Entity Linking in Web Tables with Multiple Linked Knowledge Bases (JIST 2016)

This is my accepted long paper Entity Linking in Web Tables with Multiple Linked Knowledge Bases in proceedings of the 6th Joint International Semantic Technology Conference (JIST 2016). [PDF]

这是我的 JIST2016 论文，介绍了一种基于多知识库的实体链接新技术，并与 TabEL 和 LIEGE 做了性能对比。[PDF]

---

Fig: An Example of Constructed Entity Disambiguation Graph

Paper: [PDF]

Code: [JIST2016]

Corrections:

In section 3.1 (c), there are a few mistakes about the definition of $A_{ij}$.

Actually I find out that “Aij = Aji” does not apply to all circumstances.

Now I give the new definition of $A_{ij}$ as

$$A_{ij}=\left\{ \begin{array}{rcl} \frac{SR_{me}(i,j)}{SR_{me}(i,*)} && {if \ i \neq j,\ i\ is\ a\ mention\ node\ and\ j\ is\ an\ entity\ node}\\ (1-SR_{em}(i)) \times \frac{SR_{ee}(i,j)}{SR_{ee}(i,*)} && {if\ i \neq j,\ i\ and\ j\ are\ two\ entity\ nodes}\\ \frac{SR_{me}(j,i)}{SR_{me}(j,*)} && {if \ i \neq j,\ i\ is\ an\ entity\ node\ and\ j\ is\ a\ mention\ node}\\ 0 && {otherwise} \end{array} \right.$$

where $SR_{em}(i)​$ is the mention-entity semantic relatedness between an entity node i and its only adjacent mention node.

Besides, in Iterative Probability Propagation, damping factor = 0.5 seems to make it easier to reach the convergence.