Talk:Relevance (information retrieval)

This article was nominated for deletion on 13 May 2008. The result of the discussion was keep.

The current version of this page seems to edge toward violation of the NPOV policy. Specifically, use of the term "relevancy" has a different scholarly lineage, and is not incorrect. It's just different. A second example is the insistence on a particular view of relevance - that two or more documents are never relevant to each other without a user's information need. Source needed - that is one view among many. Documents may be related in many ways. For example, two documents may be related to each other for policy reasons, whether or not specific users feel they have an information need.

I agree. The discussion of relevancy versus relevance seems irrelevant, ha ha, and should certainly not be at the top of the article. Ideally, there would be a much better summary of algorithms, discussion of relevance versus importance in search results, and so on. I might have a go at some stage... Sam Dutton 12:24, 16 November 2007 (UTC)[reply]

Binary or analog value

I have seen, in multiple IR documents, Relevance used as a binary value, where the relevance of a document to a certain query is either 1 or 0. However, all definitions of "relevance" I have seen, descbribes it as an analog value, a value which can have lots of values on a scale from "perfectly relevant" to "not relevant at all". Which is correct; A binary value or an analog value?

Another realated question: In searches with a search engine, you will often find documents which are of interest, but not an answer to your query. Are such documents to be regarded as relevant or not relevant.

Jpalme (talk)jpalme —Preceding undated comment added 12:16, 17 February 2010 (UTC).[reply]

Not quite the right definition of relevance

The section relating to Google should be removed, because it's not really relevant (heh) to the topic, but, more because it defines relevancy incorrectly. Relevance concerns how much a document matches a certain query. Google's PageRank algorithm is specifically a query independent metric. It is not a novel measure of relevancy. It is a measure of how authoritative the document is, irrespective of particular queries.

Srowen 06:29, 27 April 2007 (UTC)[reply]

Should fold into Information_retrieval page

It seems to me that this page should be folded into the Information_retrieval page, which, though not perfect, is much more thorough. And I agree that the jab at Google, is at best a violation of NPOV, and at worst misleading. It belongs in a broader context of query-independent measures of authority, such as the one started on the Link_analysis page.

Dtunkelang 19:55, 10 September 2007 (UTC)[reply]

It seems there is a consensus among the wiki-editors to keep this page rather than fold it into the Information_retrieval page. In that case, I believe it needs to be fleshed out, corrected, and made to conform to NPOV. A couple of useful starting points would be work by Gerald Salton award recipient Tefko Saracevic (link to publications) and Stefano Mizzaro (link to publications)

Dtunkelang (talk) 23:27, 25 May 2008 (UTC)[reply]

I've taken a swag at rewriting this page. I've removed the NPOV notice because the current page is an overhaul. Let's have a discussion if my changes are controversial.

Dtunkelang (talk) 14:27, 26 May 2008 (UTC)[reply]

I'd just like to point folks to the discussion taking place at The Noisy Channel on how to improve this entry.

Dtunkelang (talk) 03:12, 29 May 2008 (UTC)[reply]

I've just made another large change to the Relevance page, including removing the section on evaluation measures since these are included in the main Information Retrieval page and in the Precision and Recall page. I also re-arranged some of the sections to (hopefully) make a little more sense. --Jelsas (talk) 19:52, 29 May 2008 (UTC)[reply]

Additional Reading

Sperber & Wilson and Saracevic seem to be notable enough to qualify as additional reading on relevance. I'm inclined to add Ingwersen & Jarvelin's "The Turn" as another reference. But I'd like to avoid this section become a place for researchers to promote their own work, which is why I removed a recent addition. Anyone else want to chime in on what criteria we should use for this section?

Dtunkelang (talk) 22:32, 7 June 2008 (UTC)[reply]