Research And Implementation Of Question Answering System Based On Food Ontology

Currently, food knowledge is mainly obtained from search engines which onlyreturn reverent pages. This method results in low knowledge obtain efficiency.Otherwise, search engines use massive web pages as data sources, but they cannotreturn comprehensive and professional answers by re-integrating food knowledge fromdifferent web pages. To solve these problem, an food ontology based QA system isbuilt.Since there is no large-scale food ontology, this thesis proposes a method tosemi-automatically establish food ontology. The concept layer of food ontology is builtmanually by reorganizing the skeleton of the concerned websites. The instances arecrawled by the web crawler from relevant web pages under the guidance of the conceptrelations. The food ontology is fulfilled by integrating the concept layer and instances.The crux of an ontology based QAsystem is semantic parsing. This thesis proposea method by mapping utterance to logic form and then mapping logic form to SPARQLquery. A food domain dictionary is designed to map utterance to logical form,including food vocabularies and synonymous expressions. In addition, DCSisused as the bridge to channel utterance and SPARQL query. Logic forms is generatedby building a breadth-first tree. To chooses the optimal logical form a scoring model isadded.The food ontology based QA system is evaluated from three aspects: thecompleteness of food ontology, the accuracy of semantic parsing and the capacity ofthe system. The experimental results prove the system is efficiency and practical.