Thank you, my score is fixed score from the properties of the page, but at first we need to adjust the score for a promising result. I have tried one way of manually re-ranking all the documents by the search results. But it needs to iterate all the retrieved results and fetch the re-ranking score (stored in the index) to sum the overall score. It is inefficient. How to improve that by a new approach?
Sorry for making you misunderstanding. Thanks Best Regards, Zhou Qi Dept. Computer Science & Engineering Shanghai Jiaotong University -----邮件原件----- 发件人: Chee Wu [mailto:[EMAIL PROTECTED] 发送时间: 2007年11月15日 14:23 收件人: java-user@lucene.apache.org 主题: Re: how to effeciently implement the stastical scores like pagerank? Not sure with what you want to do .. There are many factors can affect the rank of documents.Some factors should be fixed, always the same for different query words ,such as PageRank and the ratio between amount of the links and the full text length of the pages,and VSM should be a dynamic factor.. So how do you want to implement your own score mechanism, just import some new factors or a completely new way? If add new factor, then a fixed one or a dynamic one ? 2007/11/15, Zhou Qi <[EMAIL PROTECTED]>: > > Hi Guys, > > > > I made a problem in implement some extra scores besides the VSM > model. My works entails with re-ranking the returned documents from the > extra scores like page quality or page property ( good page or navigator > page). I have tried two approaches before: > > A. Setting the document boosting attribute in indexing, but it is > hard > to adjust these weight after indexing the documents. > > B. The second way is to retrieve all the documents and getting the > score first, and then I will need to iterate all the documents to fetch > some > store fields to get the property of the page to compute the re-ranking > score. From the experiments before, this way takes too much time and it > will > need to iterate all the documents to re-ranking. This implementation is > easy > to adjust the ranking algorithem but it is too slow. > > Can anyone give a better solution to do that? Thanks in advance. > > > > Best Regards, > > > > Zhou Qi > > Dept. Computer Science & Engineering > > Shanghai Jiaotong University > > > > > > --------------------------------------------------------------------- To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
