As an important carrier of cultural metaphor and human portrayal in English literature, the symbolism of residential space is often presented through complex linguistic forms and narrative structures. In this study, we propose an innovative approach that integrates literary criticism theory and natural language processing technology, and construct the Attention-BERT sentiment analysis framework based on the attention mechanism and the BERT model, aiming to systematically analyze the multidimensional symbolism of houses in English literature. Three core analysis clues are refined at the theoretical level. On the technical level, the Attention mechanism is introduced to dynamically weight the key semantic units, and the BERT bidirectional coding capability is utilized to parse the masked language, realizing the end-to-end mapping from text to sentiment tendency. The experiments validate the model performance by comparing the baseline models such as LSTM, RoBERTa, etc. Attention-BERT takes a significant lead with an accuracy of 83.53% and an F1 value of 80.05%, which is an improvement of 10.49 percentage points compared to the 73.04% of the traditional LSTM, and verifies the validity of combining the attention mechanism with the pre-trained model. Further quantitative analysis shows that the residential spatial text presents high diversity in lexical features, with a class character shape character ratio of 12.01% versus a high density feature, and a 67.30% share of real words. The frequent use of nouns (29.15%), verbs (20.91%) and adjectives (9.50%) strengthened the spatial figurative and metaphorical expression. At the syntactic level, the significant high-frequency distributions of quadratic lexical clusters (NSPC, VSPC), with all LLR values >21.94, p<0.01, reveal the thematic association of space with power and identity.