FAKE JOB POSTING DETECTION

The researchers took part in the Russian competition of artificial intelligence named "RuCode Fake Job Postings". So this work is devoted to solving the problem of fake job posting detection. This task has some similar works that were studied by the researchers to make a better performance in the competition. The data were analyzed and processed carefully. Python libraries such as pandas and matplotlib were used for the analysis. As for data processing, the researchers decided, to use some techniques of the English text preprocessing and a couple of feature extractors such as TF-IDF and word2vec methods. Several different algorithms were trained and compared with each other. They are as follows: Logistic Regression, K-nearest neighbors, Random. Forest, Bi-directional LSTM, BERT, ALBERT and RoBERTa. The results of their performance were summarized in a single table which shows the achieved, F1-score of the quality of solving our problem. As the main result the researchers found out that the best algorithm for the task under consideration was RoBERTa, which allowed to win the competition. There is also a room for improvement. So we concluded that our solution could be improved by using larger datasets, modifying the training and validation schemas and trying to use more tips and tricks in algorithms architectures and data preprocessing.