A deep learning baseline for the classification of Chinese word semantic relations

Abstract

The classification of Chinese word semantic relations is a significant research topic in the field of natural language processing. Compared with studies which identify the relation of word-pairs in given texts, the task of context-free lexical relational classification is more challenging due to the lack of context. A common way of solving this problem is to use word embeddings and lexical features to train a classifier. In this paper, we design various combinations of deep learning models and features and propose a joint model based on convolutional neural network and highway network. The joint model has reached a f1 value of 0.58 and outperform all the other deep learning models now available. Furthermore, we design extensive experiments to analyze how the magnitude of the training data influences the model’s performance and whether the distribution of data influences model’s performance.