Answering library consultation questions with a machine learning approach: A study on academic databases

The fact that there is no single machine learning algorithm useful for solving all problems worldwide and that many scientific fields feed various machine learning algorithms and prefer these algorithms in their field-specific applications has made it a matter of curiosity what the useful algorithms are in machine learning projects carried out in the discipline of Library and Information Science. In this regard, the study aims to determine the model and appropriate algorithms developed to ensure that natural language questions describing the information source requirement on any subject are answered with the machine learning approach in the knowledge discovery process. The training data set, consisting of features obtained from academic databases, was converted into a format that the machine can learn by going through various data mining processes. Then, a test data set consisting of 7300 questions was prepared by imitating natural language questions that describe the need for information sources on a subject that are likely to be directed to library information units. It was requested to test the model with the test data set filtered using text mining and natural language processing techniques. Support Vector Machine, Probabilistic Neural Network, K-Nearest Neighbor, Naive Bayes, Fuzzy Logic, Decision Tree and Deep Learning (DL4J) algorithms suitable for data matrices were tested with both default and improved hyperparameter settings. Support Vector Machine, which was found to be the most useful algorithm, achieved a success rate of 92.7%. Alternative algorithms to this are determined as Deep Learning (72.3%) and Probabilistic Neural Network (70.5%).