Image based agorithm for automatic generation of chinese couplets.

The article focuses on the development of an image-based intelligent generative model for Chinese couplets, a traditional form of Chinese literary expression. The proposed model first extracts features from an input image to generate a descriptive text, then uses an improved hybrid algorithm combining TextRank and TF-IDF to extract keywords from the description, which serve as input for a Chinese GPT-2 model to generate the first line of a couplet. Subsequently, the second line is produced using an encoder-decoder framework with Bi-LSTM encoding and GRU decoding enhanced by an attention mechanism. Experimental results, including BLEU automatic evaluation and manual assessments by diverse evaluators, demonstrate the model's effectiveness in producing coherent, corresponding, and rhythmically appropriate couplets that reflect image content. The study acknowledges limitations such as dataset scope and emotional expressiveness, suggesting future work on end-to-end image-to-couplet datasets, enhanced evaluation methods, and expansion to Chinese Spring Festival couplets.