我如何在LightFM电影推荐系统的用户项交互矩阵上进行交叉验证？

Question

我有一个来自movielens数据集的交互矩阵（scipy.sparse.csr_matrix），具有来自用户的电影评分，并且我正在建立一个具有item_features的LightFM模型。现在，我将矩阵分为训练和测试，但是如何对其进行交叉验证？如何衡量效率？

!pip install lightfm
from lightfm import LightFM, cross_validation
from lightfm.evaluation import precision_at_k, auc_score

train, test = cross_validation.random_train_test_split(user_item, test_percentage=0.25)
model_lightfm = LightFM(loss='warp', learning_rate=0.01, k=10)
model_lightfm.fit(train, item_features=item_features, epochs=50)

def recommend(model, user_id):
  n_users, n_items = train.shape
  best_rated = ratings_df[(ratings_df.userId == user_id) & (ratings_df.rating >= 4.5)].movieId.values

  known_positives = metadata.loc[metadata['MOVIEID'].isin(best_rated)].title_clean.values

  scores = model.predict(user_id, np.arange(n_items), item_features=item_features) 
  top_items = metadata['title_clean'][np.argsort(-scores)]

  print("User %s likes:" % user_id)
  for k in known_positives[:10]:
    print(k)

  print("\nRecommended:")
  for x in top_items[:10]:
    print(x)

recommend(model_lightfm, 10)


train_precision = precision_at_k(model_lightfm, train, k=10, item_features=item_features).mean()
test_precision = precision_at_k(model_lightfm, test, k=10, item_features=item_features, train_interactions=train).mean()

train_auc = auc_score(model_lightfm, train, item_features=item_features).mean()
test_auc = auc_score(model_lightfm, test, item_features=item_features, train_interactions=train).mean()

print('Precision: train %.2f, test %.2f.' % (train_precision, test_precision))
print('AUC: train %.2f, test %.2f.' % (train_auc, test_auc))

Answer 1

可用于lightFM的评估指标是auc_score和precision @ k。您正在计算指标Precision和auc_score。您可以通过查看

来判断模型的效率

1. auc_score for test-告诉您模型在考虑所有预测的电影的情况下，在为用户预测正确的推荐方面有多好。不考虑预测电影的顺序/等级。

2. precision_at_k进行测试-告诉您模型的精度，生成最高k（在您的情况下为10）预测。如果您想在生成top-n建议时判断模型，则有帮助。