In medical image analysis, precise cell segmentation is crucial for understanding cell morphology and diagnosing diseases. Nevertheless, traditional models typically trained for specific modalities or cell types often fail to generalize to unknown categories. The Segment Anything Model (SAM) was introduced to address this but relies heavily on precise prompts and lacks integration of frequency domain features alongside multi-scale and multi-level information. To overcome these limitations, we propose Multi-scale Frequency and Boundary-enhanced Segment Any Cells (MFB-SAC), which includes a Multi-scale Frequency Convolution (MFC) module for providing multi-scale information and texture detail and a Multi-level Boundary Awareness (MBA) module to enhance boundary retention. Additionally, a Dual-representation Collaborative Attention (DCA) module optimizes feature integration. We tested MFB-SAC on 8 public datasets with various microscopy techniques and cell types. The results reveal that MFB-SAC outperforms the current general and benchmark cell segmentation models. The code is available at https://github.com/Mrliujunwen/SAC.

Relation extraction is one of the core tasks of natural language processing, which aims to identify entities in unstructured text and judge the semantic relationships between them. In the traditional methods, the extraction of rich features and the judgment of complex semantic relations are inadequate. Therefore, in this paper, we propose a relation extraction model, HAGCN, based on heterogeneous graph convolutional neural network and graph attention mechanism. We have constructed two different types of nodes, words and relations, in a heterogeneous graph convolutional neural network, which are used to extract different semantic types and attributes and further extract contextual semantic representations. By incorporating the graph attention mechanism to distinguish the importance of different information, and the model has stronger representation ability. In addition, an information update mechanism is designed in the model. Relation extraction is performed after iteratively fusing the node semantic information to obtain a more comprehensive node representation. The experimental results show that the HAGCN model achieves good relation extraction performance, and its F1 value reaches 91.51% in the SemEval-2010 Task 8 dataset. In addition, the HAGCN model also has good results in the WebNLG dataset, verifying the generalization ability of the model.

Vehicle GPS devices provide abundant trajectory data that can be exploited to generate helpful pick-up service recommendations for taxi drivers. However, existing trajectory clustering approaches struggle to perform well on trajectory data with different distribution characteristics (e.g., dense in downtown and discrete in suburbs) simultaneously. Additionally, current prediction models mainly focus on subsection prediction but fail to produce accurate multisection predictions. To this end, we propose a recommendation framework, namely TaxiGuider, that can generate accurate pick-up cluster recommendations for taxi drivers. First, historical pick-up points are extracted from the entire vehicle trajectory data after preprocessing. Then, a graph Laplacian-based multiple spatial-temporal clustering approach is presented to generate clusters that can effectively match the distribution of trajectory data. Furthermore, a pick-up frequency prediction model that employs a multi-head attention mechanism is proposed to produce accurate multisection predictions that can help taxi drivers make comprehensive considerations for their next destination. Finally, top$N$clusters with the highest predicted pick-up frequency are recommended to the target taxis according to their request. Experimental results on real-world datasets suggest that TaxiGuider outperforms state-of-the-art approaches in terms of both subsection and multisection predictions. Moreover, it produces pick-up cluster recommendations with superior prediction and classification accuracy simultaneously.

Accurate cell nuclei segmentation is crucial for characterizing cell morphology and elucidating disease types in medical image analysis. However, fully-supervised segmentation models often have limited generalization to unseen classes or domains due to being trained on specific modalities or cell types. The recent Segment Anything Model (SAM) has shown promise for interactive instance segmentation and zero-shot generalization. However, its effectiveness is hindered by the sparse and distributed nature of nuclei. It also requires a large number of user prompts that scale with the number of nuclei in the image. To overcome these challenges, we introduce Segment Any Nuclei (SAN), a novel foundation model tailored specifically for nuclei segmentation. SAN is trained on an extensive multi-modal dataset containing diverse nuclei instances from various imaging modalities, staining methods, and tissue types. Unlike previous SAM approaches that rely on manual prompts, SAN incorporates an innovative auto-prompting auxiliary segmentation network, which enables the model to make predictions without manual intervention while still allowing for manual interaction when needed. We evaluate SAN on a large dataset of 9,244 images and demonstrate state-of-the-art nuclei segmentation performance, surpassing both fully-supervised approaches and other SAM-like models. SAN represents a step towards more generalizable, efficient and interactive nuclei segmentation. Our code is available at https://github.com/Mrliujunwen/SAN.

Biomedical event detection is a pivotal information extraction task in molecular biology and biomedical research, which provides inspiration for the medical search, disease prevention, and new drug development. The existing methods usually detect simple biomedical events and complex events with the same model, and the performance of the complex biomedical event extraction is relatively low. In this paper, we build different neural networks for simple and complex events respectively, which helps to promote the performance of complex event extraction. To avoid redundant information, we design dynamic path planning strategy for argument detection. To take full use of the information between the trigger identification and argument detection subtasks, and reduce the cascading errors, we build a joint event extraction model. Experimental results demonstrate our approach achieves the best F-score on the biomedical benchmark MLEE dataset and outperforms the recent state-of-the-art methods.

Long-tail service recommendation provides an unexpected but reasonable experience for potential developers when they construct mashups. However, the lack of available information makes it difficult to recommend highly relevant long-tail services for target mashups. Collaborative tagging systems employ extensive tag records to replenish the available information of long-tail services, whereas existing tag-aware approaches are unable to learn multi-aspect embeddings from graphs with different structures and relationships for long-tail services. To this end, we present a novel approach, namely collaborative tag-aware graph neural network, to recommend satisfactory long-tail services by extracting multi-aspect embeddings. Firstly, a tensor decomposition is executed to parameterize mashups, tags, and services as low-dimensional vector representations, respectively. Then, an interaction-aware heterogeneous neighbor aggregation is presented to aggregate both neighboring node features and interaction strength to enhance the embedding quality of long-tail services. Next, a diffusion-aware homogeneous neighbor aggregation is proposed to assign higher weights for long-tail neighboring nodes so as to reduce the influence of popular neighboring nodes during the aggregation process. Furthermore, a type-aware attention network is employed to update the final node embedding by aggregating multi-aspect embeddings. Experimental results on two real-world Web service datasets indicate that the proposed approach generates superior accuracy and diversity than state-of-the-art approaches in the aspect of long-tail service recommendation.

Biomedical event extraction is an important branch of biomedical information extraction. Event detection is the most important subtask in event extraction, which has been widely concerned. Most of the existing research on event detection is based on traditional machine learning or neural network. However, they ignored the semantic information of the word itself and its event type and the insufficient features of the out-of-vocabulary neologism representation. In this paper, trigger extraction is treated as a sequence labeling problem. We propose a biomedical event detection model based on knowledge injection and model dual channel fine-tuning, which introduces an external biomedical knowledge base, UMLS. This architecture improves our model's ability to capture semantic information about the word itself and its event types, as well as information about out-of-vocabulary neologisms. The experimental results show that the proposed model improves the performance of biomedical event detection, and the F1 value on the MLEE dataset is 83.59%, which outperforms the recent state-of-the-art methods. Moreover, testing our model on GE13, the experimental results are also significantly improved.

Facial Expression Recognition (FER) is a basic and crucial computer vision task of classifying emotional expressions from human faces images into various emotion categories such as happy, sad, surprised, scared, angry, etc. Recently, facial expression recognition based on deep learning has made great progress. However, no matter the weight initialization technology or the attention mechanism, the face recognition method based on deep learning hard to capture those visually insignificant but semantically important features. To aid above question, in this paper we present a novel Facial Expression Recognition training strategy consisting of two components: Memo Affinity Loss (MAL) and Mask Attention Fine Tuning (MAFT). MAL is a variant of center loss, which uses memory bank strategy as well as discriminative center. MAL widens the distance between different clusters and narrows the distance within each cluster. Therefore, the features extracted by CNN were comprehensive and independent, which produced a more robust model. MAFT is a strategy that blindfolds attention parts temporarily and forces the model to learn from other important regions of the input image. It's not only an augmenting technique, but also a novel fine-tuning approach. As we know, we are the first to apply the mask strategy to the attention part and use this strategy to fine-tune the models. Finally, to implement our ideas, we constructed a new network named Architecture Attention ResNet based on ResNet-18. Our methods are conceptually and practically simple, but receives superior results on popular public facial expression recognition benchmarks with 88.75% on RAF-DB, 65.17% on AffectNet-7, 60.72% on AffectNet-8. The code will open source soon.

BACKGROUND: Biomedical event extraction is a fundamental task in biomedical text mining, which provides inspiration for medicine research and disease prevention. Biomedical events include simple events and complex events. Existing biomedical event extraction methods usually deal with simple events and complex events uniformly, and the performance of complex event extraction is relatively low. RESULTS: In this paper, we propose a fine-grained Bidirectional Long Short Term Memory method for biomedical event extraction, which designs different argument detection models for simple and complex events respectively. In addition, multi-level attention is designed to improve the performance of complex event extraction, and sentence embeddings are integrated to obtain sentence level information which can resolve the ambiguities for some types of events. Our method achieves state-of-the-art performance on the commonly used dataset Multi-Level Event Extraction. CONCLUSIONS: The sentence embeddings enrich the global sentence-level information. The fine-grained argument detection model improves the performance of complex biomedical event extraction. Furthermore, the multi-level attention mechanism enhances the interactions among relevant arguments. The experimental results demonstrate the effectiveness of the proposed method for biomedical event extraction.

Recommender system (RS) suggests relevant objects to generate personalized service and minimize information overload issue. User-based collaborative filtering (UBCF) plays a dominant role in practical RSs. However, traditional UBCF suffers from a recommendation overfitting problem, i.e., recommendations generated by UBCF usually concentrate on popular items, resulting in lower diversity. In addition, UBCF cannot maintain a reasonable tradeoff between the accuracy and diversity of recommendations because raising the diversity is often accompanied by a decrease in accuracy. In this article, we propose a novel approach, namely link-based collaborative filtering, to enhance the recommendation accuracy and diversity simultaneously without employing additional complex information. First, a user–item bipartite network is constructed based on the user–item rating matrix of RSs. Then, a global–local weighted bipartite modularity is presented to conduct link partition so that links with the same community can not only be relatively denser but also own the same characteristic. Furthermore, redundant links are removed from each community by utilizing a link reduction algorithm so that neighborhood of a target user can be selected according to the more efficient nonredundant links. Finally, rating prediction is executed based on the rating information of neighborhood. Also, items owning the highest predicted rating scores will be recommended to the target user. Experimental results from three real datasets of RSs suggest that, without taking advantage of special additional data, our proposed approach outperforms the state-of-the-art studies and is able to generate personalized recommendations with satisfying accuracy and diversity simultaneously.

With the development of social media sites, user credit grading, which served as an important and fashionable problem, has attracted substantial attention from a slew of developers and operators of mobile applications. In particular, multi-grades of user credit aimed to achieve (1) anomaly detection and risk early warning and (2) personalized information and service recommendation for privileged users. The above two goals still remained as up-to-date challenges. To these ends, in this article, we propose a novel regression-based method. Technically speaking, we define three natural ordered categories including BlockList , GeneralList , and AllowList according to users’ registration and behavior information, which preserve both the global hierarchical relationship of user credit and the local coincident features of users, and hence formulate user credit grading as the ordinal regression problem. Our method is inspired by KDLOR ( kernel discriminant learning for ordinal regression ), which is an effective and efficient model to solve ordinal regression by mapping high-dimension samples to the discriminant region with supervised conditions. However, the performance of KDLOR is fragile to the extreme imbalanced distribution of users. To address this problem, we propose a robust sampling model to balance distribution and avoid overfit or underfit learning, which induces the triplet metric constraint to obtain hard negative samples that well represent the latent ordered class information. A step further, another salient problem lies in ambiguous samples that are noises or located in the classification boundary to impede optimized mapping and embedding. To this problem, we improve sampling by identifying and evading noises in triplets to obtain hard negative samples to enhance robustness and effectiveness for ordinal regression. We organized training and testing datasets for user credit grading by selecting limited items from real-life huge tables of users in the mobile application, which are used in similar problems; moreover, we theoretically and empirically demonstrate the advantages of the proposed model over established datasets.

Abstract User-based collaborative filtering (UBCF) is widely used in recommender systems (RSs) as one of the most successful approaches, but traditional UBCF cannot provide recommendations with satisfactory accuracy and diversity simultaneously. Covering-based collaborative filtering (CBCF) is a useful approach that we have proposed in our previous work, which greatly improves the traditional UBCF and could provide satisfactory recommendations to an active user which often has sufficient rating information. However, different from an active user, a new user in RSs often has special characteristics (e.g., fewer ratings or ratings concentrating on popular items), and the previous CBCF approach cannot provide satisfactory recommendations for a new user. In this paper, aiming to provide personalized recommendations for a new user, through a detailed analysis of the characteristics of new users, we reconstruct a decision class to improve the previous CBCF and utilize the covering reduction algorithm in covering-based rough sets to remove redundant candidate neighbors for a new user. Furthermore, unlike the previous CBCF, our improved CBCF could provide personalized recommendations without needing special additional information. Experimental results suggest that for the sparse datasets that often occur in real RSs, the improved CBCF significantly outperforms those of existing work and can provide personalized recommendations for a new user with satisfactory accuracy and diversity simultaneously.

This paper studies a searching problem in an unknown street. A simple polygon [Formula: see text] with two distinguished vertices, [Formula: see text] and [Formula: see text], is called a street if the two boundary chains from [Formula: see text] to [Formula: see text] are mutually weakly visible. We use a mobile robot to locate [Formula: see text] starting from [Formula: see text]. Assume that the robot has a limited sensing capability that can only detect the constructed edges (also called gaps) on the boundary of its visible region, but cannot measure any angle or distance. The robot does not have knowledge of the street in advance. We present a new competitive strategy for this problem and prove that the length of the path generated by the robot is at most 9-times longer than the shortest path. We also propose a matching lower bound to show that our strategy is optimal. Compared with the previous strategy, we further relaxed the restriction that the robot should take a marking device and use the data structure S-GNT. The analysis of our strategy is tight.

Inter-transactional association rules mining is mainly used in mining significant association between different transaction, but the existing algorithm only focus on the efficiency or accuracy. In this study, we propose the inter-transactional association rules algorithm based on cluster and dual-strategy analysis model. The algorithm adopts dual-strategy interest model to judge the integrity of the inter-transactional association rules, make up for mining bugs, avoid the generation of false rules, improve the quality of the mining algorithm; And use of cluster analysis to remove a large number of redundant data in database, improve the efficiency of the algorithm. The experimental results show that the proposed algorithm improves accuracy and efficiency of inter-transactional association rules algorithm.

Clustering algorithms are the core technique of data mining, machine learning, pattern matching, bioinformatics and a number of other fields. This paper proposes a new clustering method based on attribute partitioning and a novel data visualization method. In a nutshell, the idea for our method is based on two steps: 1) cluster data set using primary and secondary attributes of data; 2) map color stimulus spectrum to RGB color space and visualize clustering using the chromaticity diagram of J.C. Maxwell (Maxwell's triangle). The experiments show that the algorithm is very efficient. In addition it is simple and easy to implement. Our visualization algorithm aims at helping the user to get an overview of data as well as in prediction and decision making processes.