最新刊期
卷 11 ， 期 2 ， 2025
-
- 摘要：The financial industry has long faced challenges in processing vast amounts of market data and information. Currently, large language models have made significant progress in general text understanding tasks, but there is still considerable room for improvement in more specialized domains, such as Chinese finance. To address the limitations of current large language models in handling professional domain-specific text tasks, a two-stage training approach based on finance knowledge-enhanced continued pre-training and supervised fine-tuning is designed. This approach improves the organization of training data and the training paradigm, thereby enhancing the model's capabilities in complex financial scenarios. Finally, experiments have validated the effectiveness of the proposed knowledge-enhanced approach in large model training.关键词：large language model;financial time series forecasting862|555|0更新时间：2025-03-15
- 摘要：Traditional knowledge graph technologies still face significant challenges in converting human knowledge, expressed in natural language, into a formal language-based knowledge graph and utilizing it effectively. In recent years, code large language models (LLMs) have demonstrated remarkable capabilities in understanding both natural and formal languages, as well as in translating between them. These advancements are expected to drive significant breakthroughs in developing next-generation knowledge graph technologies. This paper reviews the application of code LLMs in KGs. Firstly, this paper systematically analyzes the role of code LLMs in enhancing key knowledge graph technologies across three critical areas: construction, reasoning, and question-answering. Secondly, a relatively detailed introduction to the existing methodologies in these areas is provided. Finally, this paper summarizes the current state in the field and offers insights into the future of knowledge graph technologies empowered by code LLMs. In the future, knowledge representation based on programming languages is expected to enable more efficient, automated, and complex operations on knowledge graphs, realizing knowledge programming.关键词：code large language model;large language model1023|328|0更新时间：2025-03-15
-
Research on recommendation system based on knowledge graph and large language model enhancement AI导读
摘要：The core of recommendation system is user and commodity, the relationship between user and commodity can be abstracted as graph structure, so graph neural network has been widely used in recommendation field. However, graph-based recommendation interaction data is sparse, relying heavily on numbering information and graph structure information, ignoring valuable text information related to user and commodity, and less representational information. At the same time, there are some noise and bias in the implicit feedback data, which bring challenges for recommendation system to understand user behavior and preferences. To solve these problems, this paper proposes a recommendation system based on knowledge graph and large language model enhancement. Knowledge graphs can provide structured information about commodities, enabling models to learn potential relationships between commodities and understand user behavior and preferences. Large language models have excellent ability to generate and understand, and can use prompt engineering techniques to deeply analyze and mine text information, and deduce the features of commoditiy and user portraits. The features enhanced by these auxiliary information are encoded, and the representations are enhanced to align with the ID representations obtained by the graph neural network to complete the downstream recommendation task. The experimental results show that the proposed system can help to comprehensively characterize users and commodities, and has good performance.关键词：knowledge graph;recommendation system;representation learning1664|1225|0更新时间：2025-03-15 - 摘要：As an important means of achieving sustainable carbon cycling, developing high-performance electrocatalysts is the key to the sustainable development in future, and recommending innovative and valuable preparation solutions is one of the effective ways to improve the efficiency of electrocatalytic development. This paper is based on scientific and technological literature in the field of electrocatalysis, inviting domain experts to construct a knowledge system and extract knowledge, forming a domain knowledge base. In addition, this paper utilizes literature data to fine tune and enhance knowledge of universal big language models, jointly completing preparation scheme recommendations for target products, material categories, and regulation method categories, and assisting in the design of electrocatalysts. Experimental results have shown that the preparation solutions provided by knowledge enhanced large language models have certain improvements in both effectiveness and innovation.关键词：knowledge enhancement;large language model472|784|0更新时间：2025-03-15
- 摘要：In view of the problem of fixed drug indication prediction quantity and inability to fully reveal potential indication of drug in current drug repositioning research, this article proposes a generative AI drug repositioning model GenDrugShifter, which consists of a graph attention neural network and a transformer decoder module. The model uses the drug molecular structure represented in the InChI format as input, learns the potential relationship between the drug active molecular structure and indication through self-supervision, and outputs the indication of the drug using autoregressive methods. The results of the Western medicine repositioning experiment show that GenDrugShifter is superior to the other four advanced drug repositioning methods in terms of predictive performance. The GenDrugShifter model can more comprehensively reveal the potential indication of drug, with superior performance and reliability. The validation of clinical data further proves its effectiveness in practical applications.关键词：drug repositioning;large language model;artificial intelligence;drug discovery405|249|0更新时间：2025-03-15
-
Large language model-based embodied intelligence task planning: from single-agent to multi-agent AI导读
摘要：With the development of artificial intelligence, embodied intelligence and task planning have gradually become research hotspots. Traditional task planning methods lack flexibility when facing unpredictable environments, while large language models, with their powerful language understanding and multimodal capabilities, provide more comprehensive task planning solutions for intelligent agents, which offers new possibilities for addressing this issue. This paper reviews task planning methods based on large language models, covering different strategies in both single-agent and multi-agent contexts. Several representative frameworks and their performance and potential in practical applications are discussed. Specifically, this paper introduces single-agent large language model task planning methods, such as end-to-end planning, staged planning, and dynamic planning, and multi-agent large language model task planning methods, such as centralized planning, distributed planning, and hybrid planning. It also analyzes how these methods combine with reinforcement learning, multimodal perception, and other techniques to optimize the planning process. In addition, the paper discusses the characteristics, limitations, and challenges of large language model-based embodied intelligence task planning, and outlines the future development directions. This paper aims to provide valuable insights for designing more flexible and adaptive next-generation embodied intelligent systems.关键词：task planning;large language model1083|428|0更新时间：2025-03-15
TOPIC
- 摘要：GNN is adept at aggregating information from neighboring nodes in graph-structured data to learn node representations, showing immense potential in the field of citation network data mining. However, most existing GNN lack a deep exploration of the factors driving edge information, which limits a thorough understanding and interpretation of complex relationships between nodes. For instance, the citation relationships between different papers are often driven by a variety of research topics. Despite attempts to enrich node and edge feature representations by integrating LLM with their strong textual understanding capabilities, these approaches have still not effectively sloved the problem of uncovering the underlying drivers of edge information. In light of this, an innovative framework was proposed—language model-enhanced edge factor mining，aimed to enhance the differentiation of edge relationship modeling in various GNN through a plug-in approach, exploring its application value in citation network link prediction scenarios. Coarse-grained factor mining extracted explicit category-related edge factors from citation network graphs containing documents by designing structured information prompts for LLM. Fine-grained factor mining used the K-means clustering algorithm to capture more detailed semantic topic-level edge factors from graph textual data. To verify the effectiveness of the proposed framework, experiments were conducted on three public datasets. The results demonstrate a significant advantage of language model-enhanced edge factor mining framework in improving the performance of various GNN models.关键词：graph neural network;edge factor mining;link prediction;citation network204|219|0更新时间：2025-03-15
- 摘要：Data stream classification is a crucial research area within data stream mining, with the core task of swiftly capturing concept drifts from real-time incoming data stream and promptly adjusting classification models. Extreme learning machine possesses advantages such as fast training speeds and excellent generalization performance. However, existing data stream classification methods based on extreme learning machine often struggle to simultaneously address common challenges in data stream, such as multi-class imbalance, concept drift, and the expensive labeling cost. For this reason, an imbalanced data stream classification with limited labels was proposed. We defined a sample prediction certainty measure that combined the difference in predicted probabilities and information entropy. An uncertainty label request strategy was introduced. Furthermore, we defined a sample importance measure based on class imbalance ratios and sample prediction errors. We also proposed an update and reconstruction mechanism for the classifier based on the concept drift index. Comparative experiments on six synthetic data streams and three real data streams demonstrate that the proposed method outperforms six existing data stream classification methods in terms of classification performance.关键词：multi-class imbalance;extreme learning machine;concept drift;expensive labeling cost158|358|0更新时间：2025-03-15
- 摘要：Time series prediction is widely used in many fields in modern life, and its importance has become increasingly prominent. Non-stationarity is one of the main issues affecting the accuracy of time series predictions. Due to the statistical characteristics of time series data changing over time, it is difficult to stably apply the patterns learned from historical data to future predictions, thus increasing the difficulty and uncertainty of predictions. In order to solve this problem, a novel framework called StabilizeNet was proposed，which was designed to reduce the non-stationarity of time series data. The framework consisted of three parts: centralization and scaling transformation, linear transformation, and reverse transformation. By introducing a learnable normalized linear transformation matrix, it optimized data information retention and enhanced the model’s ability to capture time series dynamics. Compared with advanced time series prediction models such as Informer, SCINet, Pyraformer, FEDformer, and Crossformer, StabilizeNet demonstrated effectiveness and superiority on multiple datasets. This framework provides a new preprocessing method for time series prediction, which helps to improve the prediction performance of time series predictive models.关键词：non-stationarity;normalization219|204|0更新时间：2025-03-15
- 摘要：At present, the top-level system of ownership governance in China needs to be systematized and improved, and problems such as an imperfect transaction mechanism and a low level of data circulation and sharing are identified, which hinder the marketization process of data elements and inhibit the full release of their value. Based on the above background, firstly, the theoretical basis of ownership governance for cross-domain data flow was explored from four perspectives: value network theory, ecological synergy theory, architect theory, and financial trust theory. Subsequently, a distributed data space architecture model based on the perspective of cross-domain collaboration for data authorization operators was innovatively proposed by combining distributed architecture theory with dynamic It is found that the architecture model can effectively guide the application of cross-domain scenarios so that different dimensions' data can generate large-scale gain value through superposition and optimize ownership allocation to improve orderliness in data market transactions.关键词：cross-platform data collaboration;distributed data space;data ownership governance265|353|0更新时间：2025-03-15
STUDY
- 摘要：Weather factors, as the primary factors affecting flight delays, have an important impact on flight delay prediction. Confronting the severe weather, multi-classification prediction of flight delay duration was made, and a Stacking-based integrated flight delay prediction model was proposed for the problems of low prediction accuracy and poor stability of traditional single model. Combining flight data and weather data features, multiple heterogeneous classifiers such as LightGBM and XGBoost were used as base learners, and SVM was used as the primary learner. A stacked, two-layer integrated learning framework was constructed. To verify the model validity, multiple single models were constructed for comparison with the integrated model. The experimental results demonstrate that the Stacking integrated prediction model has the best performance with an overall accuracy of 95.25% and an F1 score of 0.9527.关键词：Stacking ensemble learning;multi model fusion;severe weather197|874|0更新时间：2025-03-15
- 摘要：Ecological environment big data, as an important part of the new quality productivity, helps promoting efficient, scientific and precise environmental governance and realizing the intelligent transformation of environmental governance. However, the current application of ecological environment big data in China's environmental governance practice faces practical difficulties: firstly, the lack of legal regulations on data utilization hinders data utilization, leading to difficulties in data retrieval; secondly, the insufficient security technology guarantee for ecological environment big data results in data security risks, such as data distortion and data leakage; thirdly, the inadequate regulatory system for big data algorithms leds to algorithmic discrimination and disrupts China's environmental governance ecosystem. To solve the current problems, this paper starts from three aspects: strengthening legislation, technical support, and supervision and management. Combined with China's environmental governance practice, we need to continuously explore new paths for ecological environment big data governance in China through strengthening the construction of basic legal systems for data, enhancing privacy protection, researching digital security technologies such as blockchain, and improving algorithm supervision methods. This paper lays a solid foundation for achieving modernization of environmental governance.关键词：environmental governance;data security;blockchain;algorithmic supervision253|285|0更新时间：2025-03-15
APPLICATION
- 摘要：From the perspective of data trading practice, the pricing machanism for data trading is not mature.This study aims to explore the impact of "exclusive pricing" and "non-exclusive pricing" on all parties involved in transactions. By doing so, it not only enhances academic research on data transactions but also provides useful insights for data product providers when they make practical pricing strategies. An economic model was constructed, with data scale and data product price as the variables. Price discrimination and the mutual influence among various parties in the trading market were taken into account. A comparative analysis was conducted between the pricing strategies of “exclusive pricing” and “non-exclusive pricing” for data products. The study finds the providers should adjust their pricing strategies based on market change in the data product trading market. When the market is inactive, “exclusive pricing“ can deprive consumers of surplus; when the market is active, “non-exclusive pricing“ can generate higher profits due to its low marginal cost. At the same time the providers should prioritize product quality and meet market demand to achieve higher profits and promote the healthy development of the market. Ignoring market change, product quality and market demand will lead to profit loss and hinder market development.关键词：exclusive pricing;non-exclusive pricing;price discrimination280|220|0更新时间：2025-03-15
FORUM
- Postal code：100079
- Tel：（010）53879208、53859533 Email：bdr@bjxintong.com.cn
- Technical support is provided by Beijing Founder electronics co., LTD 京ICP备09064830号-19
京公网安备11010802024621 - It is recommended to read the content of this site in Chrome&IE9+. Please switch to extreme mode in browser 360.
- Cookies We use cookies to help provide and enhance our service and tailor content. By continuing, you agree to the use of cookies.
- Total Visits：273110 Visits Today：16
0