Developing a text mining and artificial intelligence based method for identifying and analyzing job satisfaction factors using employees' online evaluations

This thesis study provides information about the work conducted to obtain machine learning-based analysis of employees' online reviews aimed at identifying and analyzing job satisfaction factors. A total of 3370 online reviews from 59 different companies published on tr.indeed.com were obtained, and analyses based on machine learning methods were performed. Furthermore, a larger dataset was obtained from www.kaggle.com to demonstrate that machine learning techniques can achieve high performance in classification and work effectively and scalably with large datasets. A total of 67529 comments consisting of online reviews made by employees of Google, Amazon, Netflix, Facebook, Apple, and Microsoft about their companies were analyzed. Three different weighting schemes were used to determine the features in the text document, taking into account term presence, term frequency, and term frequency-inverse document frequency as basic representation methods, along with 1-gram, 2-gram, and 3-gram models. Five different supervised learning algorithms were used for machine learning-based analyses: NB, SVM, LR, KNN, and RF. In addition to the specified machine learning methods, community learning methods such as DWEOL, CBWE, BMAE, and BGWEDU were used and compared with deep learning-based models such as CNN, RNN, LSTM, and GRU models. The word embedding-based models used in these analyses are word2vec (skip-gram model), word2vec (CBOW) model, fastText (skip-gram) model, fastText (CBOW) model and GloVe models. Different representation methods for data sets analyzed using different models were comparatively evaluated using accuracy, precision, recall and F-measure criteria.