A safe and dependable supply of energy and power is directly correlated with the quality of distribution network engineering. The assessment and diagnosis of the design quality and economic viability of a distribution network engineering process are essential for guaranteeing the steady functioning of the corresponding power system. In this paper, an intelligent assisted assessment technique for distribution network engineering is proposed to address the issues of inefficiency, high manual dependence, and low utilization of vital information in the text during the evaluation of projects related to distribution network engineering. To improve the model's contextual learning ability, the robustly optimized bidirectional encoder representations from transformers pretraining approach and whole-word masking are adopted to extract useful features from the distribution network engineering project review text. Principal component analysis is then used to downscale the high-dimensional features, thereby greatly increasing the efficiency of downstream classification. The light gradient boosting machine performs classification on the downscaled text features, and the Bayesian optimization approach is utilized to identify the best hyperparameter combinations. This significantly lessens the impacts of random parameters on the model performance. Tenfold cross-validation results demonstrate that the model can quickly and accurately identify common problems in distribution network projects' technical and economic dimensions.