Douban model save condition

- ubuntu corpus
每训练完一个epoch，就会在验证集上跑一遍`utils/evaluation.py`，根据p1@10和p2@10的累计得分判断模型性能
- douban corpus
训练阶段保存模型的条件也是用`utils/evaluation.py`吗？因为我看test阶段会把脚本改成`utils/douban_evaluation.py`

Compare the training path that uses utils/evaluation.py with the test-stage reference to utils/douban_evaluation.py. Trace where the model-save condition is selected for the Douban corpus, then document which evaluation script and cumulative score determine saving, if the repository provides enough context to establish that.

Written by the indexing model from the issue text.