Learners : Matchbox recommender : Command-line runners

Negative data generator

The concept of positive-only data and the need to generate negative data is explained in the corresponding data mapping section.

The negative data generator can be called by running “Learner Recommender GenerateNegativeData”. It takes two input parameters - the existing positive-only data file name and the name of the file to store the output to. For example:

Learner Recommender GenerateNegativeData --input-data PositiveData.dat   
                                         --output-data StarRatingData.dat

The input file, although containing only positive data, has to follow the guidelines for the runner’s data formatting. That is, it has to provide a rating descriptor, as well as a rating value. For example:

R,1,1  
u2,i3,1  
u2,i1,1  
u1,i1,1  
u3,i2,1  
u4,i4,1

The corresponding output will be:

R,0,1  
u2,i3,1  
u2,i1,1  
u2,i4,0  
u2,i2,0  
u1,i1,1  
u1,i3,0  
u3,i2,1  
u3,i1,0  
u4,i4,1  
u4,i1,0

Data in this format can now be fed into the recommender for training. If you are going to split the data into a train and test set, then you should generate negative data before splitting, so that the test data also has negative data.