convert user / item view data into 2D metrics

Question

User / item view data is available as below

User  Item 
Louis    1
Louis    2
Adam     1
Adam     3

And I want to transform it into an item by item metric like

    1  2  3
1   0  1  1
2   1  0  0
3   1  0  0

So each value represents "number of person who viewed i also viewed j". (diagonal value does not matter)

Is there any efficient way of doing this?

Below is my code but took a long time to run when there is around 50k items and 500k view records.

import pandas as pd
import numpy as np
from scipy.sparse import csr_matrix, lil_matrix
raw = pd.DataFrame(columns=['user','item'])
raw['user']=['Louis','Louis','Adam','Adam']
raw['item']=[1,2,1,3]

item_list = raw.item.unique().tolist()
user_list = raw.user.unique().tolist()
m = lil_matrix((len(raw.item.unique()),len(raw.item.unique())))
for user in user_list:
    temp = raw.loc[np.in1d(raw['user'], user)].item
    if len(temp) > 1:
        for idx1, id1 in enumerate(temp[0:-1]):
            for id2 in temp[idx1+1:]:
                m[item_list.index(id1),item_list.index(id2)]+=1
                m[item_list.index(id2),item_list.index(id1)]+=1
m.toarray()

Answer 1

You could use

In [147]: dff = pd.crosstab(df.Item, df.User)

In [148]: dff = dff.dot(dff.T)

In [149]: np.fill_diagonal(dff.values, 0)

In [150]: dff
Out[150]:
Item  1  2  3
Item
1     0  1  1
2     1  0  0
3     1  0  0