What goes to the model and how to get those features

[nirajdpandey]

I was wondering that how you have generated those files in data folder out of two KG. How to get those rel_ID's, ref_pairs, triple_1, triple_2 data. Does this repository already contains script to generate those features out of two KG. I am pretty new to this please excuse me for asking naive questions. Thank you very much.

[sunzequn]

Hi,

Thank you for your interest in our work.

To generate new datasets for entity alignment and run existing embedding-based methods, I suggest you to consider the following steps:

(1) Let us assume that there are two KGs with relation triples (e.g., triples_1 and triples_2).

(2) Extract aligned entity pairs between such two KGs and split them into training/test sets (e.g., sup_pairs and ref_pairs).

(3) Convert the URIs of entities and relations into IDs (the ent_ids and rel_ids files contain the ID-URI mappings) such that the embedding models can directly read.

For generating IDs, you can refer to the code from OpenEA.

Zequn Sun