Activities

No upcoming events

Intelligent Systems Lab Project: SEMDB

Participants

• Christiana Ogbeide
• Husam Abuhabib
• Lando Meyer
• Mert Ince
• Saleem Raza

Supervisors

• Philipp Cimiano
• Christina Unger

Motivation

• Semantic Web connects knowledge which allows to reach precise results for search queries
• On the other hand, Querying is hard and very tricky to be handled by users
• Moreover, The implementation of Semantic Web has become more trendy for searching approaches

Application Scenario

The user asks a question in natural language. The system converts this question to a query to retrieve answers from a knowledge base.

Objectives

The project goals are

• to code a system which can convert natural language questions to queries
• to provide intelligent auto-completion suggestions
• to show meaningful and precise results to user
• to use a Semantic Web based approach in a question answering system

Description

The system takes as input natural language questions from users and converts them into SPARQL queries. While the user is typing a question, it gives intelligent suggestions, guiding the user in formulating his information need. The system then returns answers.

• The grammer extractor reads a QALD file and extracts question clusters according to the similarity of linguistic structures. Based on those, it builds a grammar consisting of templates of question/query pairs.
• A state machine is initialized based on the extracted grammar.
• As soon as the user is typing, the system sends queries to a local database to get possible auto-completion suggestions.
• These suggestions are shown in the input field.
• With the typing of a question mark, the system converts the question into a SPARQL query by instantiating the template identified by the grammar.
• The resulting SPARQL query is sent to the DBpedia endpoint to get answers.
• The results are presented on the webpage.

Results of the second term (summer term 2015)

• The system can recognise natural language quesions available in the QALD benchmark.
• The system translates those natural language questions to SPARQL queries.
• The question patterns and templates are generated dynamically by means of clustering techniques.
• The system is able to give suggestions in each step of typing the question.
• The system responds with precise and meaningful results.

The following video shows how Quassk works.

Discussion and Conclusion

Using a grammer extractor based on clustering techniques, we implemented a system that converts a natural language questions into a SPARQL query. To this end, it can suggest words to the user in an intelligent way, so that the typed question is ensured to lead to a well-structured and correct query that provides the intended answers using Semantic Web technology.

Outlook

Currently the system works on the English questions available in the QALD benchmark. More work is required to port the techniques to other languages like Spanish, German, French etc.