Triple Scoring Using Paragraph Vector - The Gailan Triple Scorer at WSDM Cup 2017

TL;DR

This paper presents a method for triple scoring in knowledge bases by using Paragraph Vector to measure textual similarity between subjects and their values, effectively ranking relevant triples.

Contribution

The paper introduces a novel application of Paragraph Vector for triple scoring, demonstrating its effectiveness in ranking triples based on textual similarity.

Findings

The approach achieves promising results in triple relevance scoring.

Paragraph Vector effectively captures semantic similarity in this context.

The method is suitable for ranking triples in knowledge base tasks.

Abstract

In this paper we describe our solution to the WSDM Cup 2017 Triple Scoring task. Our approach generates a relevance score based on the textual description of the triple's subject and value (Object). It measures how similar (related) the text description of the subject is to the text description of its values. The generated similarity score can then be used to rank the multiple values associated with this subject. We utilize the Paragraph Vector algorithm to represent the unstructured text into fixed length vectors. The fixed length representation is then employed to calculate the similarity (relevance) score between the subject and its multiple values. Our experimental results have shown that the suggested approach is promising and suitable to solve this problem.