Towards Lexical Chains for Knowledge-Graph-basedWord Embeddings

Abstract

Word vectors with varying dimensionalities and produced by different algorithms have been extensively used in NLP. The corpora that the algorithms are trained on can contain either natural language text (e.g. Wikipedia or newswire articles) or artificially-generated pseudo corpora due to natural data sparseness.We exploit Lexical Chain based templates over Knowledge Graph for generating pseudo-corpora with controlled linguistic value. These corpora are then used for learning word embeddings. A number of experiments have been conducted over the following test sets: WordSim353 Similarity, WordSim353 Relatedness and SimLex-999.The results show that, on the one hand, the incorporation of many-relation lexical chains improves results, but on the other hand, unrestricted-length chains remain difficult to handle with respect to their huge quantity.

Trends in Media Coverage and Information Diffusion Over Time: The Case of the American Earth Systems Research Centre Biosphere 2

“…Concerning the automatic task of identifying the stakeholders, this one is based on graphical forms, but does not take into account pronominal anaphors and, more generally, lexical chains. Presently, this remains a challenge for NLP researchers, but recent advances based on deep learning and word embedding (e.g., Simov et al, 2017) can provide the means to make the identification of stakeholders and the computing of the SC score more accurate.…”

“…this remains a challenge for NLP researchers, but recent advances based on deep learning and word embedding (e.g., Simov et al, 2017) can provide the means to make the identification of stakeholders and the computing of the SC score more accurate.…”

Enhanced word embeddings using multi-semantic representation through lexical chains

“…Even though the applicability of lexical chains is diverse, we see little work exploring them with recent advances in NLP, more specifically with word embeddings. In (Simov et al, 2017), lexical chains are built using specific patterns found on WordNet Fellbaum (1998) and used for learning word embeddings. Their resulting vectors, as ours, are tested in the document similarity task.…”

“…We expect the proposed algorithms to produce a robust semantic representation through the use of lexical chains. Furthermore, we build our lexical chains using several synset objects 4 in the lexical database in addition to hypernyms and hyponyms, which are commonly used in the literature (Gonzales et al, 2017;Mascarell, 2017;Simov et al, 2017) (detailed in Section 3). The main idea is to bring the semantic relations of lexical chains to traditional word embeddings techniques, leveraging their vector representation, and improving the overall result in the document classification task.…”