Although indexing makes it possible to access information from very large document collections, the 

conversion from a document text to a list of weighted keywords does result in a loss of information. 

Writing a document is an intentional process; a document is intended to convey a message. The 

translation to a list of keywords retains the essential building blocks of the message, the terms 

themselves, but the message(s) that the author intended cannot be accessed by the retrieval mechanism. 

The effect of this loss of information may be ameliorated or deteriorated by the use of controlled 

vocabularies   pre defined sets of indexing terms, [Ing92, Chap 3]. However, the fact remains that when 

we talk of representing the information content of documents we are only representing the components 

of the message, not the message itself. 

The reduction of the document text into a series of keywords also transforms the task of an IR system 

from retrieving information to retrieving objects that contain information. Some authors argue that 

objects such as documents cannot be held to contain information as such, rather information is a change 

in a cognitive, or internal, state brought about by exposure to the contents of these objects. The 

following early quote by Maron, [Mar64], illustrates this concern,  

"..information is not a stuff contained in books as marbles might be contained in 

a bag   even though we sometimes speak of it in that way. It is, rather a 

relationship. The impact of a given message on an individual is relative to what 

he already knows, and of course, the same message could convey different 

amounts of information to different receivers, depending on each one s internal 

model or map."  

The degradation of the document text, necessary for computation, and the subjectivity of relevance 

results in a layer of indirection between the user and the documents. The goal of the IR system is to 

bridge this gap between the user and potentially relevant material. Indexing techniques identify and 

highlight potentially good indicators of relevant material, and retrieval techniques use these indicators 

of relevance to select which documents to present to the user. How individual retrieval systems use 

these indicators to retrieve documents is the topic of the next section. 

2.2 Retrieval and feedback

Retrieval is the process of matching a representation of an information need, usually a user supplied 

query, to an indexed document representation. Queries will be indexed in the same way as a document 

and compared with a document index to determine if a document is likely to be relevant to a query. 

How the indexed query is compared with the indexed document differentiates the major retrieval 

models. In this section we shall briefly outline the four main models of retrieval: Boolean, vector space, 

probabilistic, and logical, and describe the basic approaches to RF in each of the models. 

2.2.1 Boolean model 

The first operational IR retrieval model was the Boolean model, based on Boolean logic. In this model 

queries are keywords combined, by the user, with the conjunctive (AND), disjunctive (OR) or negation 

(NOT) operators. This is an exact match model: the system only retrieves those documents that exactly 

match the user's query formula. For example, for the query `information AND retrieval AND system' 

the system will return all documents that contain the three words `information', `retrieval' and `system', 

whereas the query `information OR (retrieval AND system)  will return those documents that contain 

the word `information' and those documents that contain both `retrieval' and `system'.   

The Boolean model has been used in a large number of on line public access catalogue (OPAC) 

systems but has been shown to demonstrate a number of difficulties. Firstly, traditional Boolean 

systems do not use term weights and consequently return the complete set of documents that match the 

query as an unordered set. This means the users may have to add or remove terms, or generate more 

complex query expressions to reduce the set of retrieved documents to a manageable size. Willie and 

Bruza, [WB95], argue that the problems with interacting with Boolean systems are not only a matter of 

the formal query language but a conceptual problem: the Boolean model does not lend itself to 

supporting how users think about searching and their individual search techniques. A further problem 

with Boolean systems is that the order in which operators are applied may not be consistent across 

systems, resulting in the fact that different systems may retrieve different documents for the same query, 

[Borg96]. Nevertheless Boolean systems do remain popular with users, perhaps because of the explicit 

control that is offered by these systems to the user. Web search engines often allow Boolean style 

querying performed on an underlying best match model (see section 2.2.2). 

 5