Large Language Model (LLM) agents trained with reinforcement learning (RL) show great promise for solving complex, multi-step tasks. However, their performance is often crippled by ''Context Explosion'', where the accumulation of long text outputs overwhelms the model's context window and leads to reasoning failures. To address this, we introduce CoDA, a Context-Decoupled hierarchical Agent, a simple but effective reinforcement learning framework that decouples high-level planning from low-level execution. It employs a single, shared LLM backbone that learns to operate in two distinct, contextually isolated roles: a high-level Planner that decomposes tasks within a concise strategic context, and a low-level Executor that handles tool interactions in an ephemeral, isolated workspace. We train this unified agent end-to-end using PECO (Planner-Executor Co-Optimization), a reinforcement learning methodology that applies a trajectory-level reward to jointly optimize both roles, fostering seamless collaboration through context-dependent policy updates. Extensive experiments demonstrate that CoDA achieves significant performance improvements over state-of-the-art baselines on complex multi-hop question-answering benchmarks, and it exhibits strong robustness in long-context scenarios, maintaining stable performance while all other baselines suffer severe degradation, thus further validating the effectiveness of our hierarchical design in mitigating context overload. Our code is available at https://github.com/liuxuanzhang718/CoDA.

Traditional conversion rate prediction models suffer from suboptimal performance due to sharing the same network parameters for all instances, failing to capture heterogeneous underlying distributions across instances. Recent parameter personalized network based models address this by grouping instances and adjust parameters for each group. However, existing parameter personalization methods face challenges: (1) taking prior information features as grouping condition for model parameter personalization leads to suboptimal performance due to human's limited understanding of data distribution, or (2) using all the features for both parameter generation module and deep neural network (DNN) of conversion prediction tasks causes gradient conflicts during backpropagation. A better approach is to automatically select features as grouping condition based on data distribution through iterative learning. Therefore, we propose Automatic Feature Partitioning-Based Parameter Personalized Network (APPNet), which consists of two components: Automatic Feature Partitioning (AFP) and Parameter Personalized Network (PPNet). The AFP module automatically partitions all the features into two parts: one part for DNN of conversion prediction tasks, and the other part for PPNet module to generate weights to adjust DNN parameters of conversion prediction tasks. Specifically, we implemented two versions of AFP: feature-wise AFP and bit-wise AFP. The feature-wise AFP partitions features at the feature field granularity, while the bit-wise AFP partitions each bit of the feature embeddings. The PPNet module adjusts model parameters of conversion prediction task for each group of instances by applying element-wise multiplication to the DNN parameters of conversion tasks. Extensive offline experiments demonstrate APPNet outperforms previous parameter personalized models. Furthermore, online A/B testing in production system achieved a 1.09% improvement on conversion rate, validating its practical effectiveness.

Personalized product search provides significant benefits to e-commerce platforms by extracting more accurate user preferences from historical behaviors. Previous studies largely focused on the user factors when personalizing the search query, while ignoring the item perspective, which leads to the following two challenges that we summarize in this paper: First, previous approaches relying only on co-occurrence frequency tend to overestimate the conversion rates for popular items and underestimate those for long-tail items, resulting in inaccurate item similarities; Second, user purchasing propensity is highly heterogeneous according to the popularity of the target item: it is less correlated with the user's historical behavior for a popular item and more correlated for a long-tail item. To address these challenges, in this paper we propose NAM, a Normalization Attention Model, which optimizes ''when to personalize'' by utilizing Inverse Item Frequency (IIF) and employing a gating mechanism, as well as optimizes ''how to personalize'' by normalizing the attention mechanism from a global perspective. Through comprehensive experiments, we demonstrate that our proposed NAM model significantly outperforms state-of-the-art baseline models. Furthermore, we conducted an online A/B test at Fliggy, and obtained a significant improvement of 0.8% over the latest production system in conversion rate.

Static user and item features constitute important information to be taken into account in the recommendation process. However, as these features are usually sparse and of large-vocabulary, existing deep learning-based methods typically construct large tables of high-dimensional feature embeddings, which is inefficient in terms of memory storage and is computationally problematic. On the other hand, while product quantization-based methods have been proposed to compress latent embeddings, they usually come at the cost of compromising recommendation performance due to the restrictive expressive power, as feature correlations and user-item interactions are not properly captured in the compression process. To address these issues, we propose a novel Dual Contrastive Learning method to generate low-dimensional discrete static feature representations that significantly reduce memory storage and computational complexity, while simultaneously producing superior recommendation performance. Extensive offline experiments on three large-scale industrial datasets demonstrate that our proposed model significantly outperforms the selected baselines. In addition, we conducted an online A/B test at Alibaba and show that the proposed model significantly improves the average video streaming time, while reducing the size of the feature embedding table by 90% over the currently deployed system.

Existing cross domain recommender systems typically assume homogeneous user preferences across multiple domains to capture similarities of user-item interactions and to provide cross domain recommendations accordingly. Meanwhile, the heterogeneity of user behaviors is usually not well studied and captured during the recommendation process, where users might have vastly different interests in different domains. In addition, previous models focus primarily on recommendation tasks between domain pairs, and cannot be naturally extended to serve for multiple domain recommendation applications. To address these challenges, we propose to utilize the idea of adversarial learning to intelligently incorporate global user preferences and domain-specific user preferences for providing satisfying cross domain recommendations. In particular, our proposed Adversarial Cross Domain Recommendation (ACDR) model first obtains the latent representations of global user preferences from their explicit feature information, and then transforms them into domain-specific user embeddings, where we take into account user behaviors and their heterogeneous preferences among different domains. By doing so, we address the differences among user representations in the domain-specific latent space while also preserving global user preferences, as we effectively segment the distributions of domain-specific user embeddings in the shared latent space. The convergence of our proposed model is theoretically guaranteed. The proposed ACDR model leads to significant and consistent improvements in cross domain recommendation performance over the state-of-the-art baseline models, which we demonstrate through extensive experiments on three real-world datasets. In addition, we show that the improvements are greater on those datasets that are smaller and more sparse, on those users that have fewer interaction records in the dataset, and when user interactions from more product domains are included in the cross domain recommendation model.

Cross domain recommender systems have been increasingly valuable for helping consumers identify useful items in different applications. However, existing cross-domain models typically require large number of overlap users, which can be difficult to obtain in some applications. In addition, they did not consider the duality structure of cross-domain recommendation tasks, thus failing to take into account bidirectional latent relations between users and items and achieve optimal recommendation performance. To address these issues, in this paper we propose a novel cross-domain recommendation model based on dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Furthermore, we combine the dual learning method with the metric learning approach, which allows us to significantly reduce the required common user overlap across the two domains and leads to even better cross-domain recommendation performance. We test the proposed model on three large-scale industrial datasets and demonstrate that it consistently and significantly outperforms the state-of-the-art baselines. We also show that the proposed model works well with very few overlap users to obtain recommendation performance comparable to the state-of-the-art baselines that use many overlap users.

Multi-criteria recommender systems have been increasingly useful for helping consumers identify the most relevant items based on different dimensions of user experiences and highlighting their most valued features. Therefore, researchers have proposed various multi-criteria models to improve recommendation performance. However, most of the existing methods utilize only multi-criteria ratings explicitly provided by the users. Note that explicit multi-criteria ratings are sparse and have the problem of missing values. User reviews, on the other hand, contain richer information of user experiences and reveal multi-dimensional user preferences. Therefore, we propose to use latent multi-criteria ratings generated from user reviews, as opposed to explicit multi-criteria ratings, to provide recommendations and capture latent complex heterogeneous user preferences. Specifically, we propose two novel models for the latent multi-criteria rating generation process: the one-stage model LatentMC-1S that utilizes document hashing method to directly compute latent ratings and the two-stage model LatentMC-2S that uses GRU and Gumbel-Softmax for indirect rating generation. Extensive experiments show that the proposed latent multi-criteria rating approaches outperform explicit ratings across different datasets and performance measures. We also show that latent multi-criteria ratings could be used for imputing missing explicit multi-criteria ratings and thus further improving multi-criteria recommender systems.

Cross domain recommender system constitutes a powerful method to tackle the cold-start and sparsity problem by aggregating and transferring user preferences across multiple category domains. Therefore, it has great potential to improve click-through-rate prediction performance in online commerce platforms having many domains of products. While several cross domain sequential recommendation models have been proposed to leverage information from a source domain to improve CTR predictions in a target domain, they did not take into account bidirectional latent relations of user preferences across source-target domain pairs. As such, they cannot provide enhanced cross-domain CTR predictions for both domains simultaneously. In this paper, we propose a novel approach to cross-domain sequential recommendations based on the dual learning mechanism that simultaneously transfers information between two related domains in an iterative manner until the learning process stabilizes. In particular, the proposed Dual Attentive Sequential Learning (DASL) model consists of two novel components Dual Embedding and Dual Attention, which jointly establish the two-stage learning process: we first construct dual latent embeddings that extract user preferences in both domains simultaneously, and subsequently provide cross-domain recommendations by matching the extracted latent embeddings with candidate items through dual-attention learning mechanism. We conduct extensive offline experiments on three real-world datasets to demonstrate the superiority of our proposed model, which significantly and consistently outperforms several state-of-the-art baselines across all experimental settings. We also conduct an online A/B test at a major video streaming platform Alibaba-Youku, where our proposed model significantly improves business performance over the latest production system in the company.

Recommender systems have been widely adopted by many firms across various industries over the past decade, as they could provide numerous economic benefits to the industry, such as influencing consumer choices, generating the lift in sales, and enhancing consumer trust. Many currently-deployed recommender systems primiarily optimize the similarity measure following the Collaborative Filtering paradigm during the recommendation performance, which focuses on the matching between targeted users and items of their interests. Although effective and useful in many cases, these still face several important challenges to fully address consumers’ concerns. In particular, they usually provide recommendations of the most similar items related to the consumption records, without taking into account the multi-faceted consumer preferences. For example, some consumers would be satisfied to receive recommendations that are novel and unexpected to broaden their horizons, while others might be interested to receive recommendations from other product categories to expand their interests. Falling to do so would lead to the filter bubble and exploration-exploitation trade-off phenomenon, which is undesirable because it would significantly reduce sales diversity and deteriorate user satisfaction with recommender systems.

Classical recommender system methods typically face the filter bubble problem when users only receive recommendations of their familiar items, making them bored and dissatisfied. To address the filter bubble problem, unexpected recommendations have been proposed to recommend items significantly deviating from user’s prior expectations and thus surprising them by presenting ”fresh” and previously unexplored items to the users. In this paper, we describe a novel Personalized Unexpected Recommender System (PURS) model that incorporates unexpectedness into the recommendation process by providing multi-cluster modeling of user interests in the latent space and personalized unexpectedness via the self-attention mechanism and via selection of an appropriate unexpected activation function. Extensive offline experiments on three real-world datasets illustrate that the proposed PURS model significantly outperforms the state-of-the-art baseline approaches in terms of both accuracy and unexpectedness measures. In addition, we conduct an online A/B test at a major video platform Alibaba-Youku, where our model achieves over 3% increase in the average video view per user metric. The proposed model is in the process of being deployed by the company.

Cross domain recommender systems have been increasingly valuable for helping consumers identify the most satisfying items from different categories. However, previously proposed cross-domain models did not take into account bidirectional latent relations between users and items. In addition, they do not explicitly model information of user and item features, while utilizing only user ratings information for recommendations. To address these concerns, in this paper we propose a novel approach to cross-domain recommendations based on the mechanism of dual learning that transfers information between two related domains in an iterative manner until the learning process stabilizes. We develop a novel latent orthogonal mapping to extract user preferences over multiple domains while preserving relations between users across different latent spaces. Combining with autoencoder approach to extract the latent essence of feature information, we propose Deep Dual Transfer Cross Domain Recommendation (DDTCDR) model to provide recommendations in respective domains. We test the proposed method on a large dataset containing three domains of movies, book and music items and demonstrate that it consistently and significantly outperforms several state-of-the-art baselines and also classical transfer learning approaches.

Unexpectedness constitutes an important factor for recommender system to improve user satisfaction and avoid filter bubble issues. In this proposal, we propose to provide unexpected recommendations using the hybrid utility function as a mixture of estimated ratings, unexpectedness, relevance and annoyance. We plan to conduct extensive experiments to validate the superiority of the proposed method.

Unexpected recommender system constitutes an important tool to tackle the problem of filter bubbles and user boredom, which aims at providing unexpected and satisfying recommendations to target users at the same time. Previous unexpected recommendation methods only focus on the straightforward relations between current recommendations and user expectations by modeling unexpectedness in the feature space, thus resulting in the loss of accuracy measures to improve unexpectedness performance. In contrast to these prior models, we propose to model unexpectedness in the latent space of user and item embeddings, which allows us to capture hidden and complex relations between new recommendations and historic purchases. In addition, we develop a novel Latent Closure (LC) method to construct a hybrid utility function and provide unexpected recommendations based on the proposed model. Extensive experiments on three real-world datasets illustrate superiority of our proposed approach over the state-of-the-art unexpected recommendation models, which leads to significant increase in unexpectedness measure without sacrificing any accuracy metric under all experimental settings in this article.

Multi-criteria recommender systems have been increasingly valuable for helping consumers identify the most relevant items based on different dimensions of user experiences. However, previously proposed multi-criteria models did not take into account latent embeddings generated from user reviews, which capture latent semantic relations between users and items. To address these concerns, we utilize variational autoencoders to map user reviews into latent embeddings, which are subsequently compressed into low-dimensional discrete vectors. The resulting compressed vectors constitute latent multi-criteria ratings that we use for the recommendation purposes via standard multi-criteria recommendation methods. We show that the proposed latent multi-criteria rating approach outperforms several baselines significantly and consistently across different datasets and performance evaluation measures.