Semantic Similarity Entity Disambiguation for Short Text

In the use of recommendation system and retrieval system, Named Entity Disambiguation (NED) is helpful to distinguish polysemy and synonym, so as to realize accurate retrieval and recommendation. Although existing statistics-based research methods achieve high accuracy, these studies have low accuracy when faced with short texts without context. To solve this problem, this essay proposes a Named Entity Disambiguation method based on Knowledge Graph similarity. The method is a semantic similarity model based on entity property similarity. The semantic similarity model based on entity property relies on embedding model and Knowledge Graph to learn the vector representation of words and features in the same shared vector space. Experimental results show that the Named Entity Disambiguation method proposed in this essay is more effective than the existing text-only statistical method when context information is scarce. In the expansibility experiment, this essay will discuss the influence of model parameter values on the precision of disambiguation results.