To stay up-to-date in your field of research or study, you should read research papers. However, reading a research paper is not like reading a newspaper. Because a paper has a formal structure that consists of several sections. Authors of research papers know the structure quite well. It is also essential for readers to be familiar with the main sections of a research paper. This helps readers to quickly find the information they are looking for in a research paper. Moreover, it helps them comprehend new ideas and methods of a paper better. Aside from gaining new knowledge, reading research papers help you find out what was done in the past to solve a particular problem. Therefore, you are not going to reinvent the wheel and probably implement a method or algorithm that is proposed in a paper.
In this post, I explain the components of each section in a research article. Also, examples from a real and open-access research paper in machine learning are provided to help you understand the components of each section.
I have chosen the paper “Speech Emotion Recognition Based on SVM and ANN” from the journal of machine learning and computing. Because this paper is open-access, short and relatively easy-to-understand, which is suitable for the purpose of this post. Furthermore, it has an interesting application (Speech emotion recognition) and used well-known classifiers such as SVM and ANN. The PDF file of the chosen paper can be downloaded from here.
Disclaimer: I don’t know the authors of the chosen paper and do not advocate their research work. The paper is solely chosen for education purpose.
A research paper has often five sections which are as follows:
- Discussion and conclusion
Some papers may have an additional section. Sometimes authors may want to explain a component in more detail. For example, previous research studies are often introduced in the introduction. However, some might want to create a dedicated section for related research papers to their work.
Next, I’m going to explain each section of a research paper and its components with examples. You may feel this is boring but bear with me! Later on, you’ll appreciate this information that helps you read a research paper more effectively. Perhaps, you may want to write your own research paper in the near future.
Two important notes before proceeding further:
- Bear in mind that this is a relatively long post. Therefore, I highly recommend you to take a break after reading a section. For example, you can read this post in a few days. It helps you grasp the main ideas better.
- While reading the post, I also recommend you to grab a copy of the selected paper or open it up in a PDF viewer (you can download it from here.). So you can follow examples that given for each component. Moreover, the whole point of this post is to encourage you to read a research paper and help you get most out of a paper. If you don’t get the paper, it will be like a situation where someone teaches me how to drive but we don’t have a real car for practice and training.
An abstract is a brief preview and description of a research paper. It provides key points, the main idea, and the results of a paper. By reading the abstract of a research paper, you can decide whether or not to read the whole paper. As the author of a research paper, the abstract is somehow very shot advertisement for your paper. Based on this, readers mainly decide to read your paper or not.
An abstract has usually four components which are explained with examples from the selected paper.
- Background/Problem: At the start of an abstract, some background information about the research problem is provided to help readers understand the problem itself and its importance. An example from the chosen paper:
Speech emotion recognition mainly includes emotion feature extraction, feature reduction and speech emotion recognition model.
The writers provided brief background information about the research problem which is “Speech emotion recognition”.
- Contribution/Methodology: After providing the research problem, the novel method of a research paper is mentioned which solves the research problem. Note that the authors were contributed to their field of research by proposing a new method or approach. Here is an example from the selected article:
This paper chooses valid emotional features and extracts the statistical values of the emotional features. Speech emotion recognition model are constructed respectively based on SVM and ANN and the recognition effect of feature reduction respectively on two types of models are compared.
Authors are given the overview of their method for speech emotional recognition problem. Their method consists of emotional features as well as SVM and ANN classifiers.
- Results/Implications: The key results of experiments that performed by the authors are mentioned in the abstract. Sometimes numerical details of the experiments are given. Moreover, the implications of the result contribute to the information in the research area of the paper. Here is an example from the chosen paper:
The experimental results show that, based on emotion features which is extracted by CASIA emotion corpus, feature reduction can improve recognition accuracy and the recognition effect of speech recognition model based on SVM is better than ANN.
Authors’ method improved the prediction accuracy for the problem of speech emotional recognition. Also, based on the results, they found that SVM classifier performs better than ANN for the research problem.
In my opinion, this section is probably more important than other sections. The value of a research paper is often judged based on the introduction. Let me explain it. First of all, what problem a research paper solves is clearly stated in the introduction. Second, the authors give an overview of the research aim and methods. By reading the introduction of a paper, you can be almost sure whether the paper is relevant to your research area or not. Each component of the introduction section is described below with examples from the chosen paper.
- Background facts/Importance of the research problem: Oftentimes an introduction starts with why the research problem is important. From the perspective of a reviewer of a research paper, the problem may not worth solving. Moreover, some definitions are provided for those readers who may not be familiar with technical terms in a field of research. However, keep in mind that when you are reading a paper in a field of research (Say machine learning), it is assumed that you know the principles of that field. Anyhow, the example from the selected paper is shown below:
Speech emotion recognition research involves the traditional speech signal processing, pattern recognition, human psychology, artificial intelligence, human-computer interaction and other fields. In human-computer interaction, speech emotion recognition is an important part to determining the emotional state of interaction objects
Authors clearly stated the importance of speech emotion recognition (research problem) by describing its application in human-computer interaction.
- Previous research or studies: After the research problem is defined, authors introduce previous work that addressed this problem. Most research papers address a problem that has been studied before. This component in a research paper is essential. Because you can’t solve a problem without doing research on what was done regarding the problem before. Moreover, you don’t want to reinvent the wheel! Here is an example from the selected paper:
The study of speech emotion recognition can be traced
back to the early 1980s. …Schuller ,  at Technical
University of Munich were conducted a research on speech
emotion recognition. A lot of studies on speech emotion
reconducted by the Voice and Emotional Group at the
University of Southern California and the Emotion Research
Laboratory at Université de Genève.
The authors of the selected papers provide an overview of previous research on speech emotion recognition.
- A gap/problem in the research: After reading carefully previous research papers on your problem, you often find a problem or a gap in previous work (Remember no scientific work is perfect! There is almost always a problem that researchers haven’t considered. ). So the purpose of the paper is to introduce a new method to solve the problem you found. In the case of the selected paper, authors didn’t mention explicitly the gap in research. However, they state that their paper is based on previous work.
- Contribution: When the research problem is defined in the introduction, the general description of the method for solving the problem is given. At this point, authors state what the method does or what is the main purpose of the research work without giving much detail. An example from the chosen paper is given below:
Based on the CASIA Chinese Emotional Corpus, this paper
analyzied key technologies in the procession of speech
emotion recognition and the effect of feature reduction on the
speech emotion recognition. We compared the emotional
classification effect of speech recognition model based on
SVM and ANN.
By reading the above example from the selected paper, you probably realize that the paper has two research aims: (1) It investigates the effect of feature reduction on the speech emotion recognition problem. (2) The second aim was to compare the performance SVM and ANN classifiers for the research problem.
Before explaining the Methodology section, I should point out that the selected paper has an additional section to those main sections that I have listed above. At times authors provide an additional section mainly for two reasons. First, to provide sufficient background information about the research problem for readers who may not be familiar with the research area. Second, to analyze previous work or studies more for grouping related ideas and methods. Hence advantages and drawbacks of previous methods are stated. The selected paper has a section for providing background information on speech emotion recognition. Here is an example from this section:
The premise of speech emotion recognition is the
description of emotion. At present, there are two kinds of
emotion descriptions: discrete emotion classification system
and dimension emotion classification system
Here authors provide a little bit of information about emotion descriptions and also the process of speech emotion recognition. Personally, I think that it is a good practice to provide background information about the research problem. Sometimes even experienced researchers want to see definitions of technical terms in your paper, although they are familiar with the research problem.
As you saw in the introduction, the authors define a research problem which is important to solve. To solve a research problem which in our case was “Speech emotion recognition”, the authors proposed a new method to solve it. By reading this section, researchers expect to learn something new which they didn’t know before (The main point of reading a research paper is to learn about new advances and trends in a field of research). In the methodology section, a detailed description of the proposed method is provided. In fact, it’s essential to provide sufficient detail for readers to replicate the research work and obtain the nearly same result.
The selected paper has two sections for the methodology. You may recall that the paper has two research aims. Hence a section was created for explaining each aim. Let’s take look at each component of the methodology section:
- Overview of the method: The methodology section usually starts the general overview of the proposed method. That is writers give readers a big picture of their method rather than the details of the method. This is usually done by referring back to previous sections or outlining the parameters of the research work. Here is an example from the chosen paper:
The recognition model is a key part of speech emotion recognition. CASIA Chinese Emotional Corpus is one of discrete speech emotional corpus. For discrete emotional recognition, emotional recognition can be modeled into a pattern classification problem. In this paper, the classifiers are constructed respectively based on SVM and ANN
Authors provide a general description of the section which is about the recognition models (ANN & SVM). Also, they introduced CASIA dataset for speech emotional recognition.
- Background information: As in introduction, authors also provide background information in the methodology section. you may ask why! Because you don’t want to re-invent the wheel. To conduct research, most authors use things that belong to somebody else such as dataset, software, algorithm, formula, etc. Therefore, the authors acknowledge the source of things that they have used. Also, background information is provided about those things so that readers can better understand what the authors did. Here is an example from the selected paper:
There are two ways to reduce the dimension of features: feature selection and feature extraction. Feature selection does not change the original eigenvalue but select a valid subset of features through removing the irrelevant or redundant features from the original set
Here the authors introduce two ways of dimensionality reduction for a pattern recognition problem.
- Specific and precise details about the proposed method: Authors provide detailed information about what they did and used. In Computer Science, if the proposed method is a new algorithm, a step-by-step procedure is often stated to solve the problem. Given detailed information about the proposed method, you should be able to replicate the research work and get similar results. Here is an example:
The activation function of hidden layers neuron is the
linear correction unit ReLU function. When a large gradient
flows through the ReLU neurons, the weights become smaller
due to the large gradient and gradually to 0, so that the
neurons have the activating effect no longer. We avoid the
problem as much as possible by setting the learning rate to
The authors of the selected paper give detailed information about the properties of ANN classifier that they used for the speech emotional recognition. For instance, they used ReLU activation function. Without this information, it’s almost impossible to obtain similar results using ANN classifier.
- Justifications for choices and decisions: Authors give reasons for why they did things in a particular way. Because professional readers (like reviewers) are usually skeptical about the research work. Hence authors try to answer critical questions that readers might have (e.g why using method X?). Here is an example from the chosen paper:
PCA simplifies the original sample data, which can eliminate the correlation between the sample data and remove the interference of noise and excess features. The application of PCA do not have parameter restrictions, which make PCA to have a very wide range of applications. This
paper chooses PCA to reduce the dimension of features.
By mentioning the advantages of PCA method, authors somehow justify why PCA was used for reducing dimensions of training data.
After introducing the proposed method in detail, authors conduct extensive experiments to show the validity of their method. Basically, they report what they found or observed. Therefore, the results sections contain tables, figures, and images to compare the proposed method with other related methods. Moreover, Authors interpret the results based on their own understanding. With results’ interpretations, they try to convince readers that their proposed method is effective and valid. The results section oftentimes has the following components:
- The general overview of the results: At the start of the results section, authors often provide general statements about what they found. This gives the reader the overall trend of the results section. Also, the authors begin the results section by referring back to something from the previous section such as research aims or research problem. However, authors of the selected paper provided a general description of the dataset that they used for speech emotion recognition. Here are their statements:
The CASIA Chinese Emotional Corpus, recoded by institute of automation Chinese academy of sciences, consists of 9,600 audios including 6 kinds of emotion (neutral, angry, fear, happy, sad and surprise). The corpus is made up of 300 identical text and 100 different texts.
- Invitation to view the results: Perhaps, the main point of the results section is to present what the authors observed or found. Therefore, they invite you (the reader) to view the results by looking at the figures/graphs/tables. Here is an example from the selected paper:
The results are shown in the Table III.
- Comparisons with the results in the other research: There is a very high chance that other researchers have approached the research problem with their own method. Therefore, the authors compare the results they have obtained with other similar studies. Otherwise, they have no clue how satisfying the results are. In the field of Machine Learning, it is essential to compare your recognition rate with that of other research papers on the same dataset. By doing the comparisons, the authors show that their proposed method performs better than other studies’ method. However, the authors of the selected paper did not compare the results with other research. To remind you of the selected paper’s research aims, the authors investigated the effect of feature reduction on the speech emotion recognition. Nonetheless, they performed various experiments to show the readers the effect of feature reduction method on the accuracy rate for the speech emotion recognition task.
Discussion and Conclusion
The conclusion section is usually quite short in most papers and consists of two paragraphs. The elements of the introduction section occur again in the conclusion. That is, the authors state what the research problem is. However, they summarize key research findings based on the results they have obtained. This means that to what extent authors solved the research problem. Moreover, the conclusion doesn’t have new information. Ideas and key findings from the previous sections are revisited here. Let me look at each component of the conclusion section below:
- Revisiting the previous sections: At the start of the conclusion section, authors most often refer back to something from the previous sections. This is often done by restating the research aims or the research problem the paper was designed to solve. Here is an example from the conclusion section of the selected paper:
Based on SVM and ANN respectively, two speech emotion classification models are constructed. In the two classification models, the effect of feature dimension reduction on accuracy is analyzed, and the two classification methods are compared.
Here, authors restated the two research aims of the paper in the conclusion section. First, they analyzed the effect of the dimension reduction on accuracy. Second, ANN and SVM classifiers are compared for speech emotion recognition problem.
- Achievements/Contributions: These two are a very important aspect of a research paper. Because achievements and contributions are what makes a research paper different and unique from other research. Therefore, the key findings of the result section, the main ideas of the research paper, and the main advantages of the proposed method are summarized in the conclusion section. Here are the achievements of the selected paper:
The experimental results show that the feature dimension reduction is helpful to improve the classification performance. In this paper, the performance of SVM in speech emotion recognition is slightly better than ANN.
The authors state that the paper has two key findings based on the experimental results. First, they found that a feature reduction method like PCA may improve classification accuracy. Second, SVM performs slightly better than ANN for speech emotion recognition problem.
- Future work: A research paper almost always ends with some suggestions for future research. Obviously, no research work is perfect. There are always some limitations in research work. Therefore, the authors point out directions for future research that can potentially solve the problems of previous work. Also, if you are looking for some research topic for your future research work, you can read the last paragraph of the conclusion section. Here is an example from the selected paper:
In the future research, we need to expand the corpus to carry out research and analyze the distinction of speech emotion features and the effect of different feature reduction methods on speech emotion recognition to improve the accuracy of speech emotion recognition.
Here, the authors of the selected paper suggest that future research work can investigate the effect of a different feature reduction method on speech emotion recognition. They used the PCA method. Actually, using another feature reduction method may improve the recognition rate for the task of speech emotion recognition. Research in machine learning is about conducting various experiments. On the paper we can write method X may improve the accuracy of task A. This is a hypothesis. It means you need to conduct an experiment to observe whether the hypothesis is true or false.
In this post, I’ve tried to explain the main sections of a research paper and the components of each section. Also, I used an open access research paper to give you some examples for each component from real-world research. The main point of writing this post was to help those who are new to reading/writing a research paper. After reading this post completely, you are now familiar with the formal structure of a research paper, which means you are able to find quickly the information you need for your particular task or project in a research paper. Moreover, I highly suggest to visit Google Scholar and find several research papers in your field of interest. Reading research papers is one of the great ways to keep yourself up-to-date and learn new advancements, methods, and algorithms.
After all, I’d like to know your questions and thoughts. Let me know them by leaving a comment below.