Talk:Recommender system/Archive 1

Recommender vs. recommendation
I've seen more people say "recommender system" than "recommendation system". Substantiated by a decade of research from GroupLens research. See 24 papers on http://grouplens.org/publications.html. dfrankow 22:38, 31 October 2006 (UTC)

Requested move
Recommendation system → Recommender system — It is a more common name, especially in research literature dfrankow 22:38, 31 October 2006 (UTC)

Example systems
There is an extensive list of recommenders on the Collaborative Filtering page. However, recommender systems are a superset of collaborative filtering systems. Should that list be instead on this page, and include more systems? Opinions? dfrankow 23:02, 31 October 2006 (UTC)


 * There were no objections, so I moved the page. — Mets501 (talk) 03:27, 11 November 2006 (UTC)

Requested merge
Recommender system → Information filtering — Both pages are still short. As recommender systems are a subset of IF systems, it would seem to me as more logical to merge them under IF (systems). IKiddo 21:30, 3 April 2007 (UTC)

Oops, wrong edit summary on my reversion
I reverted an undiscussed external link to a specific bookseller by new anonymous user 64.201.248.10, and I left a msg on his talk page inviting him to discuss the link and read WP:EL. - Dan Dank55 (talk) 04:43, 2 March 2008 (UTC)

Gilligan
I removed this from the article: Recommender systems are also sometimes known colloquially as "Gilligans". I wasn't able to find any reference to this on Google, and it just sounds fishy to me. Can someone provide a citation? --ESP (talk) 20:33, 7 November 2008 (UTC)


 * I've never heard of that. dfrankow (talk) 16:39, 30 April 2009 (UTC)

Why are all examples from collaborative filtering?
All non-collaborative filtering approaches have been thrown out of the examples. Why? 217.72.91.237 (talk) 10:48, 10 November 2008 (UTC)

Removed this paper ref
Remove paper reference from Quatse on 19 July 2007. It's just one paper out of hundreds written on recommender systems, why single out this one? It looks like it was just added around conference time, maybe by someone who knew the author or the paper?


 * More recently, a successful recommender system has been introduced for bricks and mortar superstores based upon statistical inference (see Quatse, Jesse and Najmi, Amir (2007) "Empirical Bayesian Targeting," Proceedings, WORLDCOMP'07, World Congress in Computer Science, Computer Engineering, and Applied Computing.) as opposed to the Collaborative Filtering techniques of eCommerce. Redemption rates, or "hit rates," are much higher averaging as much as 45% in chain grocery stores.

I agree with 217.72.91.237 that there should be more examples that are not collaborative filtering. I also think there should potentially be more articles about different algorithms (e.g., all the machine learning algorithms from the Netflix challenge!). I just don't think this particular paper is notable enough to be the single paper cited. Compare to the Adomavicius and Herlocker papers, which are both large survey papers covering many different aspects of recommender systems.

Disclosure: I know Herlocker, and both papers were from the University of Minnesota, where I went to school. However, GroupLens at Minnesota has done a lot of work in recommender systems (and Adomavicius is not GroupLens).

dfrankow (talk) 16:39, 30 April 2009 (UTC)

I don't think it's appropriate to call all recommendation systems "information filtering". Collaborative filtering recommendation works that way, but in many cases recommender systems work according to primitive information retrieval methods. That is, an item (book, movie, song) is used as a starting point or a primitive query, and the recommender system selects nearest neighbors according to some manner of social or content metadata. Amazon's recommended item system works like this. Sometimes filtering is applied beforehand (to remove popularity biases, etc.), sometimes it is applied afterwards (to enable personalization), but sometimes it's not really applied at all.

There's also the ACM RecSys conference, and then there's the NetFlix Recommender System prize, so it would seem RecSys is much more active these days as well. --Jjdonald (talk) 02:04, 5 March 2010 (UTC)

We added two other types of algorithms in recommender systems
Me and my friend are positive about the comment above so we added two other types of recommendation algorithms -Content-Based and Hybrid- to the page. Falakmasir (talk) 15:35, 10 November 2011 (UTC)

Preparing a major revision
As part of a seminar class on the Mathematics of Information, we have been tasked with making large-scale improvements to certain related Wikipedia pages. I have been preparing a major revision of this article.

Some of the changes include:
 * Revised introduction and larger overview to better present the core of this topic to a mainstream audience.
 * Reorganization to improve clarity and presentation of topics
 * Separation of system design approaches (content-based, collaborative filtering, hybrid) from algorithms (k-NN, etc)
 * Additional information on privacy concerns.
 * New section on mobile recommender systems.
 * Additional references.
 * Removal of extraneous external links.

As this is a fairly major revision, I have prepared a draft in my userspace:

If you are concerned with any revisions I have made, please let me know and I will further revise this draft before merging it into the main article. Greg4cr (talk) 16:51, 17 November 2011 (UTC)

Change References and Content in Approach Section
Missing references of Collaborative Filtering in this section, the previously cited reference [11] is kind of confusing. If you click the link, it will not give you important document. I changed into "Empirical analysis of predictive algorithms for collaborative filtering (ref)".

This section only mentioned the "user-based CF", while missing item-based CF. It is better not to mention both of them because they are discussed in Wikipedia article Collaborative filtering. -- Shuguanghan (talk) 02:36, 19 October 2012 (UTC)

Adding more content
I have heard several new terminologies in recommender system, such as social recommender system (ref), Trust-aware Recommender Systems (ref). However, I haven't seen any mentions in this article about this.

Also, I guess it misses one important section - the evaluation for Recommender Systems ( ref ). It is important to add the evalaution metrics, such as Accuracy, Precision, Recall, Novelty, Serendipity -- Shuguanghan (talk) 02:36, 19 October 2012 (UTC)

Suggestions for Algorithm Part
I have some confusions about the Algorithm section and Approach section, what are the difference between Algorithm and Approach ? Will the "content-based filtering" also have three different algorithms, i.e. KNN, Pearson ? Algorithms parts seem to be loosely connect with the recommender system. I think we need to remove the algorithm part. -- Shuguanghan (talk) 02:36, 19 October 2012 (UTC)

Case Based Recommendation
I suggest add another approach-Cased Based Recommendation or at least in the content of Content-Based Recommendation since they are kind of similar, but the information used are represented as case and the system recommends the cases that are most similar to a user's preference        (ref) ––XidaoW (talk) 16:33, 6 December 2012 (UTC)

Improvement of Approaches Section
In this recommender system article,I think the Approaches section might need improvement. While 2.1 Collaborative filtering is relatively complete, 2.2 and 2.3 clearly could illustrate more detailed information about each system. Greer728 (talk) 20:41, 14 October 2013 (UTC)


 * I believe the Hybrid Recommenders System could be combination of other different recommender approaches not only Content-Based(CN) and Collaborative(CF), for example Knowledge-based(KB) and Demographic(DM), I could have some description of other approaches, and add some figures related, but elaborate content of other approaches might needed as well.Leilatsai (talk) 15:50, 15 October 2013 (UTC)


 * About the Content-based filtering part, I found missing references is the biggest problem. I'll try to fix this problem by adding citations to it.Greer728 (talk) 19:49, 15 October 2013 (UTC)
 * To further improve the quality, I might also be able to add algorithms of Content-based filtering, as well as some descriptive images.Greer728 (talk) 19:49, 15 October 2013 (UTC)
 * Do you think whether it's necessary to add case-based recommendation under content-based recommendation? As I recall, case-based is another form of content-based recommendation that is well suited to many product recommendation domains where individual products are described in terms of a well deﬁned set of features.Leilatsai (talk) 23:44, 15 October 2013 (UTC)
 * I am not familiar with case-based recommendation so I did not realize this problem. I will check about this.If it is, I will add it under the content-based filtering.Greer728 (talk) 00:12, 16 October 2013 (UTC)
 * The Content-based filtering part also has some bad links in line, I will remove them.Greer728 (talk) 21:31, 16 October 2013 (UTC)


 * Since the strategies for hybrid recommendation are quite complicated, I may simply mention their types but not explain them in details.Leilatsai (talk) 22:49, 15 October 2013 (UTC)
 * I will add some in-line citations as well since there are few quotations, statement or definition needing cited.Leilatsai (talk) 03:56, 17 October 2013 (UTC)

Definition clarification
The lead-in sentence currently describes recommender systems as seeking to 'predict the "rating" or "preference" that a user would give to an item'. That seems overly specific and not in line with the more generic description in the link referenced from that statement, specifically that the seek to offer 'suggestions for items to be of use to a user.' As an example of the distinction of "rating" versus "items of use", consider what would happen if Amazon always displayed my three favourite novels at the bottom of every single page I visit. Yes, I have a really high preference for these (they are my favorites after all), but there is little utility in showing me the exact same items over and over again (and that I have likely already bought). — Preceding unsigned comment added by 141.166.159.0 (talk) 19:04, 28 October 2016 (UTC)

Major Reno April 2019
I've gone through the article and done a major cleanup. In particular, I've removed or merged redundant content, cleaned up promotional content, and updated some out of date info. I also added a cool graphic.

It would be very helpful if someone could review the approaches section and update the technical information as necessary.

I've also left in the footnotes in the lede, even though they technically shouldn't be there. I suspect the ones in the last sentence are entirely promotional. If anyone feels like cleaning up, feel free.

Finally, the External Links section might possibly just all be promotional. Thoughts? --Seazzy (talk) 17:19, 25 April 2019 (UTC)

Health Recommender Systems
Hello Wikipedians,

also healthcare is becoming (at least in the research domain) an increasingly popular application of recommender systems (health recommender systems, HRS), e.g. for therapy recommendation. You think that application should be added or a new page created (health recommender systems)?

Best regards. — Preceding unsigned comment added by 2.247.249.176 (talk) 05:18, 20 August 2021 (UTC)

Wiki Education Foundation-supported course assignment
This article was the subject of a Wiki Education Foundation-supported course assignment, between 30 October 2018 and 11 December 2018. Further details are available on the course page. Student editor(s): Rachelma114. Peer reviewers: Rachelma114.

Above undated message substituted from Template:Dashboard.wikiedu.org assignment by PrimeBOT (talk) 07:53, 17 January 2022 (UTC)