Information Technology Reference
In-Depth Information
•
Collection Representation: In this step, collections are evaluated. The evaluation
results are used to determine the relevance of the collection with respect to the
search query. In uncooperative environments, collection representation is built by
sending random queries to get samples of each collection. This process is called
Query by Sampling (QBS) [
29
,
30
]. This sampled data is then used for estimating
relevance of the represented collection for an incoming query.
•
Collection Selection: In this step, relevant collections are selected and queried.
Prior works for collection selection can be categorized into three categories [
9
,
30
]: (1) large document-based collection selection [
6
,
36
], (2) small document-
based collection selection [
28
,
34
] and (3) classification-based collection selection
[
2
].
•
Result Merging: In this step, documents retrieved from different collections are
merged to a single result list. This includes score normalization, whichmakes these
documents comparable and thus rankable as a single result list [
6
,
22
,
32
,
33
].
In the context of the application of distributed information retrieval in an enterprise
environments the various aspects, such as heterogeneous document types, access
restriction, etc., characterized in the previous section should be considered. We
describe in the following sections how the characteristics of the enterprise envi-
ronment are handled in our distributed search system.
4.2.3 Multi-Agent System
In a multi-agent system, agent interact with each other to provide different function-
alities to the users [
35
]. Concerning the concept of agent-oriented software develop-
ment, Jennings et al. [
18
] provide an extensive description of how agent's paradigm
can be compared with other software engineering paradigms. One of the advantages
of using software agents is that we can model each agent to handle different unique
tasks, such as crawling, searching, and management tasks. An example of agent
concept usage in information retrieval field is shown in [
1
]. In this paper, agent tech-
nology is used to personalize search results based on the users' profile, i.e., contents
is filtered based on the user's information need. In our prototype systemwe use JIAC
Release V [
21
] as the framework for implementing our multi-agent-system back-
end. It provides robust and established functionalities for implementing distributed
agent-system communication.
Multiple works are available on the application of multi-agent systems in the
context of information retrieval [
19
,
27
,
38
,
40
]. A similar approach of using a multi-
agent-system for implementing enterprise search was introduced by Zhou et al. [
40
],
who, however, relies on ontologies to model user access. Our system is more flexible
since its user access is managed automatically by exploiting the existing access rights
saved in LDAP.
Search WWH ::
Custom Search