Meta-search Engine (1)

8/8/2019 Meta-search Engine (1)

1/23

Evaluation of Meta-Search engine

Merge Algorithms

Presentation by


2/23

Concept

The core of Meta search engine is the result integration algorithms.

This presentation focuses on evaluating the integration algorithms

ofmeta-search engine, and compares meta-search to the general

search engine by experiments.

An experiment method to determine the priority of participant of

meta-search engine is also proposed.

The experiment result proved that the meta-search engine could get

quality searching result on average.


3/23

Existing System:

General search engines like Google, Yahoo, MSN etc.

Drawbacks:

As there are huge no. of documents on the world wide web, it is

very difficult to locate the information that is relevant to the users

interest.


4/23

Proposed System:

Meta Search engine merge algorithms.

Benefits:

The queries posted by users which are submitted to the metasearch engine.

It sends the query to multiple single search engines intern.

When retrieved items are returned by the underlying search engines

the meta-search engine further processes these items and presentsrelevant items to the user, based on a integration algorithm.


5/23

Explanation

It is very difficult to locate information that is relevant to the internet

users interest, as there are huge no. of documents on the world

wide web.

There are several approaches to help the users to find informationfrom the web.

One is meta-searchmethod.

Queries are submitted to the meta-search engine which sends the

query to multiple single search engines.


6/23

When retrieved items are returned by the underlined search

engines further processes these items and presents relevant itemsto the user based on a result integration algorithm.

JP callam proposed 4 typical synthesis algorithms to process

different situations.

Kirsch provided another typical method which required the

participant search engine to return some additional information.

The meta-search engine would use this information to re calculate

the relevance of document on the client side.

Meta crawler imported the concept credibility to determine the

relevance between documents and queries.


7/23

Profusions algorithm was one of the normalized score method.

Inquires adopted a merge strategy which also re calculated the

relevance on the client side.

I am evaluating and then compare in experiments 4 merge

algorithms through this project

1. Simple merge algorithm

2. abstract merge algorithm

3. Position merge algorithm

4. abstract / position merge algorithm


8/23

Simple merge algorithm

This approach is similar to the single search engine, and it just to

range the results from participate search engines intern by using a

multi wave merge method.

In practical user are interest in search results of the first three pagesand high probability.

The result after three pages could only enhance the completeness

of the search results but could not improve user experience

anymore.


9/23


10/23

Abstract merge algorithm:

The main idea of abstract merge algorithm is to rank search

results with the relevance between query and the abstract

information of search results.

First, we need extract the terms from query, and calculate the

relevance between terms and abstract.

Second, we calculate the relevant between query and each page.

Finally, the results are returned to users according to theirrelevance between query and abstract of one page.


11/23


12/23

Position merge algorithm:

1. Description of the algorithm

The basic idea of position merge algorithm is to make a full use of the

original position information from each single search engine.

For the same query, there are some pages which will occur in severalresult list of different participant search engines, but their position in

different result lists may not be the same.

To reconcile this contradiction , we should take the position in different

participants into account.


13/23

2. Determining the priority

In the position algorithm the priority of search engines need to be

determined in advance.

In this experiment, I used two main search engines Google and Yahoo asthe participants.


14/23


15/23

Abstract/ position merge algorithm:

The abstract and the position are very important information.

abstract/ position merge algorithm considered these two factors

synthetically to make the

integrated results to meet the user needs.

Following is the formulae to calculate the relavance bertween the

result and query.


16/23


17/23

Valuation and Analysis:

The evaluation of the results from different merge algorithms:In this section, i am going to compare the results of Simple merge

algorithm, abstract merge algorithm, Position merge algorithm and

abstract / position

merge algorithm, then estimates which result is better and morerelevant to the user query intent.

we use to main the measures accuracy, which is the ratio of the

results in the whole results. and they wait, namely the location of

the related results in the whole results.

Firstly, setting up 10 keywords which are more broadly, more

representative and more accordant with the users' queries,

such as


18/23

Beijing Olympic Games, Beijing University 2007 Bill gates,

Dell Computer, Heilongjiang University, Jackie Chan

Movie, the NBA Yaoming, Prison break introduction, The

DaVinci Code, The Great Wall, Thinking in JAVA.

Secondly, for each query , we take the first top 30 results for

three algorithms as the basis data set for analyzing. Every

result will be judged whether the

results are related to the keywords or not based on the content

of each result for each query by human.

If one result is relevant, we will mark the location of the result

which occur in the 30 result. and then calculate the weight of

the each relevant result with the weight function.


19/23

Compare meta-search engine with Google and Yahoo:

We established a meta search engine which take Google and yahoo

as the search engines.

We used the above 4 merge algorithms to fusion the results from the

participants.

We did a test, to evaluate whether the meta search engine could

improve the experience of the user.

We choose following 5 queries as the testing set:

"Bill Gates Beijing University 2007", "Heilongjiang University" , " The

great Wall", " Beijing Olympic Games" and "Dell Computer".


20/23

Firstly, we need fetch the top 30 results fromGoogle and

Yahoo respectively for every query.

Secondly, we use the same way to estimate the relevance of

these results respectively and mark the positions.

Then calculate wait for every result and accumulate the waits.and get the mean value. Then we get accuracy, (the no of

related results) / 30


21/23

The following are the evaluating results as shown in the tables

The weight and accuracy of Google and Yahoo

Key word

Google

Accuracy Weight

Yahoo

Accuracy Weight

Bill Gates Beijing

University 2007

0.6667 5.4207 0.5333 5.8663

Heilongjiang

University

0.3667 5.4580 0.3333 4.9942

The great Wall 0.6 5.3192 0.5667 5.5166

Beijing Olympic

Games

0.4333 4.6565 0.4667 5.0207

Dell Computer 0.5667 5.2430 0.5667 5.4057

Average 0.52668 5.1295 0.49334 5.3607


22/23

The comparison of the weight and accuracy between the three ranking

algorithms of Meta-Search engine and Google and Yahoo.

accuracy weight

Google 0.5467 5.2195

Yahoo 0.4933 5.3607

Summary 0.5649 5.3965

Position 0.5567 5.4083

Summary/position 0.5749 5.3966


23/23

Conclusion:

In this project we design an experiment to evaluate abstract

merge algorithm, position merge algorithm, abstract/position

merge algorithm ofmeta search

engine and compare the meta- search engine with the generalsearch engines such as Google and Yahoo.

From the experiment results it is concluded that the bettermerge

algorithm could improve the quality of searching.

Meta-search Engine (1)

Documents