如何通过 LightFM python 包生成用户对用户的推荐?
How can I generate users to user recommandation via LightFM python package?
我正在通过以下代码创建数据集:
from lightfm.data import Dataset
from lightfm import LightFM
dataset = Dataset()
dataset.fit((row['id'] for row in user_queryset.values()),
(row['id'] for row in item_queryset.values()))
num_users, num_items = dataset.interactions_shape()
(interactions_sparse_matrix, weights) = dataset.build_interactions(
(
(
row['user_id']
,row['item_id']
,row['weight']
)
)
for row in queryset.values()
)
dataset.fit_partial(
items=(x['item_id'] for x in items_list),
item_features=(x['feature_id'] for x in item_features_list)
)
dataset.fit_partial(
users=(x['user_id'] for x in users_list),
user_features=(x['feature_id'] for x in user_features_list)
)
item_features = dataset.build_item_features(
((x['item_id'], [x['property_id']])
for x in item_features_list))
user_features = dataset.build_user_features(
((x['user_id'], [x['property_id']])
for x in user_features_list))
我通过以下方式生成火车模型:
model = LightFM(loss='bpr')
model.fit(
interactions_sparse_matrix
,item_features=item_features
,user_features=user_features
)
然后我使用 sklearn
的 cosine_similarity
方法来获得相似之处:
from scipy import sparse
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
users_sparse_matrix = sparse.csr_matrix(users_embed)
similarities = cosine_similarity(users_sparse)
但是当打印 similarities.shape
它的 return :
(14, 14)
虽然我有 5 个用户,但我认为它一定是 (5,5) ,我错了吗?像这样的矩阵:
1 0.2 0.8 0.4 0.6
0.2 1 ... ... ...
0.8 ... 1 ... ...
0.4 ... ... 1 ...
0.6 ... ... ... 1
如何获取用户及其评分以推荐给用户?谢谢
我的 LightFM 版本是:1.15
我用的是python3.6
问题不在于您的代码。对user_embedding这个概念有误解。 user_embedding矩阵是以用户特征个数为行,分量个数为列的矩阵。当你有这个矩阵时,为了得到每个用户之间的相似度与余弦相似度,你需要将 user_feature 矩阵乘以 user_embedding,最后计算 [=14] 的点积的余弦相似度=] 矩阵与 user_embedding 矩阵。
我正在通过以下代码创建数据集:
from lightfm.data import Dataset
from lightfm import LightFM
dataset = Dataset()
dataset.fit((row['id'] for row in user_queryset.values()),
(row['id'] for row in item_queryset.values()))
num_users, num_items = dataset.interactions_shape()
(interactions_sparse_matrix, weights) = dataset.build_interactions(
(
(
row['user_id']
,row['item_id']
,row['weight']
)
)
for row in queryset.values()
)
dataset.fit_partial(
items=(x['item_id'] for x in items_list),
item_features=(x['feature_id'] for x in item_features_list)
)
dataset.fit_partial(
users=(x['user_id'] for x in users_list),
user_features=(x['feature_id'] for x in user_features_list)
)
item_features = dataset.build_item_features(
((x['item_id'], [x['property_id']])
for x in item_features_list))
user_features = dataset.build_user_features(
((x['user_id'], [x['property_id']])
for x in user_features_list))
我通过以下方式生成火车模型:
model = LightFM(loss='bpr')
model.fit(
interactions_sparse_matrix
,item_features=item_features
,user_features=user_features
)
然后我使用 sklearn
的 cosine_similarity
方法来获得相似之处:
from scipy import sparse
from sklearn.metrics.pairwise import cosine_similarity
import numpy as np
users_sparse_matrix = sparse.csr_matrix(users_embed)
similarities = cosine_similarity(users_sparse)
但是当打印 similarities.shape
它的 return :
(14, 14)
虽然我有 5 个用户,但我认为它一定是 (5,5) ,我错了吗?像这样的矩阵:
1 0.2 0.8 0.4 0.6
0.2 1 ... ... ...
0.8 ... 1 ... ...
0.4 ... ... 1 ...
0.6 ... ... ... 1
如何获取用户及其评分以推荐给用户?谢谢
我的 LightFM 版本是:1.15
我用的是python3.6
问题不在于您的代码。对user_embedding这个概念有误解。 user_embedding矩阵是以用户特征个数为行,分量个数为列的矩阵。当你有这个矩阵时,为了得到每个用户之间的相似度与余弦相似度,你需要将 user_feature 矩阵乘以 user_embedding,最后计算 [=14] 的点积的余弦相似度=] 矩阵与 user_embedding 矩阵。