is what proves the validity of the idea proposed by the research paper. All the statistical details, datasets, test and control conditions, sample and population details, distributions, analysis, and results will be organized under this section.
In rare cases, the proposed idea may not be a solution that can be evaluated by the implementation or presentation of results. Such papers can present their solution as concept models and discuss the plausibility of the solution, the constraints in implementing it, possible outcomes, and so on.
The following is an excerpt from a statistical evaluation conducted for a research study.
(Taken from the paper titled “Bringing Chatbots into Education: Towards Natural Language Negotiation of Open Learner Models” [5])
The participants were 30 students from the University of Birmingham Electronic, Electrical and Computer Engineering Department. 11 were final year undergraduates and 19 were MSc students. All had previously taken courses in educational technology and C programming. All were competent English language speakers, though in some cases English was not their first language. Mr. Collins [28] provided a range of strategies for negotiation: ask user if they wish to accept the system’s viewpoint; offer compromise; ask user to justify their belief (e.g. by taking a test to demonstrate knowledge); system justify its belief; or offer 186 student the opportunity to view the learner model. These strategies were adopted as the initial conversational basis of the ‘chatbot’. The ‘Wizard’ was provided with a decision tree to allow the consistent selection of appropriate responses, and 350 pre-authored ‘chatbot’ negotiation initiations and responses to user inputs [37]. These ‘canned responses’ can be seen in the left part of Figure 2, while the right shows the wizard’s view of the learner model. This view of the model enabled the wizard: (i) to compare the user and system beliefs for each topic in two columns using colour to represent the student’s and system’s beliefs about the learner’s knowledge level; (ii) to see the student’s answers to questions; and (iii) to select unanswered questions to offer the student a test by which to demonstrate their knowledge.
Conclusion
Winding up the entire work presented in the research, the conclusion needs to re-iterate the motivation, the solution presented in the research, the outcome, observations, and the findings based on the evaluation. The conclusion also needs to clearly mention how the proposed solution has managed to address the contribution stated in the beginning. Once again, the conclusion need not be elaborate. However, it has to summarize the paper in a manner similar to the abstract, but with more details related to the outcome and how the contribution has been successfully met by the solution.
The following is a comprehensive conclusion fetched from research conducted with regard to the blockchain.
(Taken from the paper titled “Survey of Consensus Protocols on Blockchain Applications” [6])
Blockchain technology is expected to revolutionize the finance and banking sectors around the world. Many banks have already started building their own blockchain application or are finding ways to initiate one. In this paper, a comparative analysis of consensus protocol on SCP, Corda, and hyper ledger is made. The SCP has already gained popularity for its ability to connect people of various trends in society. SCP proposes to provide asymptotic security and flexible trust by introducing the concept of quorum slices that ensures more freedom to the users on deciding the participants they need to trust for validating their transactions. It also manages the stuck situation that could occur during consensus by neutralizing them. But, as mentioned in SCP [3], when providing the users more freedom in choosing the nodes to trust, the system may not be able to provide the user with all its features if the user is new to the system and does not choose the nodes to trust efficiently. Corda maintains records of various business and financial contracts. These may not be open to all, but in financial institutions where data is confidential, this applicability ensures privacy. Since the contracts are automated and they follow the then legal format for all its documents, they act as a real time saver. Also, its scalability is huge, allowing institutions from various sectors to implement and maintain their financial records using Corda. Hyperledger project allows various blockchain technologies to interconnect and assures a secure plug and play environment for them. The hyper ledger does not provide the users as much freedom as the SCP does. The current programming languages supported as chain code in hyper ledger include mainly Java, Go. Hyperledger thus focuses more on interoperability of various proposed blockchain applications, but support for various other programming languages like Haskell, Perl, etc. are proposed but yet to be implemented. The applicability of blockchain technology is not confined to the bank or finance sector. New innovations have already come out of the blockchain technology. Blockchain technology is now used for various IoT applications too to secure them against intruders. Consensus protocol being the working entity of blockchain can thus be of varied implementation styles. An efficient protocol could thus produce tremendous results for the growth of the economy.
References
The reference is a list of other related work and resources such as research papers, journal articles, books, image sources, code implementations, websites, etc. There are various types of referencing formats such as MLA, APA, Chicago, Harvard, BibTex, and so on.
The main fact to remember with referencing a source is that the reference needs to be added next to the sentence or text where it is being referred at, in the format of [i] or i, where i is a number starting from 1. And then, the respective reference and can be mentioned under the reference list.
A sample text where the references have been cited and the respective reference list is shown in the following sample.
(Taken from the paper titled “Knowledge Graph Identification” [7])
Text from the Related work section, where a few citations are available:
Early work on the problem of jointly identifying a best latent KB from a collection of noisy facts was considered by Cohen et al. [8], however, they considered only a small subset of KB errors. More recently, Jiang et al. [9] perform knowledge base refinement at a broader scope by using an ontology to relate candidate extractions and exploring many different modeling choices with Markov Logic 544 J. Pujara et al. Networks (MLNs) [10]. Jiang et al. provide a crisp codification of ontological constraints and candidate facts found in a knowledge base as rules in first-order logic, contributing an attractive abstraction for knowledge bases that we adopt in our modeling. However, the choice of MLNs as a modeling framework comes with certain limitations. In MLNs, all logical predicates must take Boolean truth values, making it difficult to incorporate the confidence values. Moreover, the combinatorial explosion of Boolean assignments to random variables makes an inference and learning in MLNs intractable optimization problems. Jiang et al. surmount these obstacles with a number of approximations and demonstrate the utility of joint reasoning in comparison to a baseline that considers each fact independently. By using PSL we can avoid these representational and scalability limitations, and we build on and improve the model of Jiang et al. by including multiple extractors in our model and reasoning about co-referent entities.
Other research has used relevant techniques for problems related to knowledge graph identification. Namata et al. [11] introduced the problem of graph identification to uncover the true graph from noisy observations through entity resolution, collective classification, and link prediction. However, Namata’s approach considered these tasks iteratively and could not easily support logical constraints such as those found in an ontology. Memory et al. [12] also use PSL to resolve confounding evidence. Their model performs graph summarization across multiple ontologies and uses inference only for inferring missing links. Work by Yao et al. [13] employs joint reasoning at the extractor level by using conditional random fields to learn selectional preferences for relations.
The respective references added under the Reference list:
1. Ji, H., Grishman, R., Dang, H.: Overview of the Knowledge Base Population Track. In: Text Analysis Conference (2011)
2. Artiles, J., Mayfield, J.: Workshop on Knowledge Base Population. In: Artiles, J., Mayfield, J. (eds.) Text Analysis Conference (2012)
3. Carlson, A., Betteridge, J., Kisiel, B., Settles, B., Hruschka, E.R., Mitchell, T.M.: Toward an Architecture for Never-Ending Language Learning. In: AAAI (2010)
4. Etzioni, O., Banko, M., Soderland, S., Weld, D.S.: Open Information Extraction from the Web. Communications of the ACM 51(12) (2008)
5. Pasca, M., Lin, D., Bigham, J., Lifchits, A., Jain, A.: Organizing and Searching the World Wide Web of Facts-Step One: the One-million Fact Extraction Challenge. In: AAAI (2006)
6. Singhal, A.: Introducing the Knowledge Graph: Things, Not Strings, Official Blog, of Google (2012), http://goo.gl/zivFV
7. Broecheler, M., Mihalkova, L., Getoor, L.: Probabilistic Similarity Logic. In: UAI (2010)
8. Cohen, W., McAllester, D., Kautz, H.: Hardening Soft Information Sources. In: KDD (2000)
9. Jiang, S., Lowd, D., Dou, D.: Learning to Refine an Automatically Extracted Knowledge Base Using Markov Logic. In: ICDM (2012)
10. Richardson, M., Domingos, P.: Markov Logic Networks. Machine Learning 62(1–2) (2006)
11. Namata, G.M., Kok, S., Getoor, L.: Collective Graph Identification. In: KDD (2011)
12. Memory, A., Kimmig, A., Bach, S.H., Raschid, L., Getoor, L.: Graph Summarization in Annotated Data Using Probabilistic Soft Logic. In: Workshop on Uncertainty Reasoning for the Semantic Web (URSW) (2012)
13. Yao, L., Riedel, S., McCallum, A.: Collective Cross-Document Relation Extraction Without Labelled Data. In: EMNLP (2010)
14. Kimmig, A., Bach, S.H., Broecheler, M., Huang, B., Getoor, L.: A Short Introduction to Probabilistic Soft Logic. In: NIPS Workshop on Probabilistic Programming (2012)
15. Bach, S.H., Broecheler, M., Getoor, L., O’Leary, D.P.: Scaling MPE Inference for Constrained Continuous Markov Random Fields with Consensus Optimization. In: NIPS (2012)
16. Dixon, S., Jacobson, K.: LinkedBrainz — A project to provide MusicBrainz NGS as Linked Data, http://linkedbrainz.c4dmpresents.org/
17. Raimond, Y., Abdallah, S., Sandler, M.: The Music Ontology. In: International Conference on Music Information Retrieval (2007)
18. Davis, I., Newman, R., Darcus, B.: Expression of Core FRBR Concepts in RDF (2005), http://vocab.org/frbr/core.html
19. Brickley, D., Miller, L.: FOAF Vocabulary Specification 0.98 (2010), http://xmlns.com/foaf/spec/20100809.html
20. Kobilarov, G., Scott, T., Raimond, Y., Oliver, S., Sizemore, C., Smethurst, M., Bizer, C., Lee, R.: Media Meets Semantic Web–How The BBC uses DBpedia and Linked Data to Make Connections. In: Aroyo, L., Traverso, P., Ciravegna, F., Cimiano, P., Heath, T., Hyv¨onen, E., Mizoguchi, R., Oren, E., Sabou, M., Simperl, E. (eds.) ESWC 2009. LNCS, vol. 5554, pp. 723–737. Springer, Heidelberg (2009)
21. Bizer, C., Seaborne, A.: D2RQ–Treating Non-RDF Databases as Virtual RDF Graphs. In: ISWC (2004) 22. Suchanek, F.M., Kasneci, G., Weikum, G.: Yago: A Core of Semantic Knowledge. In: WWW (2007)
References
[1] Parnas, D. L., & Weiss, D. M. (1985, August). Active design reviews: principles and practices. In Proceedings of the 8th international conference on Software engineering (pp. 132–136). IEEE Computer Society Press.
[2]Hemmi, K., Koljonen, T., Hoelgaard, L., Ahl, L., & Ryve, A. (2013). Analyzing mathematics curriculum materials in Sweden and Finland: Developing an analytical tool. In Proceedings of the Eighth Congress of the European Society for Research in Mathematics Education. Antalya, Turkey. Feb 6th-Feb 10th.
[3] Jeyaraj. W. J. (2012). Designing Libraries based on Factors that Determine the Existence of Libraries. Indian Journal of Information Sources and Services. 2(1). 71–74.
[4] Li, C., Wu, Y., Wu, W., Xing, C., Li, Z., & Zhou, M. (2016). Detecting context dependent messages in a conversational environment. arXiv preprint arXiv:1611.00483.
[5] Kerlyl, A., Hall, P., & Bull, S. (2006, December). Bringing chatbots into education: Towards natural language negotiation of open learner models. In International Conference on Innovative Techniques and Applications of Artificial Intelligence (pp. 179–192). Springer, London.
[6] Sankar, L. S., Sindhu, M., & Sethumadhavan, M. (2017, January). Survey of consensus protocols on blockchain applications. In 2017 4th International Conference on Advanced Computing and Communication Systems (ICACCS) (pp. 1–5). IEEE.
[7] Pujara, J., Miao, H., Getoor, L., & Cohen, W. (2013, October). Knowledge graph identification. In International Semantic Web Conference (pp. 542–557). Springer, Berlin, Heidelberg.
