Solving the apparent diversity-accuracy dilemma of recommender systems

By Tao Zhou, Z. Kuscsik, Jianguo Liu, Matús Medo, Joseph R. Wakeling, Yi-Cheng Zhang. 2008

970 citations recorded by the source corpus.

Recommender systems use data on past user preferences to predict possible future likes and interests. A key challenge is that while the most useful individual recommendations are to be found among diverse niche objects, the most reliably accurate results are obtained by methods that recommend objects based on user or object similarity. In this paper we introduce a new algorithm specifically to address the challenge of diversity and show how it can be used to resolve this apparent dilemma when combined in an elegant hybrid with an accuracy-focused algorithm. By tuning the hybrid appropriately we are able to obtain, without relying on any semantic or context-specific information, simultaneous gains in both accuracy and diversity of recommendations.

Subfields: Computer Vision and Pattern Recognition, Information Systems, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Information Systems

In Signal Processing

Citations in the corpus

How algorithmic confounding in recommendation systems increases homogeneity and decreases utility (2017, background): “Among these characteristics, diversity is often framed as a counterpart to accuracy [28, 41, 64].”
Modeling mutual feedback between users and recommender systems (2015, neutral): “In this figure, we further support the results presented in the main text by using a different well-known recommendation method which has a parameter to tune the algorithm’s bias towards low or high degree items—the hybrid method combining mass diffusion and heat conduction processes [2].”
A probabilistic model to resolve diversity–accuracy challenge of recommendation systems (2015, background): “A practical recommender system should not only have a good accuracy but also provide proper novelty and diversity of items for the users [1,39].”
Recommender Systems (2012, methodology): “…recommendation scores for user i are computed as ˜ h i α = N β = 1 W H + P αβ h i β where the initial resource values are set as Results shown in [39] show that this combination of two different algorithms allows us not merely to compromise between diversity and accuracy but to simultaneously…”
Machine learning approaches and databases for prediction of drug–target interaction: a survey paper (2020, methodology): “DT-Hybrid Domain tuned-hybrid An extended NBI technique that incorporates domain-based knowledge such as drug similarities and target similarities [209] (also look [195, 210, 210, 211, 211??”
Current challenges and visions in music recommender systems research (2017, background): “where popi is the popularity of item i measured as percentage of users who rated i , Lu is the recommendation list of the top N recommendations for user u [193,195].”
Diversity, Serendipity, Novelty, and Coverage (2016, methodology): “We observe that works addressing the long-tail recommendation problem often measure algorithm performance not only in terms of novelty but also in terms of interuser diversity [Zhou et al. 2010; Liu et al. 2012] and coverage [Shi 2013].”
Fair and Diverse DPP-based Data Summarization (2018, background): “1 Related Work DPP-based sampling has been deployed for many data summarization tasks including text and images [17], videos [12], documents [20], recommendation systems [31], and sensors [16]; and the study of DPPs with additional budget or resource constraints is of importance.”
Auralist: introducing serendipity into music recommendation (2012, methodology): “[27] T. Zhuo, Z. Kuscik, J.-G. Liu, M. Medo, J. R. Wakeling, and Y.-C. Zhang.”
Diversity and Inclusion Metrics in Subset Selection (2020, background): “Geometric or distance-based measures of diversity have also been explored within the sciences, measuring the diversity of a dataset by the dataset’s volume [1, 8, 11, 17, 20, 22, 37], variance as in PCA [30], or other measures of spread.”
Exposure diversity as a design principle for recommender systems (2018, background): “However, more recent studies (Zhang & Hurley, 2008; Zhou et al., 2010) have observed that this type of approach might lead to a situation where the recommendations offered are ‘ monothematic ’ and repetitive and therefore end up being boring to the user.”
Vital nodes identification in complex networks (2016, methodology): “Following the written relations, the score of each publication is distributed to all its co-authors, which follows the mass diffusion process [349, 350].”
Managing Popularity Bias in Recommender Systems with Personalized Re-ranking (2019, background): “ake personalization of long-tail promotion into account as well. And finally, there is substantial research in recommendation diversity, where the goal is to avoid recommending too many similar items [10, 23, 24]. Personalized diversity is also another related area of research where the amount of di”
Novelty and Diversity in Recommender Systems (2015, background): “Zhou et al [21] define novelty as the average self-information of recommended items, which amounts to the average log inverse ratio of users who like the item (also known as “inverse user frequency”).”
News recommender systems - Survey and roads ahead (2018, background): “Several ways of numerically quantifying the degree of novelty are possible, including alternative ways of considering popularity information [187] or the distance of a recommendable item to the user’s profile [137, 154].”