Font Size: a A A

Q & Answers Sorting Method Based On Markov Logic Network Research

Posted on:2013-02-27Degree:MasterType:Thesis
Country:ChinaCandidate:Z P MengFull Text:PDF
GTID:2218330374465340Subject:Computer applications and technology
Abstract/Summary:PDF Full Text Request
Compared with the traditional search engines which rely on keyword matching, Question-Answering(Q&A) System with the characteristics such as convenient, fast and efficient is better to meet the user's retrieval needs. Ranking is an important problem to be sovled in Q&A system,the result of ranking directly determines the performance of the system. The main factors that affect the answer ranking are the features of the answer candidate text and the ranking method. Therefore,for the characteristics of the specific areas of Q&A,the paper made a study about the feature selection of the answer ranking and the learning to rank method with multiple features,the main work are as follows:(1) We put forward three types of features about questions and answers: context-sensitive features, density-based features and the online Knowledge Base features. The context-sensitive features include the lexical similarity between the questions and candidate answers and the deep degree of semantic similarity between the questions and candidate answers. These features can provide the information about the questions, related documents and the candidate answers, and can help to improve the ranking performance of the retrieval model.(2) We proposed a ranking model of entity answers based on MLNs.For the characteristics of factoid and list answers in domain Q&A system,we built a ranking model of domain entity answers based on MLNs.This method use predicate formulas to describe the relevant characteristics of the questions-candidate answers,the answers-knowledge base and merge these features into Markov Logic Network., adopted the learning algorithm of discriminant training and MC-SAT inference algorithm.The answer ranking method based on Markov logic network can give different weights according to the relevance of different characteristics,and the results show that the answer precision and recall rates has greatly improved compared with other methods.(3) Taking Yunnan tourism as the restricted domain, we designed and implemented the learning to rank domain Q&A system that based on the MLNs.
Keywords/Search Tags:Question-Answering, Learning to rank, Markov Logic work, restricteddomain, Multiple Features
PDF Full Text Request
Related items