NeedleSeek is a project for web semantic mining or ontology mining. It aims at automatically extracting and aggregating semantic knowledge from tera-scale data, with the goal of boosting web search and various NLP applications.

In this online research prototype, a web interface is provided for end users to search and browse the semantic knowledge-base we built.

NeedleSeek V1.0 focuses on semantic class construction and search. It receives a term or a phrase as a query and returns the semantic class(es) the term or phrase belongs to. Our system is able to distinguish different means of the same word/phrase and return multiple semantic classes for one term. For example, for query "apple", our system returns at least two semantic classes: fruits (pear, orange, etc) and companies (Microsoft, Sun, etc).

In the current version (V2.0), we support the following semantic relations:

• Peer relation (organizing coordinate terms as semantic classes)
• Hypernym/hyponym relation (i.e., the is-a relation, including the class/instance and class/subclass relations)
• Class/attribute relation (e.g., country-population, company-CEO)

Related Research Papers