Development of a Decision Support System for Helpdesk-Service Based on the Processor for Natural Language Text Classification

Text is one of the most common sources of information. Due to this, there is an urgent need for automated processing of text data. In vast majority of cases, classifiers are applied in text data processing. The object of this work is to develop a text classification processor for a decision support system in the Helpdesk-service-like applications. The modern approach to the text classification based on machine learning method with the engagement of artificial neural network is widely known. The article describes the most common methods for text processing and neural networks architecture. Considered are such text processing methods as: stemming, lemmatization, bag-of-words, N-gram count, TF-IDF. The following neural network architectures: multilayer perceptron, convolutional neural network and long short-term memory network. To find the most optimal algorithm for building up a processor there were considered 36 processors with the use of various combinations of the above methods and neural networks architectures. The article has an analysis and a comparison of a work quality of various classification processors conducted on such characteristic as accuracy. It was established that the processors learned on the data processed though stemming and N-gram count show the highest accuracy. The processor, at the construction of which were used stemming, N-gram count and a convolutional neural network, showed a maximum accuracy of multiclass classification. The accomplished work found an optimum choice of a classifier to be used in form of a module in decision support systems in the Helpdesk-service-like applications.