A Hybrid Collaborative Filtering and LDA-Based Subject Model for Bidirectional Employment

To optimize employment matching in colleges and universities, a hybrid bidirectional model was designed to recommend suitable companies to graduates and vice versa. First, the resume submission records of graduates and interview invitation data from enterprises were integrated to construct a sparse matrix, and K -means clustering was applied to fill missing values and mitigate data sparsity. Resumes and recruitment texts were analyzed using a latent Dirichlet allocation (LDA) topic model, combined with TF-IDF weighting, to generate a graduate–enterprise feature vector space. By dynamically weighted fusion of collaborative filtering (CF) similarity and topic model similarity, a hybrid recommendation coefficient was obtained to achieve efficient bidirectional recommendation. Experimental results revealed that the CF-LDA model constructed with weight coefficients ( M =0.45 and N =0.55) significantly improved recommendation performance: the overlap rate between graduate resumes and enterprise recommendations reached 92%, the overlap rate between enterprise interview invitations and recommended talents reached 96%, and user activity increased by more than threefold (graduates and enterprises spent 65 and 141 min online per day, respectively). Compared with a single CF or LDA model, its recall rate increased by 18%–28% and the RMSE reduced by 23%–45% on publicly available datasets such as MovieLens 10M, verifying the effectiveness and generalization ability of the model. The data results indicate that the research model provides effective employment recommendation information for graduates and enterprises, improving the efficiency of graduates’ job search and enterprise recruitment.