This directory contains datasets and implementations of context-guided relation embedding (CGRE) proposed in our paper: Context-guided Self-Supervised Relation Embeddings.
It also implement neural latent relational analysis (NLRA) for word-pair representations, a method proposed by Washio and Kato, 2018 in their paper: Neural Latent Relational Analysis to Capture Lexical Semantic Relations in a Vector Space.
We used python 3.6.1. These codes require tensorflow, sklearn and numpy. This software includes the work that is distributed in the Apache License 2.0.
This project contains the following data files:
Pretrained word-pairs Embeddings: the folder includes pre-trained word-pair embeddings for SemEval-2012 Task2. Embeddings for four different methods are available as follows:
1. SemEval_CGRE_Gold.npy: supervised method trained on gold relation labels for DiffVec data in DiffVec_Pairs file 2. SemEval_CGRE_Proxy.npy: self-supervised method trained on pseudo labels of DiffVec in DiffVec_Pseudo_Label.txt 3. SemEval_MnnPL.npy: no contextual patterns are used to train Multi-class neural netword penultimate layer model. 4. SemEval_NLRA.npy: Neural Latent Relational Analysis word-pair embeddings
Read_PreTrained_WordPairs.py: a python script to read aforementioned pre-trained word-pairs embeddings.
DiffVec_Pairs: a text file of word-pairs in DiffVec dataset with gold relaiton labels used to train CGRE_Gold
DiffVec_Pseudo_Label.txt: a text file of word-pairs in DiffVec dataset with pseudo relaiton labels used to train CGRE_Proxy. The labels are obtained by applying k-mean clustering with k=50. For more details, please refer to the paper.
SemEval_Pairs.txt: a text file of word-pairs in SemEval-2012 Task2 test data.
Relational Patterns: is a folder for relational patterns of DiffVec training dataset. The folder includes two pickle files as follows:
1. Patterns_Xmid5Y: a dictionary that maps pattern ids to patterns 2. Patterns_Xmid5Y_PerPair: a dictionary that maps word-pairs to list of pattern ids.
Please follow the process below.
download GloVe 300-d from http://nlp.stanford.edu/data/glove.6B.zip, zip the file and put glove.6B.300d.zip into the main folder /Context-Guided-Relation-Embeddings
To learn CGRE for SemEval-2012 Task2 word-pairs, run the self-supervised learning of our CGRE model as follows:
$ cd Context_Guided_RelRep
$ python train.py
To learn NLRA for SemEval-2012 Task2 word-pairs,run the unsupervised NLRA model as follows:
$ cd NLRA
$ python train.py
If you use the code or the proposed method, please kindly cite the following paper:
Huda Hakami and Danushka Bollegala: Context-guided Self-Supervised Relation Embeddings Proc. of the 16th International Conference of the Pacific Association for Computational Linguistics (PACLING), October, 2019.