Research On Technologies Of Query Semantic Dependency Parsing

With the rapid development of the internet, information is generated and spread atunprecedented speed. At the same time, garbage information is growing in an exponen-tial way. How to find out really useful information from it becomes great challenge ofsearch engine. In the traditional search engines, user enters query, search engine returnsa very long URL list. It doesn’t know what the user asks, what the user looks for. Itonly finds out webpages containing the keywords based on keyword matching retrievalmethods, and then rank them through page rank algorithm. Users need to identify whichpage satisfies their needs from a long list of pages. Query semantic dependency parsing(QSDP) can improve the page ranking of traditional search engine, it can reach a deepsemantic understanding of query, thus lead more accurate understanding of user’s needs.On the other hand, compared with traditional search engines, semantic search re-cently attracts wide attention of both industry and academic circles. Semantic searchorganizes all information into a huge structured knowledge base, and directly returns theanswer when user searches, thus it saves time to identify information. QSDP can helpsemantic search engine understand user’s needs and return the most accurate answer re-trieved from knowledge base. In addition, QSDP also has a wide range of applications,including automatic question answering, intelligent personal assistant, information re-trieval, information extraction and so on.This paper presents both rule-based and statistics-based QSDP techniques, researchtopics include:(1) Similarities and diferences of semantic dependency parsing technology on queryand normal sentence. Compared to normal sentence, query has shorter length and loos-er structure, therefor, semantic dependency parsing technologies on query and normalsentence difer greatly.(2) Establishment of dependencies system of QSDP. Based on the characteristicsof query and the needs of applications, establish a suitable dependencies system. Thispaper has five types of semantic dependencies, respectively attribute, constraint, agent,patient and demand. Constraint is divided into six sub-categories, namely, time constraint,location constraint, number constraint, type constraint, question constraint and negativeconstraint. (3) For the definition of six special constraint dependencies is clear and simple,this paper proposes a rule-based approach, including rule definition, rule preparation andimplementation of rule-based system.(4) Take the QSDP problem as a classification problem, this paper proposes a statistical-based approach, including semantic resources mining and design and selection of classi-fication features.Both rule-based and statistical-based approaches are verified through experiments.