Contextual bandit algorithms are crucial in various decision-making applications, such as personalized content recommendation, online advertising, and e-commerce banner placement. Despite their successful applications in various domains, contextual bandit algorithms still face significant challenges with exploration efficiency compared to non-contextual bandit algorithms due to exploration in feature spaces. To overcome this issue, model selection policies such as MetaEXP and MetaCORRAL have been proposed to interactively explore base policies. In this paper, we introduce a novel counterfactual approach to address the model selection problem in contextual bandits. Unlike previous methods, our approach leverages unbiased Off-Policy Evaluation (OPE) to dynamically select base policies, making it more robust to model misspecification. We present two new algorithms, MetaEXP-OPE and MetaGreedy-OPE, which utilize OPE for model selection policy. We also provide theoretical analysis on regret bounds and evaluate the impact of different OPE estimators. We evaluated our model on synthetic data and a semi-synthetic simulator using a real-world dataset, and the results show that MetaEXP-OPE and MetaGreedy-OPE significantly outperform existing policies, including MetaEXP and MetaCORRAL.

In one-class recommendation systems, the goal is to learn a model from a small set of interacted users and items and then identify the positively-related (i.e., similar) user-item pairs among a large number of pairs with unknown interactions. Most loss functions in the literature rely on dissimilar pairs of users and items, which are selected from the ones with unknown interactions, to obtain better prediction performance. The main issue with this strategy is that it needs a large number of dissimilar pairs, which increases the training time significantly. In this paper, our goal is to only use the similar set to train the models and discard the dissimilar set. We highlight three trivial solutions that the recommendation system models converge to when they are trained only on similar pairs: collapsed and dimensional collapsed solutions. We propose a hinge pairwise loss and an orthogonality term that can be added to the objective functions in the literature to avoid these trivial solutions. We conduct experiments on various tasks on public and real-world datasets, which show that our approach using only similar pairs can be trained several times faster than the state-of-the-art methods while achieving competitive results.

Neural networks have become popular recently in recommendation systems to extract user and item representations. Most previous works follow a two-branch setting, where user and item networks learn user and item representations in the first and second branches, respectively. In the item cold-start problem, where the usage patterns of the items do not exist, the user network uses ID/interaction vector as the input and the item network uses the item side information (content) as the input. In this paper, we will show that by using this structure, two representations are learned for each item in the training set; one is the output of the item network and the other one is hidden inside the user network and is used for learning user representations. Learning two representations makes training slower and optimization more difficult. We propose to unify the two representations and only use the one generated by the item network. Also, we will show how attention mechanisms fit in our setting and how they can improve the quality of the representations. Our results on public and real-world datasets show that our approach converges faster, achieves higher recall in fewer iterations, and is more robust to the changes in the number of training samples compared to the previous works.

Autoencoder-based hybrid recommender systems have become popular recently because of their ability to learn user and item representations by reconstructing various information sources, including users' feedback on items (e.g., ratings) and side information of users and items (e.g., users' occupation and items' title). However, existing systems still use representations learned by matrix factorization (MF) to predict the rating, while using representations learned by neural networks as the regularizer. In this paper, we define the neural representation for prediction (NRP) framework and apply it to the autoencoder-based recommendation systems. We theoretically analyze how our objective function is related to the previous MF and autoencoder-based methods and explain what it means to use neural representations as the regularizer. We also apply the NRP framework to a direct neural network structure which predicts the ratings without reconstructing the user and item information. We conduct extensive experiments which confirm that neural representations are better for prediction than regularization and show that the NRP framework outperforms the state-of-the-art methods in the prediction task, with less training time and memory.

We introduce an approach to recommending short-lived dynamic packages for golf booking services. Two challenges are addressed in this work. The first is the short life of the items, which puts the system in a state of a permanent cold start. The second is the uninformative nature of the package attributes, which makes clustering or figuring latent packages challenging. Although such settings are fairly pervasive, they have not been studied in traditional recommendation research, and there is thus a call for original approaches for recommender systems. In this paper, we introduce a hybrid method that leverages user analysis and its relation to the packages, as well as package pricing and environmental analysis, and traditional collaborative filtering. The proposed approach achieved appreciable improvement in precision compared with baselines

Rakuten recipe is a recipe site where users can submit their recipes and share with the others. Since recipe contents are generated by users, they usually contain many misspellings, abbreviations, synonyms, hypernyms and hyponyms. Identifying and normalizing these words is essential to retrieve relevant recipes to user's request. In this paper, we introduce a new approach to finding related words in a recipe domain using the data structure. Based on the observation that people usually write the main ingredient in the first position of ingredient lists of each recipe and such a ingredient is strongly related to the categories where recipes belong, we calculate relation scores of word pairs using real service data, which contains 790 categories and 405,519 recipes. The experimental result showed that we successfully found semantically related word pairs with f-score of 0.93.