Applying classifications techniques in job recommendation system for matching of candidates and advertisements

Because of the increase in the number of internet users and the increase in the amount of knowledge correspondingly in the recent years, being able to find an appropriate candidate\job has become a real tough process. Job proposal systems have been developed with an attempt to find the desired jobs and candidates among this enormous data with ease. Job proposal systems are frequently used within the field of e-trade today. The online recruitment platforms are being used as primary hiring channels by a great number of companies with the aim of decreasing the employment periods and advertisement expenses. The most common technique in job search is that a user should input his or her criteria in the job search websites and then the system should present the user the job advertisements\candidates that match with the criteria of the system. The user evaluates the advertisements\candidates according to his\her own criteria in line with the results. The high number of matching advertisements\candidates makes it difficult for users to find the most appropriate job\candidate and causes a loss of time. Job proposal systems aim to eliminate this manual searches of the users and present the optimum adverts\candidates to the users. Job proposal systems have been designed in a way to present the candidates with the list of the optimal jobs that they can apply to or the employers with the candidates that are best fit for their requirements. Typically, job proposal systems offer the candidates and adverts a list of proposals. While these systems try to find the appropriate adverts\candidates, they should also not present the adverts\candidates that are not related with the user profile to the users. For this reason, high-quality job proposal systems make use of approaches that are suitable with the characteristics of the users. There is a wide range of strategies that are used for job proposals. This study aims to discuss a job proposal system that handletheusers' profiles, interaction features and preferences all together. The proposed system primarily includes the candidates that are similar to the target candidate and the job adverts applied by similar candidates. Then the answers that the advertisements will return to the target candidate are obtained by using the preference information of the advertisements. And finally, the advertisements that prefer mutually the target candidates are suggested to the candidate in order. The success rates of the system for classification methods that is proposed using different classification methods have been measured to estimate the return values of the advertisements to the target candidate. A new method based on the preference information of the candidates is used to determine the order of the advertisements in the proposed list. The proposed system has been evaluated by using a data set received from job search sites. The data set used in the proposed model consists of 7455 applications made for 10 advertisements and there also exist the answers given by the advertisements to the applications. For this reason, the focus is centered primarily on the determination of the return values of the advertisements. For this reason two methods have been used to obtain the return values. The first method uses the activity information of the advertisements on the background of the candidates. By using this activity information, the rules governing which applications returned with positive which ones returned with negative responses were determined. Applying these rules to the data set resulted in a positive return for a total of 748 applications. The second method that was used to determine the return values of advertisements is the Trust Rank (TR) method. The TR method is a method used to detect spam users on the web. The reason why this method is used is the similarity between the problem of identifying spam users and the problem of determining the return values of advertisements in the data set used. The TR uses the approach which states that by using the source\tag similarities, the users giving the same source the same tag namely exhibiting the same behavior should be similar. With this approach, a small number of pages known to be spam users are used with TR to determine the tag values of pages that are not known to be spam users. This approach was also thought to be appropriate for the data set used, and it is assumed that the candidates for which the advertisements would return positively will be similar among themselves. With this assumption, the return values (positive / negative) of the advertisements were determined using the TR method with a small number of applicants manually tagged and a positive return values were obtained for 1967 applications. The steps of the method proposed for the results obtained by these two methods used for the determination of return values of advertisements are applied separately and the obtained values are compared. The proposed method consists of 4 parts. The first part is creating the announcement list to be proposed to the candidate. In order to make a recommendation for a candidate who is entering the system just recently, firstly the candidates similar to this candidate in question are found. Subsequently, the advertisements applied by the previous similar candidates are found and listed. The second part is the determination of the return values of the advertisements. The return values obtained from the section of the determination of the return values of the advertisements are trained by the classification methods for each advertisement, and what kind of a response the advertisements applied by similar candidates will return to the target candidate is calculated. In this way, the information whether they will be preferring the target candidate for each advertisementis obtained. The third part is determining the importance values of the advertisements. A confidence value is calculated for each candidate by using the applications in the system that were previously carried out by the candidates. Trust value is the value determined from the similarities of the advertisements made by a candidate. If the similarity among the advertisements the candidate applied to is very high, the confidence value of the candidate is high, if it is low then the confidence value of the candidate is low. Each candidate provides his/her confidence value to the advertisement they apply to. The importance value of the advertisement is determine through the sum of the candidates who are applying to it and these importance values are accepted as the score value of the advertisement. Thus, a score value for each advertisement is obtained. The fourth and the last part is the formation of a consecutive advertisement proposal list. For each advertisement to be proposed to the candidate, the return value of the advertisement for the candidate and score values are multiplied. Then the advertisements are proposed to the target candidate in sequences in line with the obtained values. Two steps of the method that was used for the determination of return values of advertisements on applications and the steps of the method proposed for the results obtained by means of two methods are separately applied and two studies were formed. The results obtained as a result of these two studies were evaluated using the MAP and TNR measurements. When looking at the results; it was seen that the experiments conducted on return values obtained using TR show higher accuracy than the results obtained using rules.