Publications

Large Language Models are increasingly deployed as autonomous agents for complex real-world tasks, yet existing systems often focus on isolated improvements without a unifying design for robustness and adaptability. We propose a generalist agent architecture that integrates three core components: a collective multi-agent framework combining planning and execution agents with critic model voting, a hierarchical memory system spanning working, semantic, and procedural layers, and a refined tool suite for search, code execution, and multimodal parsing. Evaluated on a comprehensive benchmark, our framework consistently outperforms open-source baselines and approaches the performance of proprietary systems. These results demonstrate the importance of system-level integration and highlight a path toward scalable, resilient, and adaptive AI assistants capable of operating across diverse domains and tasks.

Large language models (LLMs) have demonstrated their significant potential to be applied for addressing various application tasks. However, traditional recommender systems continue to face great challenges such as poor interactivity and explainability, which actually also hinder their broad deployment in real-world systems. To address these limitations, this paper proposes a novel paradigm called Chat-Rec (ChatGPT Augmented Recommender System) that innovatively augments LLMs for building conversational recommender systems by converting user profiles and historical interactions into prompts. Chat-Rec is demonstrated to be effective in learning user preferences and establishing connections between users and products through in-context learning, which also makes the recommendation process more interactive and explainable. What's more, within the Chat-Rec framework, user's preferences can transfer to different products for cross-domain recommendations, and prompt-based injection of information into LLMs can also handle the cold-start scenarios with new items. In our experiments, Chat-Rec effectively improve the results of top-k recommendations and performs better in zero-shot rating prediction task. Chat-Rec offers a novel approach to improving recommender systems and presents new practical scenarios for the implementation of AIGC (AI generated content) in recommender system studies.

Retrieval-augmented code generation utilizes Large Language Models as the generator and significantly expands their code generation capabilities by providing relevant code, documentation, and more via the retriever. The current approach suffers from two primary limitations: 1) information redundancy. The indiscriminate inclusion of redundant information can result in resource wastage and may misguide generators, affecting their effectiveness and efficiency. 2) preference gap. Due to different optimization objectives, the retriever strives to procure code with higher ground truth similarity, yet this effort does not substantially benefit the generator. The retriever and the generator may prefer different golden code, and this gap in preference results in a suboptimal design. Additionally, differences in parameterization knowledge acquired during pre-training result in varying preferences among different generators. To address these limitations, in this paper, we propose RRG (Retrieve, Refactor, Generate), a novel framework for effective and efficient code generation. This framework introduces a code refactorer module between the retriever and the generator to bridge them. The refactoring process transforms the raw retrieved code into a more concise, efficient, and model-friendly version. It eliminates redundant information and noise, reducing the input length. Consequently, the generator receives higher-quality context, enabling it to produce more accurate results with lower inference costs. We conducted comprehensive experiments on multiple datasets. In the experiments, we confirmed the existence of a preference gap between the retriever and the generator, and RRG effectively bridges this gap. Specifically, RRG achieved significant performance improvements, with increases of up to 28% on EM, 13% on BLEU, and 6.8% on CodeBLEU.

To safely and effectively apply fatty acids and phenolic acids to control and prevent water-bloom,the article studied the toxicity of linoleic acid,salicylic acid and p-hydroxybenzoic acid on natural algal assemblage of pond water and zooplankton Moina macrocopa.The results showed that linoleic acid had the strongest toxicity,while salicylic acid and p-hydroxybenzoic acid had moderate toxicity on the natural algal assemblage of pond water,when the concentration of linoleic acid was 0.3mg/L,the biggest inhibitory ratio was up to 87.6%.The collective activity of a mixture of two allelochemicals were also examined and the synergistic growth inhibition on the natural algal assemblage were proved,in which the mixture of linoleic acid and salicylic acid had stronger toxicity.The three allelochemicals to some extent also showed toxicity on Moina macrocopa,but linoleic acid had stronger toxicity,while salicylic acid and p-hydroxybenzoic acid had lower toxicity on Moina macrocopa,the EC50s of the three allelochemicals on Moina macrocopa was far higher than the EC50s on the natural algal assemblage,which indicated that there exists great potential for using the three allelochemicals,especially linoleic acid,to be developed biological algicide agent while no evident toxicity on environment within the concentrations of controling the growth of water-bloom algae.

<strong>List of files:</strong> entity2image.json: The entity to map images file MMpedia_triples.ttl: The triples file EntlistXX.tar: The image file MMpedia dataset is split into 132 subsets and each subset is compressed into a <code>.tar</code> file. After unziping files under the folder "MMpedia", the data structure are as following: <pre><code>|-MMpedia |-Entitylist1 |-Entity1 |-1.jpg |-2.jpg |-3.jpg ... |-Entity2 |-Entity3 ... |-Entitylist2 |-Entitylist3</code></pre> For example, the path <code>MMpedia/Entlist141/Bart Tanski/Bart Tanski+1.jpg</code> means the image corresponding to the entity "Bart Tanski" <strong>Other image files can be found in following URLs:</strong> MMpedia2 | Zenodo MMpedia3 | Zenodo MMpedia4 | Zenodo <strong>For more details, please refer to </strong>Delicate2000/MMpedia (github.com)

Few-shot learning (FSL) presents immense potential in enhancing model generalization and practicality for medical image classification with limited training data; however, it still faces the challenge of severe overfitting in classifier training due to distribution bias caused by the scarce training samples. To address the issue, we propose MedMFG, a flexible and lightweight plug-and-play method designed to generate sufficient class-distinctive features from limited samples. Specifically, MedMFG first re-represents the limited prototypes to assign higher weights for more important information features. Then, the prototypes are variationally generated into abundant effective features. Finally, the generated features and prototypes are together to train a more generalized classifier. Experiments demonstrate that MedMFG outperforms the previous state-of-the-art methods on cross-domain benchmarks involving the transition from natural images to medical images, as well as medical images with different lesions. Notably, our method achieves over 10% performance improvement compared to several baselines. Fusion experiments further validate the adaptability of MedMFG, as it seamlessly integrates into various backbones and baselines, consistently yielding improvements of over 2.9% across all results.

While several quality assessment tools focus on evaluating the quality of Linking Open Data (LOD), most tools fail to meet diverse quality assessment requirements from the users’ perspective. In this demo, we categorized quality assessments requirements into three layers: understanding the characteristics of data sets, comparing groups of data sets, and selecting data sets according to user-defined usage scenarios. We have designed KBMetrics to incorporate the above quality assessment purposes. Not only does the tool incorporate different kinds of metrics to characterize a data set, but it has also adopted ontology alignment mechanisms for comparison purposes. Most importantly, end users can define usage contexts to adapt to different usage scenarios. Both the quality assessment processes and findings in these data sets show the effectiveness of our tool.

Chinese abbreviation prediction plays an important role in natural language processing. The prevalent approach often utilizes generation models to predict abbreviations for full forms, but relying solely on a single generation model may not yield high-quality abbreviations. We emphasize the importance of introducing an evaluation model after the generation model to assess the rationality of generated abbreviations. Hence, in this paper, we propose a novel two-stage method with LLM generation and contrastive evaluation for Chinese abbreviation prediction. In the first stage, we design a type discriminator to determine the abbreviation type and then introduce a pre-trained and fine-tuned LLM to generate multiple candidate abbreviations. In the second stage, we propose a contrastive evaluation model to assess the rationality of the candidates based on the abbreviation scorer and phrase scorer with a joint learning strategy. Experiments on two public datasets indicate that our method outperforms the current state-of-the-art method, achieving improvements of 3.32% and 1.73%, respectively. More importantly, we deploy it on the Fliggy application and the 20-day online A/B testing shows a 0.65% increase in Point of Interest Recognition Rate and a 1.37% increase in Page View Click-Through Rate when using abbreviations predicted by our method in the search system.