基于上下文特征的短文本实体链接研究

Context Feature Based Entity Linking for Short Text

本文构建了一个面向短文本的实体链接系统,致力于鉴别出文本中指向Freebase实体的文本片段。本文将实体链接分为两步,利用现有的实体指称识别方法鉴别出实体指称,随后利用三类特征进行实体消岐,包括:实体指称-实体相似度、实体-实体相似度、候选实体上下文指称相似度。通过考虑所有的实体指称-实体对,选择得分最高的作为实体链接结果。

This paper design an entity linking system for short text, aiming at identifying all text fragments referring to an entity contained in Freebase. The task is organized in two steps. We adopt existing mention detection method to detect enti- ty mentions and then disambignate among candidate entities leveraging three kinds of features, including entity mention-en- tity similarity, entity-entity similarity, candidate entity context mention similarity. By considering all possible mention-enti- ty pair combinations, we select the one with highest score as the linking result.