IPRE

IPRE: a Dataset for Inter-Personal Relationship Extraction

Dataset/Format

Files	Contents
sent_train/dev/test	sentID   eh   et   sentence
sent_relation_train/dev/test	sentID   relationIDs
bag_relation_train/dev/test	bagID   eh   et   sentIDs   relationIDs

The above table shows the format of our data set. Some keywords are explained as follows:

• e_h   The head entity in the sentence.
• e_t   The tail entity in the sentence.
• sentence   A segmented sentence contains the e_h and e_t.
• sentID   A unique ID given to an instance, which consists of a ordered entity pair and a sentence.
• bagID   A unique ID given to a bag, and all setences in a bag contains the same entity pair.
• relationID   A unique ID given to a relation type.

Note that the part represented by the keywords mentioned above is separated by a tab character. Moreover, relationIDs is composed of several relationID that separated by a space character, and sentIDs is also so.

Dataset/Files

Each dataset contains three files, in which one file stores the mapping information between sentID and instance, and the remaining two files store the data in dataset at sentence-level and bag-level respectively.

For example, test set contains sent_test.txt, sent_relation_test.txt and bag_relation_test.txt. The mapping information between sentID and instance in test set is stored in sent_test.txt, and the data in test set at sentence-level and bag-level are stored in sent_relation_test.txt and bag_relation_test.txt respectively.

Reference

If you use the IPRE data, please cite the following paper:

• Haitao Wang, Zhengqiu He, Jin Ma, Wenliang Chen, Min Zhang. 2019. IPRE: a Dataset for Inter-Personal Relationship Extraction. https://arxiv.org/abs/1907.12801