For the employment intention recognition and guidance scenario of higher vocational college students, this paper constructs an accurate analysis method based on density peak clustering in the big data environment, and modeling is carried out around multi-source employment data cleaning, heterogeneous feature normalization, local density calculation and decision mapping. Students’ basic information, course performance, practice records, post browsing, delivery behavior and interview texts were uniformly coded to form a feature vector for intention clustering. Then, by introducing the adaptive truncation distance and relative neighborhood discrimination strategy, the highly similar group discovery and boundary sample identification were completed. On this basis, combined with the job tag library and clustering rules, the employment intention categories and recommendation results are output. On the validation set, the silhouette coefficient of the proposed method reaches 0.732, the CH index reaches 418.6, the accuracy of clustering decision reaches 92.4%, the hit rate of post guidance reaches 89.7%, and the average response delay is controlled at 0.41 seconds. Compared with the traditional K-means, hierarchical clustering and Gaussian mixture model, the proposed method performs better in terms of clustering stability, intention discrimination and decision consistency. It can provide computational technical support for the analysis and classification guidance of employment data in higher vocational colleges, and has continuous computing power and expansion space for post update, portrait correction and dynamic recommendation.
Povzetek: Za potrebe prepoznavanja in usmerjanja zaposlitvenih namer študentov višjih strokovnih šol ta članek v okolju velikih podatkov vzpostavlja analitično metodo, ki temelji na gručenju z gostotnimi vrhovi. Metoda izvaja čiščenje podatkov, kodiranje in gradnjo značilk na podlagi osnovnih informacij o študentih, učne uspešnosti, evidenc prakse in podatkov o vedenju pri iskanju zaposlitve. Eksperimenti so bili izvedeni na podatkovni zbirki s 4260 veljavnimi vzorci. Na validacijskem naboru je koeficient silhuete dosegel 0,732, indeks CH 418,6, natančnost skupinskega odločanja pa 92,4 %. Metoda lahko zagotovi računsko podporo in odločitveno podlago za analizo zaposlitve ter usmerjanje.
