“…Karakaya and Tevfik [15] introduced a modification of Koren et al [11]'s MF model for explicit feedback by penalizing popular items to improve diversity. The method has not been extended to implicit datasets.…”
Clicking data, which exists in abundance and contains objective user preference information, is widely used to produce personalized recommendations in web-based applications. Current popular recommendation algorithms, typically based on matrix factorizations, often have high accuracy and achieve good clickthrough rates. However, diversity of the recommended items, which can greatly enhance user experiences, is often overlooked. Moreover, most algorithms do not produce interpretable uncertainty quantifications of the recommendations. In this work, we propose the Bayesian Mallows for Clicking Data (BMCD) method, which augments clicking data into compatible full ranking vectors by enforcing all the clicked items to be top-ranked. User preferences are learned using a Mallows ranking model. Bayesian inference leads to interpretable uncertainties of each individual recommendation, and we also propose a method to make personalized recommendations based on such uncertainties. With a simulation study and a real life data example, we demonstrate that compared to state-of-the-art matrix factorization, BMCD makes personalized recommendations with similar accuracy, while achieving much higher level of diversity, and producing interpretable and actionable uncertainty estimation.
“…Karakaya and Tevfik [15] introduced a modification of Koren et al [11]'s MF model for explicit feedback by penalizing popular items to improve diversity. The method has not been extended to implicit datasets.…”
Clicking data, which exists in abundance and contains objective user preference information, is widely used to produce personalized recommendations in web-based applications. Current popular recommendation algorithms, typically based on matrix factorizations, often have high accuracy and achieve good clickthrough rates. However, diversity of the recommended items, which can greatly enhance user experiences, is often overlooked. Moreover, most algorithms do not produce interpretable uncertainty quantifications of the recommendations. In this work, we propose the Bayesian Mallows for Clicking Data (BMCD) method, which augments clicking data into compatible full ranking vectors by enforcing all the clicked items to be top-ranked. User preferences are learned using a Mallows ranking model. Bayesian inference leads to interpretable uncertainties of each individual recommendation, and we also propose a method to make personalized recommendations based on such uncertainties. With a simulation study and a real life data example, we demonstrate that compared to state-of-the-art matrix factorization, BMCD makes personalized recommendations with similar accuracy, while achieving much higher level of diversity, and producing interpretable and actionable uncertainty estimation.
“…Diversity_in_top_N) [19,20] or the distribution of recommended products among all recommendation lists (e.g. Gini_diversity) [14,15]. Diversity_ in_top_N favors recommending more products, but it does not consider the distribution of recommended products.…”
Section: Accuracy and Diversity Measuresmentioning
confidence: 99%
“…Increasing AD, therefore having a higher possibility to recommend more products in the long tail, has great potential for gaining higher profits since products in the long tail are extremely abundant. From this point of view, AD is significant to ecommerce business [14,15].…”
Section: Introductionmentioning
confidence: 99%
“…The increase of either diversity comes with a decrease in recommendation accuracy [16]. In addition, increasing ID does not necessarily lead to a significant improvement in AD, and vice versa [8,15,17]. For example, high ID can be obtained by recommending the same diverse set of products to all customers, but AD still remains low.…”
The effectiveness of product recommendations is previously assessed based on recommendation accuracy. Recently, individual diversity and aggregate diversity of product recommendations have been recognized as important dimensions in evaluating the recommendation effectiveness. However, the gain of either diversity is usually at the cost of accuracy and the increase of one diversity does not guarantee a significant improvement in the other. A few attempts have been made to achieve reasonable trade-offs either between recommendation accuracy and individual diversity or between recommendation accuracy and aggregate diversity. Little attention has been paid to obtain a balance among the three important aspects of product recommendations. To address this problem, we propose an adjustable re-ranking approach that incorporates two new ranking criteria for improving both diversities. Three ranking lists are generated to guarantee recommendation accuracy, individual diversity, and aggregate diversity, respectively. The three ranking lists are finally merged with tunable parameters to generate a recommendation list. To evaluate the proposed method, experiments are conducted on a data set obtained from Alibaba. The results show that the proposed method achieves much higher improvements in both diversities than the baseline methods when sacrificing the same amount of recommendation accuracy.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.