用余弦相似度计算未定标项

Question

我想用这种方法计算具有余弦相似性的未评定项目。

import numpy as np; import pandas as pd
from sklearn.metrics.pairwise import cosine_similarity

dff = pd.DataFrame(np.random.randint(0, 10, (5, 3)))
temp = dff.copy()
dff
    0   1   2
0   8   0   4
1   6   9   4
2   5   0   5
3   5   9   4
4   9   4   8

cossim = cosine_similarity(dff) # calculate scores.
cossim

array([[1.  , 0.62, 0.95, 0.57, 0.92],
       [0.62, 1.  , 0.61, 1.  , 0.83],
       [0.95, 0.61, 1.  , 0.58, 0.95],
       [0.57, 1.  , 0.58, 1.  , 0.81],
       [0.92, 0.83, 0.95, 0.81, 1.  ]])

我想用余弦相似度来计算0值。

for x in range(0,dff.shape[1]):
    indexes = dff.index[dff.loc[:,x]==0].tolist()
    for y in indexes:
        dff.loc[y,x] = (cossim[y]*temp.loc[:,x].to_numpy()).sum()
dff

    0   1   2
0   8   21.307945   4
1   6   9.000000    4
2   5   34.528532   5
3   5   9.000000    4
4   9   4.000000    8

我用两个循环计算它？有什么节能型的计算方法吗？

有测试数据(实际值)

import numpy as np; import pandas as pd
from sklearn.metrics.pairwise import cosine_similarity
data = [['aa',1, 10], ['aa',2,6], ['aa',3, 5],['aa',4],
        ['bb',1], ['bb',2,5], ['bb',3,8],['bb',4,6],
        ['cc',7], ['cc',2], ['cc',3,4],['cc',4,9]] 
df = pd.DataFrame(data, columns = ['full_name','user_id', 'rating'])  
repo_matrix = df.pivot_table(index='full_name', columns='user_id', values='rating')
repo_matrix.replace(np.nan, 0, inplace=True)
repo_matrix
cossim = cosine_similarity(repo_matrix)
display(cossim)
temp = repo_matrix.copy()
repo_matrix = temp.mask(temp==0, cossim@temp)
repo_matrix

所有零转换为NaN。?？

Answer 1

你的操作就是矩阵乘法。所以你可以：

 # pass the numpy array instead of dataframe
 # also, you don't need to copy to temp 
 dff = dff.mask(dff==0, cossim @ dff.values)

输出：

   0         1  2
0  8  14.35119  4
1  6   9.00000  4
2  5  14.49324  5
3  5   9.00000  4
4  9   4.00000  8