Evaluation of employee suggestions by using topic modeling approach in an automotive company

Text mining, a sub-branch of data mining, is used for feature extraction in texts written in many languages and in many different fields. Subject modeling methods, which have gained importance in recent years, have started to be preferred frequently in text mining applications. Topic modeling is one of the most powerful techniques in text mining for data mining, discovery of hidden data, and finding relationships between data and text documents. Topic modeling, which reveals the hidden structure of large documents in an unsupervised manner, has been accepted as a successful method. Phrases called subjects of documents in a set of documents are found in documents in a hidden and unstructured form. The peculiarity of these topics is that they are often seen together in the text and usually consist of words that share a common or similar theme. In topic modeling, abstract topics are produced by clustering the words that are frequently seen together in the text, and the related texts are positioned in one or more clusters closest to the words they contain. Expressions that may be close to each other are grouped in the semantic space to form an abstract subject. These documents are then clustered according to the groups created and the words they contain. The high competition in the automotive industry, which is one of the largest industries of today, necessitated the businesses to attach importance to improvement studies. Among these improvements, the suggestions of the employees occupy a very important place. The fact that the content of the recommendation systems consists of texts has made them suitable data sets for advanced text mining studies. Analyzing employee suggestions with subject modeling will make it possible to make suggestions about which subjects are the most, which subjects should be focused on, and to make predictions about future suggestions and improvements. In this study, Latent Dirichlet Allocation (LDA), one of the text mining methods, was used to analyze the recommendation systems of an automotive company. Azure Machine Learning tool was used for analysis in our project. The most common type of suggestion is "positive suggestions with no return". These recommendations are generally related to occupational health and safety. 2. The most frequently given suggestions are "suggestions", that is, those that have returns and provide profit to the company. In the 3rd rank, there are "fast kaizene suggestions", that is, high-yielding suggestions that can be achieved in a very short time. 4. Suggestions that are "referred to be evaluated" are in the rank, while "recommendations that will not be put into action" seem to be the least given type of suggestion.