Collaborative Ranking with a Push at the Top

Proceedings of the 22nd International Conference on Intelligent User Interfaces

et al. 2017

Feature-based matrix factorization techniques such as Factorization Machines (FM) have been proven to achieve impressive accuracy for the rating prediction task. However, most common recommendation scenarios are formulated as a top-N item ranking problem with implicit feedback (e.g., clicks, purchases) rather than explicit ratings. To address this problem, with both implicit feedback and feature information, we propose a feature-based collaborative boosting recommender called BoostFM, which integrates boosting into factorization models during the process of item ranking. Specifically, BoostFM is an adaptive boosting framework that linearly combines multiple homogeneous component recommenders, which are repeatedly constructed on the basis of the individual FM model by a re-weighting scheme. Two ways are proposed to efficiently train the component recommenders from the perspectives of both pairwise and listwise Learning-to-Rank (L2R). The properties of our proposed method are empirically studied on three real-world datasets. The experimental results show that BoostFM outperforms a number of state-of-the-art approaches for top-N recommendation.

Section: Experimental Setup Datasets Evaluation Metrics and Baselinesmentioning

confidence: 99%

BoostFM

Yuan¹,

Proceedings of the 22nd International Conference on Intelligent User Interfaces

et al. 2017

“…In order to speed up the experiments, we follow the common practice as in [5] by randomly sampling a subset of users from the Libimseti and Yahoo datasets, and a subset of items from the Lastfm dataset. Table 1 summarizes the statistics of the three datasets used in this work.…”

Section: Settingsmentioning

confidence: 99%

“…Since the frequency of a user listening to a song (i.e., relevance feedback) can be obtained from the Lastfm dataset, we also recommend songs by leveraging such information (denoted as FMF 11 ). Note that the frequency information has a large range compared with ratings (e.g., [1,5] interval). For example, a user may listen to a song in hundreds of times.…”

Section: Settingsmentioning

confidence: 99%

Optimizing Factorization Machines for Top-N Context-Aware Recommendations

Yuan

Lecture Notes in Computer Science

et al. 2016

Abstract. Context-aware Collaborative Filtering (CF) techniques such as Factorization Machines (FM) have been proven to yield high precision for rating prediction. However, the goal of recommender systems is often referred to as a top-N item recommendation task, and item ranking is a better formulation for the recommendation problem. In this paper, we present two collaborative rankers, namely, Ranking Factorization Machines (RankingFM) and Lambda Factorization Machines (LambdaFM), which optimize the FM model for the item recommendation task. Specifically, instead of fitting the preference of individual items, we first propose a RankingFM algorithm that applies the cross-entropy loss function to the FM model to estimate the pairwise preference between individual item pairs. Second, by considering the ranking bias in the item recommendation task, we design two effective lambda-motivated learning schemes for RankingFM to optimize desired ranking metrics, referred to as LambdaFM. The two models we propose can work with any types of context, and are capable of estimating latent interactions between the context features under sparsity. Experimental results show its superiority over several state-of-the-art methods on three public CF datasets in terms of two standard ranking metrics.

“…Interestingly, we observe that it is possible to compute the approximate rank of positive item i by sampling one incorrectlyranked item. More specifically, given a u, i pair, we repeatedly draw an item from I until we obtain an incorrectlyranked item j such thatŷ(x i )−ŷ(x j ) ≤ ε and P u ij =1, where ε is a positive margin value 5 . Let T denote the size of sampling trials before obtaining such an item.…”

mentioning

confidence: 99%

“…In this algorithm, we iterate through all positive items i ∈ Iu for each user and update the model parameters w, V until the procedure converges. In each iteration, given a user-item pair, the sampling process is first performed so as to estimate violating 5 We hypothesize that item j is ranked higher than i for user u only ifŷ(x i ) −ŷ(x j ) ≤ ε, the default value of ε in this paper is set to 1.…”

mentioning

confidence: 99%

LambdaFM

Yuan

Proceedings of the 25th ACM International on Conference on Information and Knowledge Management

et al. 2016

State-of-the-art item recommendation algorithms, which apply Factorization Machines (FM) as a scoring function and pairwise ranking loss as a trainer (PRFM for short), have been recently investigated for the implicit feedback based context-aware recommendation problem (IFCAR). However, good recommenders particularly emphasize on the accuracy near the top of the ranked list, and typical pairwise loss functions might not match well with such a requirement. In this paper, we demonstrate, both theoretically and empirically, PRFM models usually lead to non-optimal item recommendation results due to such a mismatch. Inspired by the success of LambdaRank, we introduce Lambda Factorization Machines (LambdaFM), which is particularly intended for optimizing ranking performance for IFCAR. We also point out that the original lambda function suffers from the issue of expensive computational complexity in such settings due to a large amount of unobserved feedback. Hence, instead of directly adopting the original lambda strategy, we create three effective lambda surrogates by conducting a theoretical analysis for lambda from the top-N optimization perspective. Further, we prove that the proposed lambda surrogates are generic and applicable to a large set of pairwise ranking loss functions. Experimental results demonstrate LambdaFM significantly outperforms state-of-the-art algorithms on three real-world datasets in terms of four standard ranking measures.