Diversity in Recommendation Systems

draft.md

Improvising diversity of personalized recommendation systems

Recent Research papers:

• Improving Aggregate Recommendation Diversity Using Ranking-Based Techniques:

  107 Citations : IEEE Transactions on Knowledge and Data Engineering
  we introduce and explore a number of item ranking techniques that can generate substantially more diverse recommendations across all users while maintaining comparable levels of recommendation accuracy. Comprehensive empirical evaluation consistently shows the diversity gains of the proposed techniques using several real-world rating data sets and different rating prediction algorithms

• Recommendation Diversification Using Explanations: (Data Engineering, 2009. ICDE '09. IEEE 25th International Conference)

  Traditionally, the problem is addressed through attribute-based diversification grouping items in the result set that share many common attributes (e.g., genre for movies) and selecting only a limited number of items from each group. It is, however, not always applicable, especially for social content recommendations. For example, attributes may not be available as in the case of recommending URLs for users of del.icio.us. Explanation-based diversification provides a novel and complementary alternative|it leverages the reason for which a particular item is being recommended (i.e., explanation)for diversifying the results, without the need to access the attributes of the items. In this paper, we formally define the problem of explanation-based diversification and, without going into the details of the actual diversification process, demonstrate its effectiveness on a real world data set, Yahoo!~Movies.

• Improving the Diversity of Top-N Recommendation via Determinantal Point Process (LSRS ,Italy 2017. ACM):

  improve the diversity of top-N recommendation results based on the determinantal point process (DPP), which is an elegant model for characterizing the repulsion phenomenon.
  An Accelerated model with tunable parameter allowing users to smoothly control the level of diversity

• Diverse Personalization with Determinantal Point Process Eigenmixtures (NIPS 2013 Workshop):

  Determinantal point processes are probabilistic models for selecting subsets of items which can naturally encode a preference for both diversity and quality in a principled way. However, in a personalization context we would typically like to have more control over the recommendations than DPPs afford. To address this, we introduce several approaches for blending the properties of multiple DPPs, presented in increasing order of sophistication. Using a novel interpretation of DPPs, the final proposed approach, which we refer to as the DPP eigenmixture, exploits the eigenstructure of the DPP kernel matrices in order to encapsulate the most important properties of several DPPs in a single model.

• Post Processing Recommender Systems for Diversity (KDD Research paper,2017, Canada):

  we formulate recommendation system design as a subgraph selection problem from a candidate super-graph of potential recommendations where both diversity and rating quality are explicitly optimized: (1) On the modeling side, we de- fine a new flexible notion of diversity that allows a system designer to prescribe the number of recommendations each item should receive, and smoothly penalizes deviations from this distribution. (2) On the algorithmic side, we show that minimum-cost network flow methods yield fast algorithms in theory and practice for designing recommendation subgraphs that optimize this notion of diversity. (3) On the empirical side, we show the effectiveness of our new model and method to increase diversity while maintaining high rating quality in standard rating data sets from Netflix and MovieLens.

• Dealing with diversity and novelty in group recommendations using Hesitant fuzzy sets (2017 IEEE International Conference on Fuzzy Systems) :

  our proposal uses Hesitant Fuzzy Sets to model the group information. A case study is performed to evaluate the proposal, whose results show its performance regarding recommendation diversity, novelty and accuracy.

• A new Collaborative Filtering technique to improve recommendation diversity ( Computer and Communications (ICCC), 2016 2nd IEEE International Conference ):

  In this study, an enhanced collaborative filtering method is proposed, which combined both of the User-Based and Item-Based collaborative filtering to improve diversity. This model divides the two algorithms into different weights, and uses different diversity methods to generate the recommendation lists. Then recommender system provides each user the diversity recommendations under the guarantee of accuracy. Experiments on 100K Movielens have proved that our scheme is effective, which has little increment in accuracy and much more increment in individual diversity and aggregate diversity.

• Combining user-based and item-based collaborative filtering techniques to improve recommendation diversity

  Web Intelligence and Intelligent Agent Technology (WI-IAT), 2012 IEEE/WIC/ACM International Conferences
  The proposed approach provides a flexible solution that incorporates multidimensional clustering into a collaborative filtering recommendation model to provide a quality recommendation. This facilitates to obtain user clusters which have diverse preference from multi-view for improving effectiveness and diversity of recommendation. The presented algorithm works in three phases: data preprocessing and multidimensional clustering, choosing the appropriate clusters and recommending for the target user. The performance of proposed approach is evaluated using a public movie dataset and compared with two representative recommendation algorithms. The empirical results demonstrate that our proposed approach is likely to trade-off on increasing the diversity of recommendations while maintaining the accuracy of recommendations.

• Relative similarity based approach for improving aggregate recommendation diversity:

  India Conference (INDICON), 2015 Annual IEEE
  Recommender systems solve the problem of information overload, by helping to find the most suitable items from a large set. Evaluating recommender system and made recommendations are equally important in an efficient recommender system. Though quality assessment of recommender system can be done using various measures, accuracy is the most important one. Sometimes accuracy may lead to a lack of user satisfaction since the user may not always be interested in the trending items. Diversity, one of the important aspects of the recommender system, eliminates such problems. Diversity is all about distinct recommendations, which are to be suggested to the user. This article presents a new metric relative similarity index (RSI) to improve the aggregate diversity of a system at a minimal loss of accuracy using nearest neighbor (NN) based collaborative filtering. The proposed algorithm is verified using two datasets namely Jester and Movie Lens.

• Using latent features to measure the diversity of recommendation lists:

  Information and Communication Technology, Electronics and Microelectronics (MIPRO), 2015 38th International Convention
  One of the well-known issues with content recommender system is that they tend to become over-specialized, which often has a negative influence on user experience. This can be solved by diversification of the recommendation list, a process that implements a tradeoff between accuracy and diversity of recommended items. Normally, item metadata is used in the diversity measure. In certain cases however, the item metadata may not be available thus a different approach to measure diversity is required. The aim of this preliminary study is to determine whether latent features can be used to measure the diversity of recommended items. In order to resolve this we generated recommendation lists for 43 different users using the LDOS-CoMoDa dataset. We then evaluated the diversity of these lists using the standard intra-list diversity measure. In addition we calculated the diversity of each list by comparing the latent features (calculated using the matrix factorization approach) of each item on the list. The comparison of both value sets showed that they show similar characteristics which implies that latent feature space offers an alternative method of evaluating item diversity when no metadata is present.

• Enhancing Diversity-Accuracy Technique on User-Based Top-N Recommendation Algorithms:

  Computer Software and Applications Conference Workshops (COMPSACW), 2013 IEEE
  In this paper we demonstrate how each item in top-N recommendation list has an impact on total diversity of the list in recommender systems. We proposed a new recommendation ranking method, namely "Total Diversity Effect Ranking", based on the total diversity effect of each item. Typically, diversity and accuracy in recommender systems are conversely. The finding shows that the proposed ranking method can guarantee the improvement of diversity quality in top-N recommendations. We proposed two new hybrid ranking approaches to make them more balance and designed experiments to evaluate their performance in comparison with the standard ranking approaches. To make results of our experiments more reliable, we apply the 5-fold cross validation technique to the MovieLens dataset provided by the GroupLens Research Project. In an evaluation process, we suggest using harmonic mean to measure the quality of recommendation approaches. The findings show that the proposed approaches give better performance than the standard approaches.