Jobs from Headhunter Portal

The purpose of this work is to develop a model that allows one to predict wages based on the job description. The project consists of the following stages: setting the task, collecting data, preprocessing, creating a model and implementing a service. The Head Hunter aggregator is used as a data source, providing its own API for parsing information characterizing vacancies. The formed dataset consists of 300,000 vacancies posted in 2021. Appropriate preprocessing methods have been applied to all features, depending on the type of data, representation, distribution of values, etc. The paper presents the stages of preprocessing to prepare the data for use in various models. To solve the problem, a number of machine learning models of different architectures and based on several sets of features have been developed. The values of the selected metrics of the developed algorithms in comparison with the metrics of the analytical solution created as a baseline demonstrate the importance of the set of features used by the model and its general generalizing ability. With the use of the best model, a service based on the telegram bot has been created that allows one to predict wages by providing a link to a vacancy on the Head Hunter portal. In order to achieve that, the entire pipeline described earlier is reproduced: data collection from the Head Hunter portal, preprocessing of features and application of the model. The described interface allows one to implement an inference model for new vacancies.