Applied Research On Employment Of University Students Based On Semi-supervised Learning

With development and expansion of college and university, the number of students is increasingly growing; accordingly students’ employment has became a serious problem. At the same time, there is a large amount of data in university students information systems. On basis of these data, some operations such as query, modification, addition and deletion are not enough to get the interesting information, thus data mining is used to deal with these data in order to discover potential applications in university students employment service. It is one of the most focuses in the field of education research in recent years.There are some algorithms in analyzing employment data through data mining technology, such as, decision tree, clustering analysis and association rules. The core problem of decision tree algorithm is how to classify by using decision attribute data set. But it is difficult for method of decision attribute to apply to the large training set. Although ID3, ID4 and C4.5 are proposed, they do not consider correlation between attributes and error accumulation problem, and it could not fundamentally improve the classification accuracy rate. While the decision tree algorithm is a supervised learning algorithm, and university students employment data with unlabeled data, the robustness of the algorithm cannot be guaranteed. Clustering algorithm is unsupervised learning with no label guidance. Common used algorithm for analysis of college students employment data is K-means algorithm, but the clustering result of this kind of algorithm is depended on the initialization parameters. Students were clustered into one group when their employment units are similar, so it has not formed the employment guidance for the other students. Association rules algorithm can dig out the related factors that affect the employment of university students on the basis of these factors can provide reference for the cultivation of college students. We used to consider Apropri algorithm and its improved algorithm, but these algorithms generate a huge number of candidates, coupled with too many I/O operations and more execution, so the efficiency of algorithm is very low.In order to solve the aforementioned problems, the guiding role of the labeled data in the employment data and the supporting role of unlabeled data are considered to train the classifier with high generalization. The classifier is used to classify and predict students employment data, which can become the foundation of recommending jobs for students. The main work in this paper is as the following five points:(1)To analyze employment and relevant factors in different universities, we unite three tables, namely student grade, student management system and employment management system respectively, then we attain comprehensive data.(2)Considering graph-based method, we discuss graph-based through introducing kernel function into semi- supervised learning, it makes use of advantage of kernel, which solves the question that it did not divide using linear method in low dimension space, and then a graph semi-supervised learning algorithm based on kernel is proposed. At last, this article has carried on the contrast experiment, and the experimental results demonstrate the effectiveness and the feasibility of the proposed algorithm.(3)Based on sparse graphs, we discussed effectiveness of the method which can reflect the characteristics of the geometry and spatial structure between the data, proved the validity and rationality. At last semi-supervised learning via non-symmetric sparse graph is proposed. Validated by the symmetric-graph Laplace algorithm, the experimental results confirmed the feasibility of the proposed algorithm.(4)On the basis of Bayesian classification algorithm, we weaken the condition independent hypothesis of feature attributes, and made the decision attribute which had a more reasonable weight, that impacted on the classification more accurate. So improved weighted Bayesian classification algorithm is proposed. Finally, the experimental results validated the effectiveness of the improved algorithm.(5)The four points were proposed to put forward the problem of employment of university students. At last this paper pointed out the research goal and the direction of further work.