same prediction for every user

Hi, I am really new to recommender system and to lightfm. I am trying to see what the model I build recommender for all my users (considering all my items), but I keep getting the very same prediction (recommendation) for every single user.

Here the code I used :

param = {'no_components': 314,
         'learning_schedule': 'adagrad',
         'loss': 'warp-kos',
         'learning_rate': 0.010372998003563394,
         'item_alpha': 1.0041412735137758e-06, 
         'user_alpha': 4.364266509627352e-09, 
         'max_sampled': 20,
         'num_epochs': 57}
num_epochs = param.pop("num_epochs")
model = LightFM(**param)

model = model.fit(interaction,
                  user_features=user_features,
                  item_features=item_features,
                  epochs=num_epochs,
                  num_threads=4)

mapp = dataset.mapping()
dict_user_id = mapp[0]
dict_item_id = mapp[2]

pid_array = np.arange(len(dict_item_id), dtype=np.int32)
d_user_pred = {}

for user in dict_user_id.keys():
    d_user_pred[user] = []

for index, user_id in dict_user_id.items():    
    sys.stdout.write("\rProcessing user " + str(index + 1) + "/ " + str(len(dict_user_id)))
    sys.stdout.flush()
    scores = model.predict(index, np.arange(len(dict_item_id)),
                           user_features=user_features,
                           item_features=item_features)
    top_items = item['item_id'][np.argsort(-scores)]
    d_user_pred[index] = top_items

My dataset shape is : 20.741 users and 2530 items. Also, I have really low p@k, maybe that the problem ?

[clementechiu]

Hey @RaphLot ! I’m having a similar problem. Did you manage to solve it? What did you do? I’d appreciate any help. Thanks in advance!

[clementechiu]

No, that means there’s something wrong with your model or data. You should have different recommendations, otherwise using a model is pointless.