interactive querying

Full reference:

Alan Dix (1998).
Interactive Querying - locating and discovering information
Second Workshop on Information Retrieval and Human Computer Interaction, Glasgow, 11th September 1998.
http://www.hcibook.com/alan/papers/IQ98/

Read the full paper (Adobe pdf)

See also:

• day conference on 'The Active Web' to be held in January 1999
• my research topics page on Query by Browsing and information visualisation
• my paper on the Active Web

The once distinct technologies of information retrieval, databases and hypertext are converging and overlapping. In addition, a plethora of interactive visualisation techniques have been developed over recent years. This paper will examine the nature of interactive querying and retrieval in order to understand the common features between apparently diverse retrieval techniques and in so doing help to address the needs of the emerging hybrid information repositories.

Convergence of technologies

Traditionally the information retrieval community and database community worked on very different kinds of data and using very different retrieval techniques. IR focused on free text documents with keyword and similarity-based retrieval. Databases focused on structured data with precise formal queries whether expressed in a program-like (e.g. SQL) or tabular (e.g. query by example) fashion. Hypertext retrieval stands opposed to both these approaches with its network structure and directed browsing for retrieval.

However, these barriers have been dissolving, for example, commercial databases now include large text fields with free-text search. The web although starting out as a hypertext is increasingly borrowing traditional IR techniques for search engines as well as pioneering new technologies such as crawlers and recommender systems. In addition, the chaos of the web has lead to a demand for greater structure and semantics. This has been partly satisfied by 'META' tags, but will be extended radically as XML becomes widely used, which will allow database-like queries over published XML document types.

Learning from each other

If we are to develop mechanisms for effective query and retrieval from the emerging hybrid information storage systems, then we must understand the similarities in the semantic models of databases, IR and hypertext. Furthermore, we need to understand how these fit into the human process of interactive retrieval.

Personal context

Some years ago I developed an intelligent database querying system called Query-by-Browsing (QbB). Although aimed at traditional databases, its focus was on the selection and relevance ranking of specific records - far more similar to IR techniques. However, the differences between the two domains were important. In particular, for database retrieval it is important that the retrieved records are not just a useful or suitable set, but they are precisely the right set. This makes it important to be able to feedback not just the selected records, but also the query generated by the system to retrieve them.

QbB is in the process of being redeveloped in order to include different machine learning algorithms, both to improve the interactive style and to extend the kinds of data managed by it. A sound understanding of the interactive retrieval process is thus essential.

Understanding the problem

We need to look at the interactive retrieval process at two levels.

First we need to look at the outside picture:

• what kind of data are we dealing with (e.g. numeric, text, hierarchical)
• what kind of result do we want (in terms of its structure)
• how are our goals formulated (precise, unknown, exploratory)

• who starts it (a blank screen for a user query, or an initial display)?
• what sort of display does the system produce (lists of records, summary/aggregate data, graphical views)?
• what sort of feedback does the user give to the system (revised query, relevance feedback on records, parameter values)?

The interactive querying process

Read the full paper (Adobe pdf)