Empowering college students to select ideal advisors: a text-based recommendation model

College students often encounter numerous challenges throughout their academic journeys, making the guidance and support from educators indispensable. Recommendation systems can significantly reduce the difficulties students face when identifying suitable academic advisors. This paper proposes a A d V isor R ecommen D ation (AVRD) model based on textual data regarding college students' interests. AVRD first adopts Chinese Bidirectional Encoder Representations from Transformers (BERT) and unsupervised Simple Contrastive Learning of Sentence Embeddings (SimCSE) to train the corpus of advisors' records. The time decay factor is then introduced as the weight of the text record vectors, and the representation vectors of advisors are obtained using the weighted mean. Finally, the similarities between the advisor and student vectors are computed, and an advisor list is recommended to student according to the designed pooling and matching criteria. The questionnaire data from 170 college students are collected to evaluate the proposed model. Experimental results demonstrate the effectiveness of AVRD. The model outperforms other LLMs such as Qwen and DeepSeek by a significant margin, as well as the commonly used models like TF-IDF, LSA, and Word2Vec. Moreover, the ablation studies reveal that the SimCSE component of AVRD is crucial to the model's performance.