This paper addresses the challenge of insufficient sampling diversity in large language models (LLMs) under the conventional autoregressive decoding framework. Our research reveals that, in some tasks where LLMs underperform, they actually possess the capability to provide correct answers. However, this potential is limited by inadequate sampling diversity in the decoding process, which prevents the models from effectively and fully leveraging their reasoning capabilities. To address this issue, we propose the Adaptive Trigram Model-Assisted Decoding Strategy (ATM-ADS), a method designed to enhance the reasoning abilities of large models without requiring fine-tuning. By dynamically providing diverse candidate vocabularies, combined with a decoding dynamic decision-making mechanism and an adaptive smoothing strategy, our approach substantially enhances sampling diversity during the decoding process. Experimental results demonstrate that our ATM-ADS significantly improves model performance across multiple task scenarios, highlighting its broad potential for cross-task and cross-linguistic applications.

Relation triplet extraction aims to identify entity pairs and their relations from unstructured text. Traditional zero-shot learning methods are hindered by predefined relation types and the lack of large-scale annotated data, limiting generalization to unseen relations. To address these challenges, we propose the KOMADF (Knowledge-Optimized Multi-Agent Dynamics Framework). KOMADF constructs task-specific knowledge graphs and incorporates a distillation module to refine relation labels, filtering out irrelevant candidates while preserving semantically similar or multi-type labels. These refined labels generate problem representations that guide a multi-agent mechanism to dynamically create context-aware prompt templates. By aligning task objectives with pre-trained language models, these templates enable generalization to unseen relations.Experiments on two public datasets, FewRel and Wiki-ZSL, show that the proposed method substantially improves precision, recall, and F1 score. These findings confirm KOMADF’s adaptability and robustness, establishing it as an effective solution to the challenges of zero-shot relation triplet extraction. Furthermore, the proposed framework lays a solid theoretical foundation for the automated construction of dynamic knowledge graphs.

Introduced by Google in 2016, federated learning (FL) is a distributed machine learning framework to ensure data privacy amid the surge in big data. FL enables secure data sharing without accessing local data. Despite its advantages, it faces challenges due to the limited participation of the data owner. To address this, this article proposes the model value transfer incentive (MVTI) to enhance FL incentives for Artificial Intelligence of Things (AIoT). MVTI allows active participation of data requesters in FL training, addressing limited data owner engagement, and facilitating personalized model construction. The integrated model bail and contribution assessment mechanism ensures fair benefit redistribution. Using smart contracts (SCs) and interplanetary file system (IPFS) enhances security and reliability, ensuring transparent and tamper-resistant execution for secure transactions and data integrity. Our experiments highlight MVTI’s superiority in addressing FL incentive challenges for AIoT compared to state-of-the-art baselines on real-world datasets. We also demonstrate the compatibility of multiple gradient protections with incentive mechanisms, especially with gradient compression. The proposed SC-MVTI scheme is resilient and demonstrates the potential to significantly improve the overall efficacy of the FL system within incentive frameworks.

The goal of continuous few-shot relation extraction is to enable the model to continuously learn new relation types under conditions with limited labeled training data while avoiding the forgetting of previously learned relations. The primary challenges include catastrophic forgetting of old relations and overfitting due to data sparsity. To address these challenges, this article proposes a hierarchical distillation model that innovatively combines contrastive distillation with orthogonal adversarial distillation techniques. Specifically, we introduce a contrastive distillation approach in the feature distillation layer, integrating adaptive cosine techniques and negative sampling strategies to ensure that the model effectively retains and utilizes knowledge from previous tasks when learning new ones. Additionally, we employ orthogonal adversarial distillation in the hidden distillation layer to alleviate overfitting in low-resource scenarios. Experimental results demonstrate that our proposed method significantly outperforms the current state-of-the-art models for continuous few-shot relation extraction on two benchmark datasets, validating its effectiveness in handling few-shot data and knowledge transfer.

With the rise of large-scale language models (LLMs), it is currently popular and effective to convert multimodal information into text descriptions for multimodal multi-hop question answering. However, we argue that the current methods of multi-modal multi-hop question answering still mainly face two challenges: 1) The retrieved evidence containing a large amount of redundant information, inevitably leads to a significant drop in performance due to irrelevant information misleading the prediction. 2) The reasoning process without interpretable reasoning steps makes the model difficult to discover the logical errors for handling complex questions. To solve these problems, we propose a unified LLMs-based approach but without heavily relying on them due to the LLM's potential errors, and innovatively treat multimodal multi-hop question answering as a joint entailment tree generation and question answering problem. Specifically, we design a multi-task learning framework with a focus on facilitating common knowledge sharing across interpretability and prediction tasks while preventing task-specific errors from interfering with each other via mixture of experts. Afterward, we design an iterative feedback mechanism to further enhance both tasks by feeding back the results of the joint training to the LLM for regenerating entailment trees, aiming to iteratively refine the potential answer. Notably, our method has won the first place in the official leaderboard of WebQA (since April 10, 2024), and achieves competitive results on MultimodalQA.

Abstract Conversational recommender system (CRS) needs to be seamlessly integrated between the two modules of recommendation and dialog, aiming to recommend high-quality items to users through multiple rounds of interactive dialogs. Items can typically refer to goods, movies, news, etc. Through this form of interactive dialog, users can express their preferences in real time, and the system can fully understand the user’s thoughts and recommend corresponding items. Although mainstream dialog recommendation systems have improved the performance to some extent, there are still some key issues, such as insufficient consideration of the entity’s order in the dialog, the different contributions of items in the dialog history, and the low diversity of generated responses. To address these shortcomings, we propose an improved dialog context model based on time-series features. Firstly, we augment the semantic representation of words and items using two external knowledge graphs and align the semantic space using mutual information maximization techniques. Secondly, we add a retrieval model to the dialog recommendation system to provide auxiliary information for generating replies. We then utilize a deep timing network to serialize the dialog content and more accurately learn the feature relationship between users and items for recommendation. In this paper, the dialog recommendation system is divided into two components, and different evaluation indicators are used to evaluate the performance of the dialog component and the recommendation component. Experimental results on widely used benchmarks show that the proposed method is effective.

Document-level relational extraction requires reading, memorization, and reasoning to discover relevant factual information in multiple sentences. It is difficult for the current hierarchical network and graph network methods to fully capture the structural information behind the document and make natural reasoning from the context. Different from the previous methods, this article reconstructs the relation extraction task into a machine reading comprehension task. Each pair of entities and relationships is characterized by a question template, and the extraction of entities and relationships is translated into identifying answers from the context. To enhance the context comprehension ability of the extraction model and achieve more precise extraction, we introduce large language models (LLMs) during question construction, enabling the generation of exemplary answers. Besides, to solve the multi-label and multi-entity problems in documents, we propose a new answer extraction model based on hybrid pointer-sequence labeling, which improves the reasoning ability of the model and realizes the extraction of zero or multiple answers in documents. Extensive experiments on three public datasets show that the proposed method is effective.

Event detection tasks can enable the quick detection of events from texts and provide powerful support for downstream natural language processing tasks. Most such methods can only detect a fixed set of predefined event classes. To extend them to detect a new class without losing the ability to detect old classes requires costly retraining of the model from scratch. Incremental learning can effectively solve this problem, but it requires abundant data of new classes. In practice, however, the lack of high-quality labeled data of new event classes makes it difficult to obtain enough data for model training. To address the above mentioned issues, we define a new task, few-shot incremental event detection, which focuses on learning to detect a new event class with limited data, while retaining the ability to detect old classes to the extent possible. We created a benchmark dataset IFSED for the few-shot incremental event detection task based on FewEvent and propose two benchmarks, IFSED-K and IFSED-KP. Experimental results show that our approach has a higher F1-score than baseline methods and is more stable.

Spelling error detection serves as a crucial preprocessing in many natural language processing applications. Unlike English, where every single word is directly typed by keyboard, we have to use an input method to input Chinese characters. The pinyin input method is the most widely used. By intuition, pinyin should be helpful in detecting spelling errors. However, when detect spelling errors, most of the current methods ignore the pinyin information and adopt a pipeline framework that leads to error propagation. In this article, we propose a fusion lattice-LSTM model under the end-to-end framework to integrate character, word, and pinyin features for error detection. Experiments on the SIGHAN Bake-off-2015 dataset show that pinyin is a discriminating feature, and our end-to-end model outperforms the baseline models obviously.

Erasure codes are widely used in cloud storage systems, such as in Google File System and Windows Azure. However, most of existing erasure codes focus on homogeneous storage device, but ignore that heterogeneous devices are in majority in cloud storage system. In this paper, we propose a new erasure code framework termed Device Placement Aware Framework (DPAF), to integrate existing erasure codes to generate DPAF-Codes, in order to gain good performance on heterogeneous storage devices. The key insight is to detect the device performance to generate a series of coefficients, and use these coefficients to choose proper devices to construct DPAF-Codes. We utilize simulated annealing algorithm to optimize the selection process, in order to maintain the balance among devices by considering the coefficients. The experiment results show that, our proposed DPAF-Codes gain up to 70.3%, 48.7%, and 48.5% improvements on single failure recovery, normal read, and degraded read speed, compared to RS code and LRC code with different configurations.