Large Language Models (LLMs) are revolutionizing conversational recommender systems (CRS) by effectively indexing item content, understanding complex conversational contexts, and generating relevant item titles. However, the autoregressive nature of LLMs, which outputs item titles as a long sequence of subtokens, hinders the ability to efficiently obtain and control recommendations across the entire item set. This challenge in calculating probabilities over all items limits LLMs' potential, such as (1) limiting control over recommendation popularities and (2) preventing the synergy of marrying LLMs and traditional recommender systems (RecSys).

Matrix factorization is a widely used approach for top-N recommendation and collaborative filtering. When implemented on implicit feedback data (such as clicks), a common heuristic is to upweight the observed interactions. This strategy has been shown to improve performance for certain algorithms. In this paper, we conduct a systematic study of various weighting schemes and matrix factorization algorithms. Somewhat surprisingly, we find that training with unweighted data can perform comparably to-and sometimes outperform-training with weighted data, especially for large models. This observation challenges the conventional wisdom. Nevertheless, we identify cases where weighting can be beneficial, particularly for models with lower capacity and specific regularization schemes. We also derive efficient algorithms for exactly minimizing several weighted objectives that were previously considered computationally intractable. Our work provides a comprehensive analysis of the interplay between weighting, regularization, and model capacity in matrix factorization for recommender systems.

Conversational recommender systems (CRS) aim to provide personalized recommendations via interactive dialogues with users. While large language models (LLMs) enhance CRS with their superior understanding of context-aware user preferences, they typically struggle to leverage behavioral data, which have proven to be important for classical collaborative filtering (CF)-based approaches. For this reason, we propose CRAG-Collaborative Retrieval Augmented Generation for LLM-based CRS. To the best of our knowledge, CRAG is the first approach that combines state-of-the-art LLMs with CF for conversational recommendations. Our experiments on two publicly available movie conversational recommendation datasets, i.e., a refined Reddit dataset (which we name Reddit-v2) as well as the Redial dataset, demonstrate the superior item coverage and recommendation performance of CRAG, compared to several CRS baselines. Moreover, we observe that the improvements are mainly due to better recommendation accuracy on recently released movies. The code and data are available at https://github.com/yaochenzhu/CRAG.

Conversational recommender systems (CRS) should understand users’ expressed interests, which are frequently semantically rich and knowledge-intensive. Prior works attempt to address this challenge by using external knowledge bases or parametric knowledge in large language models (LLMs). In this paper, we study a complementary solution, exploiting item knowledge in the training data. We hypothesize that many inference-time user requests can be answered by reusing popular crowd-written answers associated with similar training queries. Following this intuition, we define a class of neighborhood-based CRS that makes recommendations by identifying items commonly associated with similar training dialogue contexts. Experiments on Inspired, Redial, and Reddit-Movie benchmarks show our method outperforms state-of-the-art LLMs with 2 billion parameters, and offers on-par performance to 7 billion parameter models while using over 170 times less GPU memory. We also show neighborhood and model-based predictions can be combined to achieve further performance improvements1.

In this paper, we present empirical studies on conversational recommendation tasks using representative large language models in a zero-shot setting with three primary contributions. (1) Data: To gain insights into model behavior in "in-the-wild" conversational recommendation scenarios, we construct a new dataset of recommendation-related conversations by scraping a popular discussion website. This is the largest public real-world conversational recommendation dataset to date. (2) Evaluation: On the new dataset and two existing conversational recommendation datasets, we observe that even without fine-tuning, large language models can outperform existing fine-tuned conversational recommendation models. (3) Analysis: We propose various probing tasks to investigate the mechanisms behind the remarkable performance of large language models in conversational recommendation. We analyze both the large language models' behaviors and the characteristics of the datasets, providing a holistic understanding of the models' effectiveness, limitations and suggesting directions for the design of future conversational recommenders.

The recommendation-accuracy of collaborative filtering approaches is typically improved when taking into account higher-order interactions [5, 6, 9, 10, 11, 16, 18, 24, 25, 28, 31, 34, 36, 41, 42, 44]. While deep nonlinear models are theoretically able to learn higher-order interactions, their capabilities were, however, found to be quite limited in practice [5]. Moreover, the use of low-dimensional embeddings in deep networks may severely limit their expressiveness [8]. This motivated us in this paper to explore a simple extension of linear full-rank models that allow for higher-order interactions as additional explicit input-features. Interestingly, we observed that this model-class obtained by far the best ranking accuracies on the largest data set in our experiments, while it was still competitive with various state-of-the-art deep-learning models on the smaller data sets. Moreover, our approach can also be interpreted as a simple yet effective improvement of the (linear) HOSLIM [11] model: by simply removing the constraint that the learned higher-order interactions have to be non-negative, we observed that the accuracy-gains due to higher-order interactions more than doubled in our experiments. The reason for this large improvement was that large positive higher-order interactions (as used in HOSLIM [11]) are relatively infrequent compared to the number of large negative higher-order interactions in the three well-known data-sets used in our experiments. We further characterize the circumstances where the higher-order interactions provide the most significant improvements.

Autoencoders (AE) aim to reproduce the output from the input. They may hence tend to overfit towards learning the identity-function between the input and output, i.e., they may predict each feature in the output from itself in the input. This is not useful, however, when AEs are used for prediction tasks in the presence of noise in the data. It may seem intuitively evident that this kind of overfitting is prevented by training a denoising AE, as the dropped-out features have to be predicted from the other features. In this paper, we consider linear autoencoders, as they facilitate analytic solutions, and first show that denoising / dropout actually prevents the overfitting towards the identity-function only to the degree that it is penalized by the induced L2-norm regularization. In the main theorem of this paper, we show that the emphasized denoising AE is indeed capable of completely eliminating the overfitting towards the identity-function. Our derivations reveal several new insights, including the closed-form solution of the full-rank model, as well as a new (near-)orthogonality constraint in the low-rank model. While this constraint is conceptually very different from the regularizers recently proposed, their resulting effects on the learned embeddings are empirically similar. Our experiments on three well-known data-sets corroborate the various theoretical insights derived in this paper.

The Sparse Linear Method (SLIM) is a well-established approach for top-N recommendations. This article proposes several improvements that are enabled by the Alternating Directions Method of Multipliers (ADMM), a well-known optimization method with many application areas. First, we show that optimizing the original SLIM-objective by ADMM results in an approach where the training time is independent of the number of users in the training data, and hence trivially scales to large numbers of users. Second, the flexibility of ADMM allows us to switch on and off the various constraints and regularization terms in the original SLIM-objective, in order to empirically assess their contributions to ranking accuracy on given data. Third, we also propose two extensions to the original SLIM training-objective in order to improve recommendation accuracy further without increasing the computational cost. In our experiments on three well-known data-sets, we first compare to the original SLIM-implementation and find that not only ADMM reduces training time considerably, but also achieves an improvement in recommendation accuracy due to better optimization. We then compare to various state-of-the-art approaches and observe up to 25% improvement in recommendation accuracy in our experiments. Finally, we evaluate the importance of sparsity and the non-negativity constraint in the original SLIM-objective with sub-sampling experiments that simulate scenarios of cold-starting and large catalog sizes compared to relatively small user base, which often occur in practice.

In this paper, we model the dependencies among the items that are recommended to a user in a collaborative-filtering problem via a Gaussian Markov Random Field (MRF). We build upon Besag's auto-normal parameterization and pseudo-likelihood, which not only enables computationally efficient learning, but also connects the areas of MRFs and sparse inverse covariance estimation with autoencoders and neighborhood models, two successful approaches in collaborative filtering. We propose a novel approximation for learning sparse MRFs, where the trade-off between recommendation-accuracy and training-time can be controlled. At only a small fraction of the training-time compared to various baselines, including deep nonlinear models, the proposed approach achieved competitive ranking-accuracy on all three well-known data-sets used in our experiments, and notably a 20% gain in accuracy on the data-set with the largest number of items.

Combining simple elements from the literature, we define a linear model that is geared toward sparse data, in particular implicit feedback data for recommender systems. We show that its training objective has a closed-form solution, and discuss the resulting conceptual insights. Surprisingly, this simple model achieves better ranking accuracy than various state-of-the-art collaborative-filtering approaches, including deep non-linear models, on most of the publicly available data-sets used in our experiments.

When a user has watched, say, 70 romance movies and 30 action movies, then it is reasonable to expect the personalized list of recommended movies to be comprised of about 70% romance and 30% action movies as well. This important property is known as calibration, and recently received renewed attention in the context of fairness in machine learning. In the recommended list of items, calibration ensures that the various (past) areas of interest of a user are reflected with their corresponding proportions. Calibration is especially important in light of the fact that recommender systems optimized toward accuracy (e.g., ranking metrics) in the usual offline-setting can easily lead to recommendations where the lesser interests of a user get crowded out by the user's main interests-which we show empirically as well as in thought-experiments. This can be prevented by calibrated recommendations. To this end, we outline metrics for quantifying the degree of calibration, as well as a simple yet effective re-ranking algorithm for post-processing the output of recommender systems.

The ranking quality at the top of the list is crucial in many real-world applications of recommender systems. In this paper, we present a novel framework that allows for pointwise as well as listwise training with respect to various ranking metrics. This is based on a training objective function where we assume that, for given a user, the recommender system predicts scores for all items that follow approximately a Gaussian distribution. We motivate this assumption from the properties of implicit feedback data. As a model, we use matrix factorization and extend it by non-linear activation functions, as customary in the literature of artificial neural networks. In particular, we use non-linear activation functions derived from our Gaussian assumption. Our preliminary experimental results show that this approach is competitive with state-of-the-art methods with respect to optimizing the Area under the ROC curve, while it is particularly effective in optimizing the head of the ranked list.

Evaluation is a cardinal issue in recommender systems; as in any technical discipline, it highlights to a large extent the problems that need to be solved by the field and, hence, leads the way for algorithmic research and development in the community. Yet, in the field of recommender systems, there still exists considerable disparity in evaluation methods, metrics and experimental designs, as well as a significant mismatch between evaluation methods in the lab and what constitutes an effective recommendation for real users and businesses. Even after the relevant quality dimensions have been defined, a clear evaluation protocol should be specified in detail and agreed upon, allowing for the comparison of results and experiments conducted by different authors. This would enable any contribution to the same problem to be incremental and add up on top of previous work, rather than grow sideways. The REDD 2014 workshop seeks to provide an informal forum to tackle such issues and to move towards better understood and shared evaluation methodologies, allowing one to leverage the efforts and the workforce of the academic community towards meaningful and relevant directions in real-world developments.

The literature on recommender systems distinguishes typically between two broad categories of measuring recommendation accuracy: rating prediction, often quantified in terms of the root mean square error (RMSE), and ranking, measured in terms of metrics like precision and recall, among others. In this paper, we examine both approaches in detail, and find that the dominating difference lies instead in the training and test data considered: rating prediction is concerned with only the observed ratings, while ranking typically accounts for all items in the collection, whether the user has rated them or not. Furthermore, we show that predicting observed ratings, while popular in the literature, only solves a (small) part of the rating prediction task for any item in the collection, which is a common real-world problem. The reasons are selection bias in the data, combined with data sparsity. We show that the latter rating-prediction task involves the prediction task 'Who rated What' as a sub-problem, which can be cast as a classification or ranking problem. This suggests that solving the ranking problem is not only valuable by itself, but also for predicting the rating value of any item.

Online social network information promises to increase recommendation accuracy beyond the capabilities of purely rating/feedback-driven recommender systems (RS). As to better serve users' activities across different domains, many online social networks now support a new feature of "Friends Circles", which refines the domain-oblivious "Friends" concept. RS should also benefit from domain-specific "Trust Circles". Intuitively, a user may trust different subsets of friends regarding different domains. Unfortunately, in most existing multi-category rating datasets, a user's social connections from all categories are mixed together. This paper presents an effort to develop circle-based RS. We focus on inferring category-specific social trust circles from available rating data combined with social network data. We outline several variants of weighting friends within circles based on their inferred expertise levels. Through experiments on publicly available data, we demonstrate that the proposed circle-based recommendation models can better utilize user's social trust information, resulting in increased recommendation accuracy.

Recommendation accuracy can be improved by incorporating trust relationships derived from social networks. Most recent work on social network based recommendation is focused on minimizing the root mean square error (RMSE). Social network based top-k recommendation, which recommends to a user a small number of items at a time, is not well studied. In this paper, we conduct a comprehensive study on improving the accuracy of top-k recommendation using social networks. We first show that the existing social-trust enhanced Matrix Factorization (MF) models can be tailored for top-k recommendation by including observed and missing ratings in their training objective functions. We also propose a Nearest Neighbor (NN) based top-k recommendation method that combines users' neighborhoods in the trust network with their neighborhoods in the latent feature space. Experimental results on two publicly available datasets show that social networks can significantly improve the top-k hit ratio, especially for cold start users. Surprisingly, we also found that the technical approach for combining feedback data (e.g. ratings) with social network information that works best for minimizing RMSE works poorly for maximizing the hit ratio, and vice versa.

Various methods have been developed for indoor localization using WLAN signals. Algorithms that fingerprint the Received Signal Strength Indication (RSSI) of WiFi for different locations can achieve tracking accuracies of the order of a few meters. RSSI fingerprinting suffers though from two main limitations: first, as the signal environment changes, so does the fingerprint database, which requires regular updates; second, it has been reported that, in practice, certain devices record more complex (e.g bimodal) distributions of WiFi signals, precluding algorithms based on the mean RSSI. In this article, we propose a simple methodology that takes into account the full distribution for computing similarities among fingerprints using Kullback-Leibler divergence, and that performs localization through kernel regression. Our method provides a natural way of smoothing over time and trajectories. Moreover, we propose unsupervised KL-divergence-based recalibration of the training fingerprints. Finally, we apply our method to work with histograms of WiFi connections to access points, ignoring RSSI distributions, and thus removing the need for recalibration. We demonstrate that our results outperform nearest neighbors or Kalman and Particle Filters, achieving up to 1m accuracy in office environments. We also show that our method generalizes to non-Gaussian RSSI distributions.

Recommendations from the long tail of the popularity distribution of items are generally considered to be particularly valuable. On the other hand, recommendation accuracy tends to decrease towards the long tail. In this paper, we quantitatively examine this trade-off between item popularity and recommendation accuracy. To this end, we assume that there is a selection bias towards popular items in the available data. This allows us to define a new accuracy measure that can be gradually tuned towards the long tail. We show that, under this assumption, this measure has the desirable property of providing nearly unbiased estimates concerning recommendation accuracy. In turn, this also motivates a refinement for training collaborative-filtering approaches. In various experiments with real-world data, including a user study, empirical evidence suggests that only a small, if any, bias of the recommendations towards less popular items is appreciated by users.

Matrix factorization (MF) has evolved as one of the most accurate approaches to collaborative filtering. In this paper, we extend the probabilistic MF framework as to account for multiple observations for each matrix element. This significantly improves the accuracy of recommender systems in several areas: (1) aggregation of ratings concerning items organized hierarchically, (2) (partial) compensation for the selection bias in the observed data by using an appropriate prior with virtual data points, and (3) improved recommendations of TV shows. While our framework applies to explicit and implicit feedback data, we outline in detail the latter application in this paper: we present the first approach that takes into account also negative feedback when training on implicit feedback data. Moreover, we shed light on the implicit assumptions underlying the most successful approach to IP-TV (Internet Protocol Television) recommendations in [Hu et al. 2008]. In our experiments, we obtain significant improvements over the existing approach.

Users typically rate only a small fraction of all available items. We show that the absence of ratings carries useful information for improving the top-k hit rate concerning all items, a natural accuracy measure for recommendations. As to test recommender systems, we present two performance measures that can be estimated, under mild assumptions, without bias from data even when ratings are missing not at random (MNAR). As to achieve optimal test results, we present appropriate surrogate objective functions for efficient training on MNAR data. Their main property is to account for all ratings - whether observed or missing in the data. Concerning the top-k hit rate on test data, our experiments indicate dramatic improvements over even sophisticated methods that are optimized on observed ratings only.

In this paper, we show that classical survival analysis involving censored data can naturally be cast as a ranking problem. The concordance index (CI), which quantifies the quality of rankings, is the standard performance measure for model \emph{assessment} in survival analysis. In contrast, the standard approach to \emph{learning} the popular proportional hazard (PH) model is based on Cox's partial likelihood. In this paper we devise two bounds on CI--one of which emerges directly from the properties of PH models--and optimize them \emph{directly}. Our experimental results suggest that both methods perform about equally well, with our new approach giving slightly better results than the Cox's method. We also explain why a method designed to maximize the Cox's partial likelihood also ends up (approximately) maximizing the CI.

We propose a Bayesian undirected graphical model for co-training, or more generally for semi-supervised multi-view learning. This makes explicit the previously unstated assumptions of a large class of co-training type algorithms, and also clarifies the circumstances under which these assumptions fail. Building upon new insights from this model, we propose an improved method for co-training, which is a novel co-training kernel for Gaussian process classifiers. The resulting approach is convex and avoids local-maxima problems, unlike some previous multi-view learning methods. Furthermore, it can automatically estimate how much each view should be trusted, and thus accommodate noisy or unreliable views. Experiments on toy data and real world data sets illustrate the benefits of this approach.

The bootstrap has become a popular method for exploring model (structure) uncertainty. Our experiments with artificial and real- world data demonstrate that the graphs learned from bootstrap samples can be severely biased towards too complex graphical mod- els. Accounting for this bias is hence essential, e.g., when explor- ing model uncertainty. We find that this bias is intimately tied to (well-known) spurious dependences induced by the bootstrap. The leading-order bias-correction equals one half of Akaike’s penalty for model complexity. We demonstrate the effect of this simple bias-correction in our experiments. We also relate this bias to the bias of the plug-in estimator for entropy, as well as to the differ- ence between the expected test and training errors of a graphical model, which asymptotically equals Akaike’s penalty (rather than one half).

A common objective in learning a model from data is to recover its network structure, while the model parameters are of minor in(cid:173) terest. For example, we may wish to recover regulatory networks from high-throughput data sources. In this paper we examine how Bayesian regularization using a product of independent Dirichlet priors over the model parameters affects the learned model struc(cid:173) ture in a domain with discrete variables. We show that a small scale parameter - often interpreted as "equivalent sample size" or "prior strength" - leads to a strong regularization of the model structure (sparse graph) given a sufficiently large data set. In par(cid:173) ticular, the empty graph is obtained in the limit of a vanishing scale parameter. This is diametrically opposite to what one may expect in this limit, namely the complete graph from an (unregularized) maximum likelihood estimate. Since the prior affects the parame(cid:173) ters as expected, the scale parameter balances a trade-off between regularizing the parameters vs. the structure of the model. We demonstrate the benefits of optimizing this trade-off in the sense of predictive accuracy.