Recommender systems face the challenge of processing massive amounts of interaction data. The underutilization of relationship information in these data, as well as the presence of noise, reduces the accuracy of recommendations and the generalization ability of recommendation models. To address this, we propose KGDRec—a multi-graph knowledge-enhanced denoising recommendation algorithm. First, a relation-aware multi-graph fusion strategy integrates user-item interactions, user-user trust relationships, and item-item relationships to enhance semantic representations and alleviate data sparsity. However, while multi-graph fusion enriches relational information, it also introduces noise. To tackle this, we design a denoising self-enhancement module that combines edge perturbation, stability weighting, and contrastive learning to suppress unreliable signals and improve representation consistency. Experiments on the Ciao, Epinions, and Yelp datasets demonstrate that KGDRec outperforms baseline models in terms of Hit Ratio (HR) and Normalized Discounted Cumulative Gain (NDCG) metrics, and enhances accuracy and generalization capability.

The recommendation algorithm suggests products to users, improving their experience, however, it encounters a challenge of insufficient diversity in the recommended results. This paper proposes Product Path and Time decay enhanced Product-based Neural Network recommendation algorithm. Firstly, establishes three types of product paths: User Purchase History Path, Product Similarity Calculation Path, and Product Bundles Path, integrates them to form a comprehensive product relation network, thereby enhancing the diversity of the recommended results. Then, a time decay function is introduced to further improve recommendation accuracy of the recommended products. Finally, fuses the product path and time decay function as a new R component to the Product layer of the PNN model. Experimental results show that the Product Path and Time decay enhanced PNN model improves the AUC from 0.8605 to 0.8772 and reduces the cross-entropy loss from 0.2228 to 0.2155. Meanwhile, the intra-list diversity (ILD) increases from 0.8581 to 0.8832, and the entropy rises from 4.15 to 4.74, demonstrating superiority over the standard PNN model in both accuracy and recommendation diversity.

With the advancement of text summarization technology, the issue of hallucinations in summaries has garnered increasing attention. Pretrained models often incorporate additional factual information to minimize the occurrence of hallucinations. In research on text summarization using LLMs, accurate samples are typically provided via a chain-of-thought approach, enabling the model to learn the implicit relationship between the source text and the target summary. To address the hallucination problem in text summarization, this paper proposes CLSeq and Nscp, which are specifically designed for pre-trained models and LLMs, respectively. CLSeq integrates the strengths of human-generated summaries and model-generated summaries to produce high-quality positive samples as target, while refining the loss function to handle negative samples more effectively. Nscp provides negative examples and explanatory information through the chain-of-thought mechanism. These strategies aim to enhance the model’s understanding of the characteristics and causes of hallucinations, thereby reducing the likelihood of factual inconsistencies in the summaries. Experimental results demonstrate that both methods effectively mitigate the hallucination problem in text summarization and exhibit a certain degree of robustness.

Pre-training language models are increasingly being utilized for multi-document summarization (MDS) tasks. However, pre-training typically necessitates large-scale, domain-specific data. Most MDS pre-training models process multiple documents as a single document, while ignoring the subsection relationship information among different documents. In this paper, we focus on pre-training objectives for MDS, which assume key information appears across multiple documents that point to the same topic. We segment each document into several subsections based on text structure features. Then, we compare the subsections of different documents, extract key sentences from these subsections using text similarity, and generate a proxy summary. Experimental results on Multi-News and WikiSum demonstrate that our proposed model outperforms compared MDS models in terms of ROUGE scores and maintains strong performance even with limited data samples.

Document-level entity relation extraction aims to identify all potential relations between entity pairs from unstructured text. To address challenges such as multi-mention entities, long-distance entity relations, and complex relation reasoning, this paper proposes a three-stage document relation extraction model (TSDRE) based on text coreference resolution and relation reasoning. Specifically, we first employ the CorefQA model to resolve referential pronouns within the document. Next, the document is segmented into sentences, and the PRGC model is utilized to extract sentence-level relations. Finally, leveraging information on potentially complex relations between entity pairs, relation reasoning, and supporting evidence encodes this information and uses it as input for the BERT model to determine possible relation categories of entity pairs. Experimental results demonstrate that on the DocRED, TSDRE achieves improvements of at least 0.18, 0.15, and 0.21 over other models in terms of validation set F1, test set F1 and lgnF1, respectively. On the DWIE, TSDRE achieves improvements of at least 0.81 and 0.35 for the validation set lgnF1 and the test set lgnF1, respectively. Results indicates that TSDRE excels in accurately and comprehensively extracting entity pair correlation information from documents.

The basic requirement of text summarization is that the facts in the summary and the original text must be consistent. At present, most of the summarization models choose to introduce fact information in the decoding stage. With the increase of the text content, the ability to process fact information becomes weak, which leads to fact consistency errors in the model. From the perspective of data fusion in input section, this paper proposes a LTSum-FTL (Long Text Summarization model with Fact Triples Labeling) to improve factual consistency, which help readers obtain more accurate information. Firstly, use fact triples to represent the factual information of the original text. Then annotate the three attributes in the triple, and vector the annotation information and fuse into the input vector. Finally, use the improved masking mechanism to mask or replace the triple attributes of the input part, to improve the model summary ability. The experimental results show that the proposed model can effectively reduce the probability of fact consistency errors,it is at least 2.4%,1.1% and 0.3 higher than that of other comparison models in the metrics of Pre1,Pre2 and FactCC-Socre.

Since the outbreak of COVID-19, it has seriously endangered the health of human beings. Computer automatic segmentation of COVID-19 X-ray images is an important means to assist doctors in rapid and accurate diagnosis. Therefore, this paper proposes a modified FOA (EEFOA) with two optimization strategies added to the original FOA, including elite natural evolution (ENE) and elite random mutation (ERM). To be specific, ENE and ERM can effectively speed up the convergence and deal with the problem of local optima, respectively. The outstanding performance of EEFOA was confirmed by experimental results comparing EEFOA with the original FOA, other FOA variants, and advanced algorithms at CEC2014. After that, EEFOA is implemented for multi-threshold image segmentation (MIS) of COVID-19 X-ray images, where a 2D histogram consisting of the original greyscale image and the non-local means image is used to represent the image information, and Rényi's entropy is used as the objective function to find the maximum value. The evaluation results of the MIS segmentation experiments show that, whether high or low threshold, EEFOA can achieve higher quality segmentation results and greater robustness than other advanced segmentation methods.

The slime mould algorithm (SMA) is a population-based optimization algorithm that mimics the foraging behavior of slime moulds with a simple structure and few hyperparameters. However, SMA has some limitations, such as getting trapped in local optima when dealing with multimodal or combinatorial functions. To overcome these limitations and improve the algorithm's exploration and exploitation abilities, a local dimensional mutation strategy and an all-dimensional neighborhood search strategy for SMA, known as LASMA, were introduced. To evaluate the performance of LASMA, experiments were conducted on 30 benchmark functions from the CEC2014 competition, and the results were compared with up to 27 peers. The experimental results were then synthesized, and the Wilcoxon signed-rank test was used to evaluate the performance of LASMA. The results showed that LASMA outperformed other algorithms in terms of solution accuracy, stability, and convergence speed, with at least a 53.3% improvement in optimization performance on the 30 tested functions. Moreover, to demonstrate the applicability of LASMA to feature selection problems, a binary version of LASMA called bLASMA was developed and compared with eight binary classification algorithms on 18 datasets from the UCI repository. The experimental results showed that bLASMA not only had faster convergence speed and higher convergence accuracy in handling optimization problems but also performed well in feature selection applications. Thus, LASMA is a promising optimization tool for handling global and binary optimization problems, and its binary version, bLASMA, can be used for feature selection tasks. By addressing the limitations of SMA and improving the algorithm's exploration and exploitation abilities, LASMA provides a robust and effective solution for various optimization problems.

The traditional knowledge graph research field focuses on static knowledge such as entities and entity relationships. Events are dynamic and have the characteristics of actions, participants, and time and space. It is a coarse-grained way of knowledge representation. This paper first uses Python's xml.dom module to parse the marked corpus text in the Chinese emergency corpus CEC to obtain event semantic information, and then realizes the complete mapping of event semantic information to Neo4j graph database to store knowledge graphs, and finally designs and implements events A knowledge graph platform, which can construct the event knowledge in the database into an event knowledge graph, and complete the basic functions of adding, deleting, modifying, and checking event nodes and event relationships. The research in this paper can provide favorable support for event-oriented knowledge processing applications.

随着中医药学在国际医疗地位中的不断提高,中草药的使用广泛度也在增加.智能化的中草药图像识别可以进一步推进中医药学的现代化和国际化.传统卷积神经网络的参数量过多,需要大量计算资源,提出一种轻量型卷积神经网络中草药分类算法.新算法使用可分离卷积方式对原始图像与其梯度图的组合进行特征提取,在降低参数量的情况下提高分类的精度.

In recent years, due to the rise of deep learning, most researchers are trying to combine the traditional recommendation system with deep learning, however the traditional recommendation model is still suitable for most scenes or be widely used as an auxiliary model of deep learning recommendation model because of its strong interpretability, fast training and deployment. When the recommendation algorithm calculates the recommended goods for users, this paper establishes the category path of goods, then find the goods according to the category path of goods and recommend them. At the same time, according to the characteristics of the data obtained in this paper, the similarity of goods is calculated by the combination of TF-IDF and Jaccard similarity coefficient, so as to infer the list of goods recommended to users.

文章采用多目标规划方法,通过对铁路大型装车点选址模型进行假设,定义相关变量,以最小化成本和最大化客户满意度为目标,对模型进行构建,利用优先级与软件优化对模型进行求解,可以降低成本,推动铁路货运业务的发展.

为解决当前无线传感网流量优化算法传输效率不高、流量控制能力不强等问题,提出一种基于超参数区域匹配机制的无线传感网流量优化算法.首先,Sink节点按照剩余能力最大化原则优选核心控制节点,采用广播-反馈机制匹配全网符合要求的节点,能够迅速实现网络分割,提高网络流量初始优化能力.随后,网络综合能量剩余、拓扑距离等因素匹配备用节点,采用参数设计节点更新周期以降低因核心控制节点失效而导致网络传输抖动现象,最大限度匹配出具有较强传输能力的核心控制节点.最后,采用最小生成树方案对匹配出的核心控制节点实现快速组网,优化核心控制节点间的传输链路,稳定网络传输质量.仿真实验表明,所提算法与寿命感知聚类算法和能量模糊逻辑修正算法相比,具有流量控制强度高及最低传输能力较强等特点,具有较好的实际部署价值.

To break the bottleneck of computing speed, bandwidth, and power consumption of electronic computers, in recent years, as an emerging computer, the ternary optical computer (TOC) has attracted much attention because of its own mega-parallelism, and as the theory and practice of optical computing technology continue to develop and enrich, the exploration of applications based on ternary optical computer is also advancing. Logistics distribution center (LDC) problem is one of the core issues in the study of logistics networks. A sensible logistics center distribution can improve the efficiency of goods transportation and the operation efficiency of logistics enterprises. Traditional immune algorithms are prone to premature convergence and insufficient local search capability. By introducing a linear selection strategy, the algorithm maintains population diversity. The introduction of Gaussian and Cauchy mutation (G&C mutation) allows the algorithm to jump out of the local optimum. The experimental results show that the algorithm has higher solution quality and better robustness and can solve the logistics center location problem under the given constraints efficiently, and compared to electronic computers, TOC is more efficient.

针对文本分类中长短时记忆网络在提取局部信息时存在的局限性问题,本文提出了融合LSTM-Attention和CNN的文本分类模型.在结构上,首先使用LSTM提取全局序列信息,并通过注意力机制对LSTM输出添加权重,再通过三层卷积神经网络提取原文的局部信息.此外,在卷积神经网络中采取的是串行结构并且有选择性地将原始输入信息与CNN输出相融合.然后,将两者的输出信息进行组合以得到全新特征并用Softmax得到每个类别的概率,最终在THUCNews数据集上准确率达到96.8％.通过实验对比发现,本文提出的模型在文本分类任务中可以得到更好的分类效果.

The realms of modern medicine and biology have provided substantial data sets of genetic roots that exhibit a high dimensionality. Clinical practice and associated processes are primarily dependent on data-driven decisionmaking. However, the high dimensionality of the data in these domains increases the complexity and size of processing. It can be challenging to determine representative genes while reducing the data's dimensionality. A successful gene selection will serve to mitigate the computing costs and refine the accuracy of the classification by eliminating superfluous or duplicative features. To address this concern, this research suggests a wrapper gene selection approach based on the HGS, combined with a dispersed foraging strategy and a differential evolution strategy, to form a new algorithm named DDHGS. Introducing the DDHGS algorithm to the global optimization field and its binary derivative bDDHGS to the feature selection problem is anticipated to refine the existing search balance between explorative and exploitative cores. We assess and confirm the efficacy of our proposed method, DDHGS, by comparing it with DE and HGS combined with a single strategy, seven classic algorithms, and ten advanced algorithms on the IEEE CEC 2017 test suite. Furthermore, to further evaluate DDHGS' performance, we compare it with several CEC winners and DE-based techniques of great efficiency on 23 popular optimization functions and the IEEE CEC 2014 benchmark test suite. The experimentation asserted that the bDDHGS approach was able to surpass bHGS and a variety of existing methods when applied to fourteen feature selection datasets from the UCI repository. The metrics measured-classification accuracy, the number of selected features, fitness scores, and execution time-all showed marked improvements with the use of bDDHGS. Considering all results, it can be concluded that bDDHGS is an optimal optimizer and an effective feature selection tool in the wrapper mode.

Ternary optical computer (TOC) has become a research hotspot in the field because of the advantages such as inherent parallelism, numerous trits, low power consumption, extendibility, bitwise allocability and dynamical bitwise reconfigurability. Meanwhile, its performance evaluation attracts more and more attentions from potential users and researchers. To model its computing ecology more accurately, this paper first builds a three-staged TOC service model by introducing asynchronous multi-vacations and tandem queueing, and then proposes a task scheduling algorithm and an optical processor allocation algorithm with asynchronous vacations of some small optical processors after dividing equally the entire optical processor into several small optical processors which can be used independently. At the same time, the analytical model was established to obtain important performance indicators such as response time, the number of tasks and utilization of optical processor, based on M/M/1 and M/M/n queuing system with asynchronous multi-vacations. In addition, relevant numerical simulation experiments are conducted. The results illustrate that the number of small optical processors, vacation rate and the number of small optical processors allowed to be on vacation have important effects on the system performance. Compared with synchronous vacation, asynchronous vacation not only ensures the system to obtain better maintenance but also improves the system performance to some degree.

Human beings recognize and understand the real world in units of events. In recent years, events have been used as the basic unit to process unstructured text in the field of natural language processing, but there is often a connection between events and events. Therefore, recognizing the relationship between events and events in unstructured text has become an important task in the field of natural language processing and has attracted more and more researchers’ attention. This paper first introduces the evolution of the method of event temporal relation and causal relation in the extraction research, comparing the advantages and disadvantages and method performance; Then, the event relation extraction model based on deep learning can be divided into strong supervision method and weak supervision method, and the extraction methods of event relation are analyzed, compared and summarized respectively, among them, the method of strong supervision based on deep learning can be further divided into pipeline method and joint learning method, and the method of weak supervision based on deep learning can be divided into semi-supervised learning method, remote learning supervised method and unsupervised learning method. Finally, this paper summarizes the methods of event relation extraction and points out the future research direction.

Background : The objective of this research was to screen metabolites with specificity differences in the lung tissue of paraquat-poisoned rats by metabolomics technology and chi-square test method, to provide a theoretical basis for the study of the mechanisms of paraquat poisoning, and to use machine learning technology to construct a paraquat poisoning diagnosis model. This provided an intelligent decision-making method for the diagnosis of paraquat poisoning. Methods: 18 paraquat-poisoned rats (36 mg/kg) and 16 positive control rats were selected. Lung tissue from each rat from both groups was extracted and analyzed by GC-MS. The chi-square test for feature evaluation was used to screen the difference in specific metabolites in the lung tissue between the paraquat-poisoned rats and the control group, and the SVM classification machine learning algorithm was used to construct an intelligent diagnosis model. Results: In the end, a total of 14 significant metabolic differences were identified between the two groups (P < 0.05). The sensitivity, specificity, and accuracy of the constructed SVM paraquat poisoning diagnostic model reached 95%, 95% and 96.67%, respectively. Conclusion: Based on metabolomics technology, the chi-square test for feature evaluation was used to successfully screen the changes of specific metabolites produced in the lungs after paraquat- poisoning, and the diagnosis model based on SVM was constructed to provide an intelligent decision for the diagnosis of paraquat poisoning.