Network user two-stage recommendation method and device based on large language model optimization

The invention discloses a network user two-stage recommendation method and device based on large language model optimization, and the method comprises the following steps: firstly, in a recall stage, a user accelerates the feature extraction according to the browsing history of the user in combination with a recursive parallelization mode; then, through a double-layer MLP network, modeling is carried out according to interaction behaviors of users; then, 20 items with the highest scores are calculated through a prediction module and serve as a candidate set; secondly, in the sorting stage, the open source large language model and the closed source large language model are combined; by means of a closed-source large language model, a prompt template is designed, information such as interaction history and social relations of users is deeply analyzed, and user interest portraits are constructed. Meanwhile, by means of the reasoning ability of a large language model, text content is summarized, new description i