Exploring the Impact of Negative Sampling on Patent Citation Recommendation

• pcr_patents.csv is the dataset which is generated by collecting samples randomly from Google Patents by exploiting a Python library. The dataset comprises around 250,000 US patents and their titles, abstracts, and citations.  Each patent has roughly on average 27 citations.

The zip file contains 3 different datasets for training and testing patent citation recommendation systems. These datasets were generated by utilizing the main dataset. They consist of around 1 million instances which are positive as well as negative samples.  

• pcr_cpc_negative_sample_data.csv  consists of negative samples that were generated based on CPC subclass codes. 

• pcr_random_negative_sample_data.csv consists of negative samples that were generated randomly. 

• pcr_sem_sim_negative_sample_data_2.csv consists of negative samples that were generated based on nearest neighbor relation.