Performance comparison over different item groups

[ksh960910]

Hi,
Thanks for the great work with SimGCL.
I'm trying to replicate the results of Figure 7 from SimGCL

Is there any source code that I can refer to?
I would appreciate your reply.

Best,
Joe

[Coder-Yu]

I just have the code for this part in another paper of mine https://arxiv.org/abs/2209.02544

For your reference

def group_evaluation(self,recList,top):
     popularity = {}
     for item in self.data.training_set_i:
         popularity[item] = len(self.data.training_set_i[item])
     testset = []
     for user in self.data.test_set:
         for item in self.data.test_set[user]:
             if item in popularity:
                 testset.append((user,item,popularity[item]))
     testset = sorted(testset,key=lambda d:d[2])
     p_idx = np.linspace(0,len(testset),11)
     p_idx = [round(n) for n in p_idx]
     groups = []
     for i in range(10):
         group = {}
         for user in recList:
             group[user]={}
         for k in testset[p_idx[i]:p_idx[i + 1]]:
             group[k[0]][k[1]] = 1
         groups.append(group)
     measures=[]
     for group in groups:
         ms = ranking_evaluation_group(group, recList, self.data.test_set,top)
         measures.append(ms)

[ksh960910]

Thanks for the reference, however can I also get the code for 'ranking_evaluation_group' function?
I'm also having trouble getting recall scores from each groups.
I would really appreciate it. Thanks

Best,
Joe

[Coder-Yu]

The core code has been provided. Do it yourself