<p>This article comprehensively utilises the BERT (Bidirectional Encoder Representations from Transformers) and ViT (Vision Transformer) deep learning algorithms to explore the intrinsic relationship between college student organisations and ideological and political education. It aims to promote more efficient and personalised teaching practices. Initially, semantic features are extracted from textual data, and sentiment analysis and topic modelling are conducted. Subsequently, key visual information is extracted from images, and the participation level and interaction patterns in activities are analysed. Textual and visual features are integrated, and a Multilayer Perceptron (MLP) is utilised to construct a comprehensive evaluation model. This model analyses students’ behaviours and emotional tendencies in ideological and political education and organisational activities. It quantitatively assesses students’ participation, emotional tendencies, and the effectiveness of organisational activities. The research findings indicate that the satisfaction level with ideological and political education in some student organisations reaches 75%, while the level of knowledge mastery is only 68%. There is a need to give more consideration to the unique needs of this group in the future. Based on deep learning algorithms, an effective analysis of the intrinsic relationship between college student organisations and ideological and political education can be conducted.</p>