R.S

@lingzhang0319 Multi-processing is better.In Python,since there is a GIL(Global Interpreter Lock),multi threads cant use multi cpu cores,but multi-processing can. As to original paper,I dont know why use multi threads ,...

On Taobao, my running cmd is `python3 -u ./src/train.py --model_type MIND --dataset taobao 2>&1 | tee MIND_1_taobao` and my result is as below ``` test recall: 0.070153, test ndcg: 0.208209,...

ComiRec_SA #### book ``` test recall: 0.079056, test ndcg: 0.048871, test hitrate: 0.162605, test diversity: 0.221267 test recall: 0.079164, test ndcg: 0.049336, test hitrate: 0.162953, test diversity: 0.208704 test recall:...

ComiRec_DR #### book ``` test recall: 0.078207, test ndcg: 0.065589, test hitrate: 0.169878, test diversity: 0.190251 test recall: 0.074753, test ndcg: 0.061694, test hitrate: 0.161645, test diversity: 0.189489 test recall:...

run cmd is as follows [run_cmd.zip](https://github.com/THUDM/ComiRec/files/7225665/run_cmd.zip) Most of the results do not achieve the effect of the paper, could the author reproduce it and upload the run command and log?...

Hi, @cenyk1230 , first thank you for your reply. In your mostpop code, ndcg and recall is different from train.py(evaluate_full), below I just take recall as an example. Suppose the...

If you can't show your strength, then you can't get anything at the negotiating table.

For me: 第一次用管理员权限运行, 之后就正常了

And in tf.nn.sampled_softmax_loss, use `num_sampled=self.neg_num * self.batch_size` is incorrect, you can refer to https://github.com/THUDM/ComiRec/issues/3

In your paper, you write `we treat all the items that the user has not interacted with as negative items` But in your code, you treat the result of `index.search(user_embs,...