下面是使用GroupBy和pandas.merge的方法：

g = df.groupby(by=['source', 'type'], sort=False)
​​
out = (
            pd.merge(df.loc[g['similarity'].idxmin()]
                       .rename(columns= {'similarity': 'sim_min', 'target': 'target_min'}),
                     df.loc[g['similarity'].idxmax()]
                       .rename(columns= {'similarity': 'sim_max', 'target': 'target_max'}),
                     on=['source','type'])
              .assign(min=lambda x: x.pop('target_min') + ': ' + x.pop('sim_min').astype(str),
                      max=lambda x: x.pop('target_max') + ': ' + x.pop('sim_max').astype(str))
        )

另一种变体：

g = df.groupby(by=['source', 'type'], sort=False)
​
​
out = (
            pd.merge(df.loc[g['similarity'].idxmin()]
                         .assign(min= lambda x: x[['target', 'similarity']]
                                                     .astype(str).agg(": ".join, axis=1)),
                     df.loc[g['similarity'].idxmax()]
                         .assign(max= lambda x: x[['target', 'similarity']]
                                                     .astype(str).agg(": ".join, axis=1)),
                     on=['source','type'], suffixes=('', '_'))
                .loc[:, ['source', 'type', 'min', 'max']]
      )

#输出：

print(out)

    source    type             min             max
0  Person1     bow  Person2: 0.636  Person3: 0.699
1  Person1  bigram   Person2: 0.04  Person3: 0.171
2  Person1   tfidf  Person2: 0.433  Person3: 0.522

示例

data = [['Person1', 'Person2', 'bow', 0.636],
        ['Person1', 'Person2', 'bigram', 0.04],
        ['Person1', 'Person2', 'tfidf', 0.433],
        ['Person1', 'Person3', 'bow', 0.699],
        ['Person1', 'Person3', 'bigram', 0.171],
        ['Person1', 'Person3', 'tfidf', 0.522]]
df = pd.DataFrame(data, columns=['source', 'target', 'type', 'similarity'])

df

source  target  type    similarity
0   Person1 Person2 bow     0.6
1   Person1 Person2 bigram  0.0
2   Person1 Person2 tfidf   0.4
3   Person1 Person3 bow     0.7
4   Person1 Person3 bigram  0.2
5   Person1 Person3 tfidf   0.5

进程

df.groupby(['source','type']).agg([min, max])

结果：

target              similarity
                min     max         min     max
source  type                
Person1 bigram  Person2 Person3     0.0     0.2
        bow     Person2 Person3     0.6     0.7
        tfidf   Person2 Person3     0.4     0.5

使结果到你想要的输出
将:加到df的目标列的值上，将result的值改为str，然后将它们组合起来。

完整代码和输出

(df.assign(target=df['target'] + ': ')
 .groupby(['source','type']).agg([min, max]).astype('str')
 .groupby(level=1, axis=1, sort=False).sum().reset_index())

输出：

source  type    min             max
0   Person1 bigram  Person2: 0.04   Person3: 0.171
1   Person1 bow     Person2: 0.636  Person3: 0.699
2   Person1 tfidf   Person2: 0.433  Person3: 0.522