Better living through Hybrid-Search

  Let's back up for a moment and look at our problem from one level higher. What we would like to accomplish in general terms is a system that assists users remember and organize. We are not interested in having the user think in the constrained grammar rules we set for our system. Rather we would like our system to act within the rules and models a human would apply to finding information.

The tool we would like to provide to users is one that is flexible and robust enough to handle the user's internal methods of recall. This means that we would like to provide users with a set of tools that allows a diverse set of queries in which the user may express the same set of hints that occur in their own minds. We call this tool hybrid-search.

This thesis was proposed within the context of the Haystack project[23], as a means to extend the functionality of this adaptive personal information repository system. Details of Haystack's core implementation, which have made this thesis possible, will be discussed in great details in chapters  and .

  
Figure: The Hybrid-Search Query Model

The Haystack architecture provides the means to rapidly integrate a variety of information systems into a user's functional Haystack. From the query perspective, it possible through Haystack to build a ``multiplexer'' that selects information systems that directly match the users needs, and a method of combining the results. From the indexing side, we can build tools to centrally maintain a data repository. This functionality will allow for the hybrid-search mechanism we are aiming for. Figure  reflects the goal of breaking up a query into different components for processing by the optimal information system.

This revised model splits the information system into three parts: I₁ + F₁ (the IR system), I₂ + F₂ (the database), and I₃ + F₃ (the hypertext system). Additionally, the model introduces two new functions:

• A Multiplexing function, M(q^f)®q^f₁,q^f₂,q^f₃ which routes different parts of the original query to the information system(s) best suited to deal with it.
• A combination function, S(r₂₁,r₂₂,r₂₁)®r₂ which merges the results of the different filter functions into one result set.

If it is not entirely clear yet what type of query a user could issue to such a system it may be helpful, in conclusion, to provide some examples of hybrid-search type queries (which independent information systems would not be able to answer):

• Which documents do I have about four legged animals that I wrote last year? The first part of the query, ``about four legged animals,'' is an unstructured query (we are assuming that we didn't pre-categorize the documents in the collection). The second part, which limits the set of documents to a certain time period, is a structuerd (database) query.
• Which email did I send to David in regard to his request for clarification on Lore? This is very similar to the first query. The first part limits us to emails that David sent, again a structured query. The second part, which may require some semantic understanding on the part of the information system, is answerable by an unstructured (information retrieval) system.
• Which web pages did I visit after looking at the MIT homepage last week? The first part of the query, ``web pages ...after ...MIT homepage,'' is an associative search. The second part of the query constrains the results to ``last week,'' and is a database type query.