/DAT

Source code for Degree-Aware Alignment for Entities in Tail

Primary LanguagePython

DAT

Source code for the SIGIR paper, Degree-Aware Alignment for Entities in Tail

Data

Place the wiki.multi.de.vec, wiki.multi.fr.vec, wiki.multi.en.vec (obtained from MUSE) under "./data"

Run

Step 1: Generate concatenated power mean embedding

  1. cpm.py. The inputs are multilingual/monolingual word embeddings; the outputs are the word embeddings merely containing the words in the names of ent1 and ent2 (name2embed1.pkl and name2embed2.pkl).
  2. cpm2.py. The outputs are the embeddings of names (1-average, 3-cpm, 6-cpm(multilingual)).

Step 2: Choose a structural model

We use RSNs in our paper. As pointed out in the paper, other models are also viable, e.g., GCN and JAPE.

This code is based on GCN due to its simplicity. It can be easily replaced with RSNs.

Run bash run.sh to get the results.

CITATION

If you find our work useful, please cite it as follows:

@inproceedings{DAT,
	Author = {Weixin Zeng and Xiang Zhao and Wei Wang and Jiuyang Tang and Zhen Tan},
	Booktitle = {SIGIR 2020},
    Pages = {811--820},
	Title = {Degree-Aware Alignment for Entities in Tail},
	Year = {2020}
}