Under the background of digital economy, the slow employment behavior of college students shows the characteristics of platform and continuous. The traditional recognition methods relying on questionnaires, static records and manual judgment are difficult to reflect the change of job hunting status. This paper constructs a framework for dynamic identification and strategy optimization of slow employment, integrates data such as recruitment platform logs, on-campus employment records, resume text, job semantics, consultation interaction and ability portrait, and based on 84,560 behavior samples of 2436 students in 8 months, after missing repair, feature coding, normalization and sliding window segmentation, it constructs a framework for dynamic identification and strategy optimization of slow employment. Local pattern extraction, temporal dependence modeling and attention aggregation are used to realize state recognition, risk prediction and intervention recommendation. Experimental results show that the accuracy of the model is 93.4%, the F1 value is 91.3%, and the comprehensive response efficiency is improved by 16.7%, which can provide computational support for digital employment services in colleges and universities.
