One of five Student Work Projects on the Indian-summer school on Linked Data (http://lod2.eu/Article/ISSLOD2011). We implemented a simple entity disambiguation approach and made the results visible on a Web UI.

Introduction

• Given:
• Reference knowledge base(s) K
• Text fragment T
• Set E of Named Entities
• Task:
• Find URI for each of the Named Entities

URI lockup

• Input: String I (label of an entity)
• Get all entities with rdfs:label l
• SELECT DISTINCT ?uri WHERE { ?uri rdfs:label “Paris”@en. }
• or each entity ?e, merge all labels
• SELECT ?label WHERE { ?e rdfs:label ?label. }

Disambiguation Approach

• Input: input text T, list of candidates C (Stanford NER http://nlp.stanford.edu/ner/index.shtml)

1. Remove stopwords in T
2. Stem each word in T
3. For each c in C
1. Remove stopwords in c
2. Stem each word in c
3. Calculate Jaccard coef. between c and T
4. Return ranked list of entities

Web UI