Research means finding something new. Great academicians say it is easy to pursue research. A novel invention is carried out based on the existing information about the product. Using bits and bytes to compile a new technique is a research. To solve something for welfare of society and to discover something innovative is termed as research.
Literature survey is the most important step in academic research. It may bring out many useful insights, using which one can find the logic to propose an effective approach for a problem statement.
In any stream, one should finalize the subject or the problem domain. One should read related research papers. The research papers which one should read during literature survey should be of good journals and conferences. Once you find out an interesting research topic for your PhD, you should focus on identifying a base paper and retrack it in backward direction from references as indicated in our previous article.
What should I identify in Literature Survey?
One should record the following observations from the set of research papers.
An algorithm is the heart of any research work. It determines how precise results are and what is the logic behind using this approach. The algorithm is proposed to fill a research gap. There can be more than one algorithms which may be used for solving a problem. For example, for solving clustering based problem, k-means, DBSCAN, k-NN and many more clustering algorithms can be used. You should identify the type of problem and why this algorithm is used for the problem. Open your mind and think beyond this algorithm. Try to find alternative problems and data specific logics to use those alternative algorithms. This is usually followed by those researchers who are beginners.
The research gap is determined by using research challenges. Research challenges could be anything which needs to be studied with respect to the problem domain. For instance, the problem of using automatic learning machine for predicting results and so on. There could be some areas in which many algorithms are not applied and tested. You may swap or integrate problem domains for similar areas. This is usually an approach carried out by experts.
The environment of research work is very important factor. Depending upon your research field, for instance, in speech recognition, the pitch and frequency of sound, the environmental noise removal, separating two different sounds which were mixed in one recording are some of the major challenging areas. Similarly, in research lab, experimental analysis with chemicals and other mechanical machines depends upon the environment in which it has been carried. You may add important parameters or remove unnecessary environment dependency using logical inferences.
The datasets used for experimental evaluation is usually old and may lack some features. It is very common to find better features for experimental analysis. This feature selection and feature extraction may add novelty to your work. This is used by those researchers who are using machine learning and deep learning in their work. Use of multiple datasets, data from multiple domains and in case of natural language processing, language independent datasets (corpus from multiple languages) can be helpful in identifying language insensitive approach for solution to problem domain. Using publicly available dataset is always and added feature of your research work which proves unbiased results in your work. However, there could be some problem domains in which dataset might not be present. You may add novelty by introducing a new dataset and then you may implement standard research algorithms on it.
Many researchers use standard performance measures for instance precision, recall and f-measure. However, there could be some problem domains which are evaluated using manual evaluation only. You may make good attempts to identify and propose new performance measures for such problem domains. For instance, to identify if text summarization is useful or not, manual evaluation is performed or ROUGE score is used. You may attempt to propose new performance measures.
You may use alternative algorithms to improve results for accuracy, precision, recall and other performance measures. This approach is usually followed by inferences, imagination and hit and trial methods. One can also work on future Scope of the existing research work as indicated by the author.