HIR, HyperContext information indexing and retrieval model : a new model for information retrieval based on interpretations of information

Abstract

The World Wide Web (WWW) is a gigantic information resource that is growing daily. As more and more data are added to the WWW, it is becoming increasingly difficult to retrieve useful information. Many attempts have been made to develop tools which effectively locate information on a user's demand. Such tools are commonly known as search engines. Users of the WWW would most certainly have come in contact with some of these search engines and more often than not, were dissatisfied with the results. In HyperContext: Information Indexing and Retrieval Model (HIR) we stop and rethink why users are dissatisfied with the results returned by common search engines, despite the fact that these documents may actually contain the terms and phrases the user chose to express his/her information need. We concentrate on the idea that one possible reason for this dissatisfaction is that information is subjective. Due to this subjective nature, the same piece of information can have different meanings. We call each different meaning an interpretation of the information. In HIR we are concerned with the information that resides on the WWW in the form of Web documents. Our aim is to automatically identify and represent interpretations of web documents. This is achieved by exploiting the inherent structure of the WWW and the documents themselves - both of which are prominent sources of latent information that can be (and should be) put to use. Having done this, we use HIR to experiment with the usefulness of indexing these interpretations. The results indicate that this is beneficial to the retrieval process because documents are represented by each of their descriptions - this is a great shift from traditional information retrieval systems where each document is represented only once. Upon submission of a user's query, the index of interpretations is used to retrieve relevant interpretations which are mapped back to their original documents and presented to the user. HIR was evaluated by applying it to a subset of a chosen website. A set of users were recruited to submit a number of queries to HIR and give relevance judgements on the documents that were returned. The evaluation showed that indexing interpretations does increase the effectiveness of retrieval over an IR system that indexes the exact same document set without interpretations.