Information Retrieval

“Information retrieval is finding material of an unstructured nature that satisfies an information need from within large collections.” (IRBook)

• Such large collections include the world wide web, a personal hard drive, academic publications, etc.
• The user input, an information need is inside her head; she translates it to a query (in the form of text, a photograph, a drawing, or some other media); the system must then find the most relevant documents to return to her.
• The challenge of retrieval is to design and implement a relevance score that reflects a user’s judgments of relevance.

Although they are often used interchangably, there is a difference between retrieval and ranking (slides):

• Ranking: Given a test item, return an list of similar items from a background collection ordered according to similarity score. (See introduction.)
• Retrieval: Given a query that represents a user information need, return a list of relevant items ordered according to relevance score.

A typical information retrieval system is structured as follows:

(Image source: Scientific and Acedemic Publishing)

The retrieval functions (or document ranking) are responsible for finding documents in the document collection that most closely match our query. It can be implemented using several techniques, such as the vector space model or the unigram language model, depending on the application.

Recall that we can also objectively evaluate information retrieval systems.

[ Home ]