Despite the assortment optimization problem has been widely studied in the past decades, the interplay between advertising and its implications for this issue remains under-explored. This study seeks to bridge this research gap by tackling the combined challenge of advertising and assortment optimization. We assume that advertising can increase the awareness of specific products, and the magnitude of this effect is jointly depends on the product-specific effectiveness of advertising and the allocated advertising budget. For this joint problem, our objective is to maximize the expected revenue by finding the optimal advertising strategy and the displayed assortment. In this work, we analyze the structure of this problem and propose efficient approaches to solve it across different scenarios. In the unconstrained setting, we demonstrate that the optimal assortment includes products whose revenue exceeds a certain threshold. When there is a cardinality constraint for the assortment, we consider a relaxed problem and propose an efficient method to identify a near-optimal solution. We also examine the joint assortment, pricing, and advertising problem in both unconstrained and cardinality-constrained settings, incorporating the fairness constraint for the advertising strategy and extending our findings to account for consumer sequential decision-making patterns. Through a series of numerical tests, we confirm the validity of our methods and demonstrate that they outperform existing heuristic approaches.

Large Language Models (LLMs) have demonstrated amazing capabilities in language generation, text comprehension, and knowledge reasoning. While a single powerful model can already handle multiple tasks, relying on a single perspective can lead to biased and unstable results. Recent studies have further improved the model's reasoning ability on a wide range of tasks by introducing multi-model collaboration. However, models with different capabilities may produce conflicting answers on the same problem, and how to reasonably obtain the correct answer from multiple candidate models has become a challenging problem. In this paper, we propose the multi-model brainstorming based on prompt. It incorporates different models into a group for brainstorming, and after multiple rounds of reasoning elaboration and re-inference, a consensus answer is reached within the group. We conducted experiments on three different types of datasets, and demonstrate that the brainstorming can significantly improve the effectiveness in logical reasoning and fact extraction. Furthermore, we find that two small-parameter models can achieve accuracy approximating that of larger-parameter models through brainstorming, which provides a new solution for distributed deployment of LLMs.

Imparting human-like commonsense to machines is a long-term goal in the artificial intelligence community.To achieve this goal,constructing large-scale commonsense knowledge resources is an important step.In recent years,due to increasing demand,com-monsense knowledge has become a rapidly growing research field,resulting in a surge of new acquisition methods and corresponding re-sources.These advances have empowered a variety of downstream AI tasks.However,constructing large-scale commonsense knowledge resources remains an ongoing and challenging task.It is still difficult to efficiently collect large-scale,high-quality commonsense know-ledge.In this paper,we systematically review recent advances in commonsense knowledge acquisition methods and resources,providing a comprehensive summary of recent research scope,the characteristics of different resources,and unsolved challenges.

As Large Language Models (LLMs) rapidly advance, we introduce Hunyuan-TurboS, a novel large hybrid Transformer-Mamba Mixture of Experts (MoE) model. It synergistically combines Mamba's long-sequence processing efficiency with Transformer's superior contextual understanding. Hunyuan-TurboS features an adaptive long-short chain-of-thought (CoT) mechanism, dynamically switching between rapid responses for simple queries and deep "thinking" modes for complex problems, optimizing computational resources. Architecturally, this 56B activated (560B total) parameter model employs 128 layers (Mamba2, Attention, FFN) with an innovative AMF/MF block pattern. Faster Mamba2 ensures linear complexity, Grouped-Query Attention minimizes KV cache, and FFNs use an MoE structure. Pre-trained on 16T high-quality tokens, it supports a 256K context length and is the first industry-deployed large-scale Mamba model. Our comprehensive post-training strategy enhances capabilities via Supervised Fine-Tuning (3M instructions), a novel Adaptive Long-short CoT Fusion method, Multi-round Deliberation Learning for iterative improvement, and a two-stage Large-scale Reinforcement Learning process targeting STEM and general instruction-following. Evaluations show strong performance: overall top 7 rank on LMSYS Chatbot Arena with a score of 1356, outperforming leading models like Gemini-2.0-Flash-001 (1352) and o4-mini-2025-04-16 (1345). TurboS also achieves an average of 77.9

When a platform has limited inventory, it is important to have a variety of products available for each customer while managing the remaining stock. To maximize revenue over the long term, the assortment policy needs to take into account the complex purchasing behavior of customers whose arrival orders and preferences may be unknown. We propose a data-driven approach for dynamic assortment planning that utilizes historical customer arrivals and transaction data. To address the challenge of online assortment customization, we use a Markov decision process framework and employ a model-free deep reinforcement learning (DRL) approach to solve the online assortment policy because of the computational challenge. Our method uses a specially designed deep neural network (DNN) model to create assortments while observing the inventory constraints, and an advantage actor-critic algorithm to update the parameters of the DNN model, with the help of a simulator built from the historical transaction data. To evaluate the effectiveness of our approach, we conduct simulations using both a synthetic data set generated with a pre-determined customer type distribution and ground-truth choice model, as well as a real-world data set. Our extensive experiments demonstrate that our approach produces significantly higher long-term revenue compared to some existing methods and remains robust under various practical conditions. We also demonstrate that our approach can be easily adapted to a more general problem that includes reusable products, where customers might return purchased items. In this setting, we find that our approach performs well under various usage time distributions.

Non-Abelian gauge fields, characterized by their non-commutative symmetry groups, shape physical laws from the Standard Model to emergent topological matter for quantum computation. Here we find that moiré exciton dimers (biexcitons) in the twisted bilayer MoTe[Formula: see text] are governed by a genuine non-Abelian lattice gauge field. These dipolar-bound exciton dimers, formed on bonds of the honeycomb moiré superlattice, exhibit three quadrupole configurations organized into a Kagome lattice geometry, on which the valley-flip biexciton hoppings through electron-hole Coulomb exchange act as link variables of the non-Abelian lattice gauge theory. The emergence of the gauge structure here is a new possibility for composite particles, where the moiré electronic structure and interactions between the electron and hole constituents jointly enforce the underlying geometric constraint. The quadrupole nature of the biexciton further makes possible local gate controls to isolate designated pathways from the extended lattice for exploiting consequences of non-commutative gauge structure, including the genuine non-Abelian Aharonov-Bohm effect. This also provides a new approach for quantum manipulation of the excitonic valley qubit. We show that path interference on a simplest loop can deterministically transform the computational basis states into Bell states.

With the development of deep learning, natural language processing technology has effectively improved the efficiency of various aspects of the traditional judicial industry. However, most current efforts focus on tasks within individual judicial stages, making it difficult to handle complex tasks that span multiple stages. As the autonomous agents powered by large language models are becoming increasingly smart and able to make complex decisions in real-world settings, offering new insights for judicial intelligence. In this paper, (1) we propose a novel multi-agent framework, AgentsCourt, for judicial decision-making. Our framework follows the classic court trial process, consisting of court debate simulation, legal resources retrieval and decision-making refinement to simulate the decision-making of judge. (2) we introduce SimuCourt, a judicial benchmark that encompasses 420 Chinese judgment documents, spanning the three most common types of judicial cases. Furthermore, to support this task, we construct a large-scale legal knowledge base, Legal-KB, with multi-resource legal knowledge. (3) Extensive experiments show that our framework outperforms the existing advanced methods in various aspects, especially in generating legal articles, where our model achieves significant improvements of 8.6% and 9.1% F1 score in the first and second instance settings, respectively.

Adjusting the outdated behaviors of large langugae models (LLMs) after deployment remains a significant challenge. It motivates the model editing research, which is however mainly explored in a restricted task form with triplebased edit requests. Recent works have initiated a transition to a more practical and unified editing task that takes free-form text as edit requests. However, there are gaps in nuanced benchmark designs and re-evaluation of existing methods. To bridge the gaps, we introduce a multi-level benchmark for free text model editing ( MULFE). The benchmark categorizes probe queries into three levels of generalization, ranging from basic literal memory to deeper understanding and reasoning. Based on the benchmark, we conduct extensive experiments across various base models, edit sizes, and editing methods, including adaptations of mainstream locate-and-edit and hypernetwork methods. The results highlight the inconsistent behaviors of edited models on different generalization levels. Higherlevel generalization remains a significant challenge. Based on the findings, we propose SIDE, a simple yet effective method based on in-context distillation to enhance the generalization performance. The benchmark dataset and evaluation scripts are publicly available at http://github.com/wchrepo/mulfe.

When a retailer has limited inventory and is operating on a periodic selling schedule, it is important to have a variety of products available for each customer. To maximize revenue over the long term, an optimal assortment policy is required that takes into account the complex purchasing behaviors of customers whose arrival order and preferences are unknown. By analyzing historical customer arrival and transaction data, we propose a data-driven approach for dynamic assortment planning. To address the challenge of online assortment customization, we utilize a Markov decision process (MDP) framework and employ a model-free deep reinforcement learning (DRL) approach to learn a policy that is nearly optimal. Our method involves using a specialized deep learning model called Gated-DNN to create assortments while adhering to constraints, and a modified version of the Advantage Actor-Critic (A2C) algorithm to adjust the parameters of the Gated-DNN model. The updates to the model’s parameters are done by simulated interactions with historical sequences of customer arrivals. The feedback we receive from simulated customers can take any form and should match the historical transaction data as closely as possible to ensure the effectiveness of the policy we learn. To evaluate the effectiveness of our approach, we conduct simulations using both a synthetic data set generated with a pre-determined customer type distribution and choice model, as well as a real-world data set. Our extensive experiments demonstrate that our approach produces significantly higher long-term revenue compared to existing methods and remains robust under various conditions. We also demonstrate that our approach can be easily adapted to a more general problem that includes reusable products, where customers return purchased items after a period of time. In further experiments, we find that our approach performs best under various usage time distributions.

Large language models (LLMs) sometimes demonstrate poor performance on knowledge-intensive tasks, commonsense reasoning is one of them. Researchers typically address these issues by retrieving related knowledge from knowledge graphs or employing self-enhancement methods to elicit knowledge in LLMs. However, noisy knowledge and invalid reasoning issues hamper their ability to answer questions accurately. To this end, we propose a novel method named eliciting, filtering and integrating knowledge in large language model (LINKED). In it, we design a reward model to filter out the noisy knowledge and take the marginal consistent reasoning module to reduce invalid reasoning. With our comprehensive experiments on two complex commonsense reasoning benchmarks, our method outperforms SOTA baselines (up to 9.0% improvement of accuracy). Besides, to measure the positive and negative impact of the injected knowledge, we propose a new metric called effectiveness-preservation score for the knowledge enhancement works. Finally, through extensive experiments, we conduct an in-depth analysis and find many meaningful conclusions about LLMs in commonsense reasoning tasks.

Large language models (LLMs) inevitably memorize sensitive, copyrighted, and harmful knowledge from the training corpus; therefore, it is crucial to erase this knowledge from the models. Machine unlearning is a promising solution for efficiently removing specific knowledge by post hoc modifying models. In this paper, we propose a Real-World Knowledge Unlearning benchmark (RWKU) for LLM unlearning. RWKU is designed based on the following three key factors: (1) For the task setting, we consider a more practical and challenging unlearning setting, where neither the forget corpus nor the retain corpus is accessible. (2) For the knowledge source, we choose 200 real-world famous people as the unlearning targets and show that such popular knowledge is widely present in various LLMs. (3) For the evaluation framework, we design the forget set and the retain set to evaluate the model's capabilities across various real-world applications. Regarding the forget set, we provide four four membership inference attack (MIA) methods and nine kinds of adversarial attack probes to rigorously test unlearning efficacy. Regarding the retain set, we assess locality and utility in terms of neighbor perturbation, general ability, reasoning ability, truthfulness, factuality, and fluency. We conduct extensive experiments across two unlearning scenarios, two models and six baseline methods and obtain some meaningful findings. We release our benchmark and code publicly at http://rwku-bench.github.io for future work.

Large language models exhibit high-level commonsense reasoning abilities, especially with enhancement methods like Chain-of-Thought (CoT). However, we find these CoT-like methods lead to a considerable number of originally correct answers turning wrong, which we define as the Toxic CoT problem. To interpret and mitigate this problem, we first utilize attribution tracing and causal tracing methods to probe the internal working mechanism of the LLM during CoT reasoning. Through comparisons, we prove that the model exhibits information loss from the question over the shallow attention layers when generating rationales or answers. Based on the probing findings, we design a novel method called RIDERS (Residual decodIng and sERial-position Swap), which compensates for the information deficit in the model from both decoding and serial-position perspectives. Through extensive experiments on multiple commonsense reasoning benchmarks, we validate that this method not only significantly eliminates Toxic CoT problems (decreased by 23.6%), but also effectively improves the model's overall commonsense reasoning performance (increased by 5.5%).

Event ontology provides a shared and formal specification about what happens in the real world and can benefit many natural language understanding tasks. However, the independent development of event ontologies often results in heterogeneous representations that raise the need for establishing alignments between semantically related events. There exists a series of works about ontology alignment (OA), but they only focus on the entity-based OA, and neglect the event-based OA. To fill the gap, we construct an Event Ontology Alignment ( EventOA ) dataset based on FrameNet and Wikidata, which consists of 900+ event type alignments and 8,000+ event argument alignments. Furthermore, we propose a multi-view event ontology alignment (MEOA) method, which utilizes description information (i.e., name, alias and definition) and neighbor information (i.e., subclass and superclass) to obtain richer representation of the event ontologies. Extensive experiments show that our MEOA outperforms the existing entity-based OA meth-ods and can serve as a strong baseline for Even-tOA research.

Recently, there are a lot of works trying to probe knowledge in pre-trained language models (PLMs). Most probing works use data in knowledge bases to create a "fill-in-the-blank" task form and probe entity knowledge in auto-encoding PLMs (e.g. BERT). Though these works have got success, their methods can not be applied to some complicated knowledge like event-based commonsense knowledge and other PLMs such as auto-regressive models (like GPT). In this paper, we develop a new knowledge probe based on confidence sorting and detect event-based commonsense knowledge with it. To make the probe suitable for different types of PLMs, we integrate different knowledge scoring methods with a new method called probability difference log-likelihood score (PDL) among them. Finally, we conduct extensive experiments on several representative PLMs, explore their commonsense abilities and analyze the factors that influence their performances.

Empirical evidence suggests that consumers commonly focus their attention on a subset of available products and evaluate them in batches to identify a satisfactory option. To capture this phenomenon, we introduce the Attention-Based Satisficing choice rule, which encompasses several specific cases such as the Sequential MNL (e.g., Gao et al. 2021), Click-Based MNL (e.g., Aouad et al. 2019), and Random Consideration Set models (e.g., Gallego and Li 2017). Previous research on these cases indicates that finding the optimal revenue-maximizing assortment and estimating certain parameters fall within the NP-hard complexity class. Despite these challenges, we demonstrate that the proposed model can be approximated by a Cascade model (e.g., Kempe and Mahdian 2008) with substantially fewer parameters in two ways: (1) the overall likelihood of purchasing from any given assortment under the proposed model can be estimated within a certain range, multiplied by that in the Cascade model, and (2) by utilizing the optimal assortment from the approximated model as a heuristic for the proposed model, the worst-case revenue is consistently at least a tightly predetermined constant (3/8) of that obtained from an optimized assortment under the best parameter configuration. Finally, leveraging the technique established in this study, we also analyze the constrained assortment optimization problem, the categorized attention-based assortment optimization problem, and the joint assortment and pricing problem.

Being able to infer possible events related to a specific target is critical to natural language processing. One challenging task in this line is