You are on page 1of 4

IMACST: VOLUME 4 NUMBER 1 MAY 2013 15

Recommendation Diversity for Web


Personalization: A Survey
Chetan B. Patil and R. B. Wagh 1

to enter exact and specific query that represents his or her


Abstract— Today the tremendous information overload is a big interest. Otherwise IR may come with large set of generalized
concern in information retrieval field. Most of the time user get results and expected result may be displayed somewhere at the
confused and often get lost while surfing on the web. later pages of results which is more cumbersome. Instead of
Recommender systems that do the job of web personalization this what if users do not have to enter any query at all and
customize the web sites and recommend the web products by someone intelligently recommend customized information for
understanding the user needs. The recommender systems are
intelligent one as they do not explicitly ask to users about their user. Well web personalization is the area where all such care
interests every time. Currently recommender systems focus on is taken about users and their likes-dislikes.
accuracy of systems to improve the quality. But diversity of Web Personalization is a technique that customizes a
recommendation list is also an important factor that brings web website in accordance of user’s interests and likes-dislikes.
personalization in true sense. We explore here overview of
different techniques to improve diversity while maintaining
Web personalization does not need any explicit data instead it
adequate level of accuracy. We observed some are content-based collects web data in web context which can be structural,
and some are collaborative techniques. We insist researchers to content or user profile data. Web Personalization guides the
propose hybrid techniques that can take advantage of users to achieve better web experience. Web experience can be
contentbased techniques as well as collaborative techniques to simply browsing of a web site or it can be as important as
enhance the diversity. In addition one should also look for purchasing some products or downloading some items from
enhancing both individual as well as aggregate diversity
simultaneously. We believe comparative analysis among that web site. And this experience can be enhanced by
performances of such techniques can justify the quality rearranging the web site content or structure, highlighting web
enhancement. links, insertion of some runtime links or creation of new
windows or pages. Uses of proper web mining techniques that
Index Terms— information retrieval, accuracy, diversity are intelligent and computationally efficient are required.
Accurate utilization of such techniques will lead to web
personalization in true sense. In addition lots of problems
I. INTRODUCTION associated with web personalization also need to be addressed
simultaneously. They are cold-start problem, scalability,
I NFORMATION retrieval is a system that retrieves
information from web resources on behalf of user requests.
adapting to user context, managing dynamics in user interests,
robustness, information security.
User type something on search engine and the requested query
is resolved according to some relevance and ranking Recommender system is a simulation of Web
algorithms on server end to reply the requested information of Personalization. Recommender system is a special kind of
that user. The fact that user has to enter his query explicitly has personalized information retrieval system that retrieves or say
itself became serious problem with information retrieval recommends products/items in accordance of user’s interests
systems. This is because today tremendous information is and likes-dislikes but without explicit requests from users.
available on web out of which user may be unaware about They are also seen as internet-based software tools, designed
most of it. to help users find their way through today’s complex online
shops and entertainment websites. Recommender engine is an
User will request only for that information that he or she had important part for any such commercial website today. The
heard before this. Other existing information in which user implicit data obtained from intelligent mining techniques is the
may find interest never comes to the user screen because IR input for recommender engine and the algorithm implemented
returns only relevant information. But unfortunately IR can’t inside do the rating predictions and ranking of information to
understand user’s interests intelligently and hence lots of be recommended. One way to classify recommender systems
useful information remains behind the scene. Another crucial as content based type, collaborative type and hybrid type
point is that though we accept IR in its current form; user has [2][3][10]. Another way is heuristic based type and model
Chetan B. Patil, PG Student, Computer Engineering Department, R. C.
based type recommender systems.
Patel Institute of Technology, Shirpur, affiliated to North Maharashtra
University (NMU), Maharashtra, India. email: chetan.rcpit@gmail.com R. B. Lots of different factors are important in deciding the
Wagh, Associate Professor, Computer Engineering Department, R. C. Patel quality of a recommender system. These factors are seen as
Institute of Technology, Shirpur, affiliated to North Maharashtra University
(NMU), Maharashtra, India. email: raj_wagh@rediffmail.com
more or

1857-7202/1305072
16 C.B. Patil and R. B. Wagh: Recommendation Diversity for Web Personalization: A Survey

less important depending on the application where you are cost and caused minimal loss of similarity among target query
going to use the recommender system. As stated earlier and recommendations.
recommender system is simulation of web personalization so C. Ziegler et. al. [3] proposed topic diversification, a
improvements in problems of web personalization will new heuristic approach to optimize the balance between
improve the quality of recommender system also. Some of accuracy and diversity so as to keep accuracy in a certain level
factors are accuracy, diversity, scalability, reliability, when increasing diversity, specifically for recommendation
serendipity etc. lists obtained as a result of some item based collaborative
Accuracy is the property of recommender system that filtering algorithm. Topic diversification resembles to Osmotic
decides whether generated recommendations are accurately Pressure analogy where selective permeability is the key
identified to user’s interests and likes-dislikes or not. If user is criteria for optimization. Taxonomies are created for various
receiving what he or she is expecting then the accuracy of your domains, arranged in a hierarchical way. Each product belongs
recommender system is high. to one or more taxonomies and they also have content
Diversity is the property of recommender system that forces descriptions relating to these domain taxonomies. The authors
to recommend various different items as possible. Many it so also propose intra-list similarity, a new metric which is well
happens that user get bored if received recommendations are suited to capture the diversity using proposed algorithm.
all of same type and in this situation if he or she found items According to authors effective use of content descriptions
that stands out differently among other items he or she may go along with relevance weights of products has effective impact
for that outstanding member. Basic reason behind this is the while ranking items and that is where the proposed method
human mind set which is easily attracted to odd figure in differs from other existing ones. Their experimental results
search of something new than regular routine. shown that users preferred the altered diversified list even
some loss of accuracy occurred, than the accurate unaltered
Diversity can be inherent or perceived [12]. Inherent
list.
diversity decides the recommendations based on dissimilarity
among items to be recommended. It can be further classified as D. Fleder et. al. [4] showed how basic design choices
individual or aggregate diversity. Individual diversity means a_ect the outcome, and thus managers can choose
dissimilarity among items recommended to a single individual recommender designs that are more consistent with their sales
user. Aggregate diversity means dissimilarity among items goals and consumers' preferences. They found that
being recommended across all users under considerations. recommenders can increase sales, and recommenders that
Perceived diversity decides the recommendations based on discount popularity appropriately may increase sales more.
actual experience by users regarding the differentness among M. Zhang et. al. [5] proposed a approach that seeks to find
items being recommended. It can be again classified as current out best possible subset of items to be recommended over all
perceived diversity or temporal perceived diversity. Current possible subsets. Here resultant list’s similarity to target query
perceived diversity considers experience of a user at a point of and diversity within list these two are taken as a binary
time. Temporal perceived diversity considers the user optimization problem. A new evaluation metrics, item novelty,
experience over a period of time. is proposed. Item novelty means how much an item is different
The survey encompasses a thorough description of both than existing items list. Item novelty depends upon other
content based and collaborative techniques for representing existing items in user profile. Item novelty brings certain level
items and user profiles. Then we discussed about various of difficulty for recommendations and hence can be used to
efforts by researchers in individual as well as aggregate generate useful test cases. By adjusting the novelty value the
diversity. Finally, it discusses trends and future research which tolerance in accuracy loss is balanced. Author points out that
might lead towards the next generation of recommender probability of recommending novel items is low whenever
systems. similarity is the basic selection criterion.
P. Castells et. al. [6] proposed that though various novelty
II. LITERATURE SURVEY and diversity metrics are popular in literature of recommender
systems, they do not address two important characteristics item
A. Individual diversity ranking and relevance. Author concentrated on two ground
concepts, namely item similarity and user-item interaction.
B. Smyth et. al. [1] proposed some ad-hoc strategies to
User-item interaction is modeled based on three conditions
rank items for inclusion in recommendation list. According
choice, discovery and relevance. They tried to cover and
to author maximum similarity in target query and cases to be
generalize the old metrics and put in better format.
retrieved is the general strategy in many domains but it
doesn’t work in some domains. S. Vergas et. al. [7] proposed that intent oriented information
retrieval diversity can be applied improving Recommendation
K. Bradley et. al. [2] proposed three new algorithms for
diversity. They formalized the diversity and novelty metrics
improving individual diversity. According to author diversity
and their results showed that resulting diversification
problem is always been limitation for content based
techniques can give best results. In addition these proposed
recommendation techniques and the proposed algorithms have
metrics are well aware of ranking and relevance issues pointed
formed a benchmark on this concern. Out of these Bounded
out in [6].
Greedy Selection algorithm has greatly reduced the retrieval
IMACST: VOLUME 4 NUMBER 1 MAY 2013 17

A. Said et. al. [8] proposed an approach that is completely III. CONCLUSION AND FUTURE SCOPE
orthogonal to the standard k nearest neighbor (kNN)
The web personalization systems are really useful today as
algorithms followed in traditional recommender systems. The they have ability to understand the user's needs explicitly as
variant they proposed is k furthest neighbors (kFN) algorithm well as implicitly. Though it has several issues of concerns still
that uses an augmented inverted similarity measure. The least well balanced systems have already captured the market.
liked items of neighborhoods are recommended. Their results Quality improvement can be achieved not only by accuracy but
on standard datasets have shown that diversity is improved also by diversity gains. This will bring personalization in true
with insignificant loss in recommendation accuracy. sense. Though the concept is very clear actual balance
A. Aggregate Diversity maintained among accuracy and diversity is challenging. Use
of proper re-ranking technique and proper measurement metric
T. Zhou et. al. [9] developed an approach that combines a will improve the system. The division into rating prediction
accuracy focused algorithm and a diversity focused algorithm. and re-ranking has brought flexibility in implementation of
According to author such collaborations can yield best results various state of the art techniques.
balancing both accuracy and diversity, without relying on any Such quality systems which balance not only the accuracy
semantic or context specific information. They used averaging but also the diversity can be made optimized using any well-
process in their algorithm that supports diversity known optimization algorithm. Genetic algorithm or Particle
enhancements. swarm optimizations are well known optimization techniques
G. Adomavicius et. al. [10] designed some ranking based that can significantly enhance the efficiency of the system.
techniques that can improve aggregate diversity of Instead of increasing diversity of individual items we can
recommendation lists across all users. They conducted online think of the diversity improvement of the some sequence of
experiments on MoviLense, Netix and YahooMovies datasets, items or say bundle of items also. Now as multiple items are
each operated by different rating prediction techniques in involved in a bundle so we have to think about the aggregate
combination with the proposed seven different ranking based characteristics of all items or services. Proposed system can be
extended in this direction also.
techniques. For item based popularity ranking approach they
REFERENCES
formed a parameterized function through which the level of
accuracy and diversity to be maintained is controlled. They [1] B. Smyth and P. Maclave,”Similarity Vs. Diversity”, 4th International
proposed precision-in-top-N, a metric to measure the accuracy Conference on case-based reasoning, 2001. pp. 348-361.
of the recommendation list and diversity-in-top-N, a metric to [2] K. Bradely and B. Smyth, “Improving recommendation diversity,"
measure the aggregate diversity of recommendation list. The Proceeding 12th Irish Conference Artificial Intelligence and Cognitive
Science,2001.
analysis shows that popularity of items can be a good tool to
[3] C.-N. Ziegler, S. McNee, J. Konstan, and G. Lausen, ”Improving
enhance the diversity and hence lead to user satisfaction.
recommendation list through topic diversification,” Proceeding 14th
C. Other Related Work International WWW conference, 2005. pp 22-32.
[4] D. Fleder and K. Hosanagar,”Blockbuster culture's next rise or fall: The
K. Alodhaibi et. al. [11] built a recommender algorithm that impact of recommender system on sales diversity,” Proceeding of 8th
works for compound products and services instead of just ACM conference, 2007. pp 192-199
individual items. For this they implemented CARD framework [5] M. Zhang and N. Hurley,” Avoiding monotony: Improving the diversity
which basically separates the utility space and diversity space of recommendation list”, 2008.
to avoid the tradeoff between similarity and diversity. The [6] P. Castells, S. Vargas, and J. Wang,” Novelty and diversity metrics for
algorithm he designed is computationally efficient and recommender system: Choice, discovery and relevance,” Proceeding of
outperforms in terms of diversity. International workshop on diversity in document retrieval (DDR), 2011.
pp. 29-37.
M. Ge et. al. [12] proposed an approach where position of
items in recommendation list has given most importance. [7] S. Vargas, “New approaches to diversity and novelty in recommender
systems," the 4th BCS IRSG Symposium on Future Directions in
Authors said that though diversity weather individual or
Information Access, 2012.
aggregate is achieved, it is not necessary that user will perceive
[8] A. Said, B. Kille, B. Jain, and S. Albayrak, 2012.,”Increasing diversity
its advantages. A well diversifies list may display the similar
through furthes neighbor-based recommendation,".
items at the top of the list and diverse items may be at bottom
[9] T. Zhou, Z. Kuscsik, J.-G. Liu, M. Medo, J. Wakeling, and Y.-C. Zhang,
which is not desirable always. Especially it is highly important “Solving the apparent diversity-accuracy dilemma of recommender
in case of small screen devices where user is interested in only systems," Proceeding of National Academy Sciences of the United States
top few recommendations. of America, 2010. pp. 4511-4515.
B. Wang et. al. [13] said that user's interests are always full [10] G. Adomavicius and Y. Kwon,”Improving aggregate recommendation
of uncertainty which could not be addressed by top-N list of diversity using ranking-based techniques," IEEE Transactions On
recommendations easily. Instead of this author proposed a Knowledge And Data Engineering, 2012.
cloud model which is powerful at solving knowledge
uncertainties.
18 C.B. Patil and R. B. Wagh: Recommendation Diversity for Web Personalization: A Survey

[11] K. Alodhaibi, A. Brodsky, and G. Mihaila, “A randomized algorithm for


maximizing the diversity of recommendations," Proceedings of the 44th
Hawaii International Conference on System Sciences, 2011.
[12] M. Ge, F. Gedikli, and D. Jannach,”Placing high-diversity items in topn
recommendation lists," Proceedings of International Joint Conferences
on Artificial Intelligence, 2011.
[13] B. Wang, Z. Tao, and J. Hu, “Improving the diversity of user-based tonn
recommendation by cloud model,” 2010.

You might also like