Preprocessed Dataset. After preprocessing the resumes, all resumes are in the form of an ID and their content information about each section. For each section, three pieces of information on the section name, the section content, and the lines that corresponding to the content are extracted. For section name, there are ‘Education’, ‘Work Experience’, ‘Activities’, ‘Skills’, ‘Profile, and ‘Other’. The section content is the text content under one section and the corresponding lines are the content about all the lines in the section content and the tokenization form of the lines. In other words, the preprocess data is like this. See figure 3.1. Figure 3.1: Cleaned Resume Content corresponding to different sections Chapter 4 Approach Approaches are various related to text classification. However, since the specialty of this dataset that every job has its own job admission standards, rule-based method can be used before doing unsupervised learning. Rule based method is mainly about using string matching to catch the impor- tant information about the job requirement description, then grabbing key information [15]from the preprocessed resumes, then make comparison, and finally judge whether the resume has all the required information in the job requirement description (Section 4.1). In addition to rule-based methods, unsupervised learning methods can be used on feature vector, which is a vector with several dimension to represent the information of the key words in different sections (Section 4.2). Instead of representing key information of the sections, augmenting the matrix allows us to use matrices to represent the whole text information in each section, which is known as Bags of Words method (Section 4.3). Finally, by mentioning the three approaches, ensemble models are also worthy trying and will introduce more in detail about the decision on choosing the components in the ensemble models in Section 4.4.
