MIRRec

Files

The main function is in MyRecEditTrain.py. Use the 'recommendByMyRecEdit' for hypergraph-base model training and use the 'evaluationAlgorithm' to get the evaluation results.

The other .py files are invocated methods for HyperGraph construction.

the .zip files are the preprocessed data of different projects in the dataset, named after the corresponding project.

RD disscussion.pdf is the result of Recommendation distribution of MIRRec and baselines

Steps

Input: The pre-processed training dataset (zip file).

Source code: MyRecEditTrain.py

Runing example：

parameters setting

# time interval selection
dates =[(2018, 6, 2019, 6), (2018, 6, 2019, 7), (2018, 6, 2019, 8), (2018, 6, 2019, 9), (2018, 6, 2019, 10),
             (2018, 6, 2019, 11), (2018, 6, 2019, 12), (2018, 6, 2020, 1), (2018, 6, 2020, 2), (2018, 6, 2020, 3)]
#projects selection
projects = ['electron', 'opencv', 'xbmc', 'react', 'angular', 'django',
             'symfony', 'rails', 'scala']
alpha = 0.9
K = 10
re_arr = [1, 2, 3, 4, 5]#Top-k
c = 0.8

training

for re in re_arr:
    for p in projects:
          MyRecEditTrain.recommendByMyRecEdit(train_data, train_data_commit, train_data_issue_comment,
                                   train_data_review_comment, train_data_y, train_data_y_workload,
                                   train_data_committer, train_data_issue_commenter, train_data_review_commenter,
                                   test_data, test_data_commit, test_data_y,
                                   test_data_y_workload, test_data_committer, date,
                                   project, convertDict, recommendNum=5,
                                   K=10, alpha=0.8, c=1,
                                   TrainPRDisIsComputed=False,
                                   HyperGraphIsCreated=False
                                   , re=4, ct=3, ic=1, rc=1)

Evaluation for metrics

for re in re_arr:
    for p in projects:
        HyperGraphIsCreated = True  # continue training without initialization, or construct the initial hypergraph(False)
        TrainPRDisIsComputed = True  # pr-pr weight updated (for optimal k)
        MyRecEditTrain.evaluationAlgorithm(p, dates, alpha=alpha, K=K, c=c, TrainPRDisIsComputed=TrainPRDisIsComputed,
                                     HyperGraphIsCreated=HyperGraphIsCreated, re=4, ct=3, ic=1, rc=1)

cufeinfor / MIRRec