Dissertation > Excellent graduate degree dissertation topics show

Encyclopedic knowledge combined with statistical methods query intent classification

Author: HuGang
Tutor: LiuBingQuan
School: Harbin Institute of Technology
Course: Computer Science and Technology
Keywords: Intent classification Encyclopedic knowledge Explicit semantic analysis Logistic regression
CLC: TP391.1
Type: Master's thesis
Year: 2011
Downloads: 39
Quote: 0
Read: Download Dissertation

Abstract

With the resources and services on the Internet increasing, people often need the help of a search engine to find relevant information. General search engine returns results contain a large amount of impurities information, users often need to meet their own query intent from screening results. Although vertical search engine to return precise answers to a particular area, but when the user has multiple query intention, it is necessary to submit to multiple vertical search engines to get a more comprehensive search results. If the general search engine can accurately classify the user 's intention, and thus the targeted integration of one or more vertical search engine results, and to show a different way, it is possible to improve the satisfaction of the user's search. The traditional classification methods are usually based on intent statistical machine learning, if you want to get good results you need a lot of human-annotated corpus. This paper introduces encyclopedic knowledge, without a lot of manual annotation in the case from the perspective of non-statistical and statistical two intentions to solve classification problems. The main contents include the following aspects: First, the paper analyzes the traditional classification algorithms intent faces several major problems, we propose a classification based on the intent of encyclopedic knowledge algorithms. And intent of the user query algorithm class are mapped to Wikipedia representation space, and in this representation space using non-statistical methods to classify query intent. Finally, with the traditional classification algorithms comparative experiments intended to illustrate the method's effectiveness and superiority. Second, this paper statistical classification method requires a lot of manual annotation data limitations, the use of large-scale seed each category entry intent to simulate real user queries, and in order to train statistical classifiers. Under the same label data size, with a real user queries trained classifier comparison shows that the method is effective. Third, the intention of this fusion of two different classification method has the advantage of a better performance combined intent classifier, and the same data set by comparison of experiments to illustrate the benefits of the fusion. In this paper, traditional search engines, based on the results of the first classification according to the intention to select the appropriate vertical search engines, and then the intent of the search results relevance scoring recommended to make search results more in line with the user's query intent.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Text Processing
© 2012 www.DissertationTopic.Net Mobile