Information Processing & Management最新文献

With the growing prominence of interdisciplinary research and heightened concerns surrounding extended prepublication timelines, we still lack of understanding regarding the interplay between interdisciplinary level and the duration of the peer review process. This study aims to untangle the relationship between interdisciplinarity and the time manuscripts spend navigating the peer review phase. Leveraging a large-scale bibliometric dataset comprising over three million journal articles, we uncover a robust positive association between paper interdisciplinary level and the duration of the peer review process. This relationship persists across diverse fields, journals with various impacts, and articles with different citation impacts. Moreover, we find that conventionality and reference age partly contribute to such delay. Notably, our investigation indicates that journal editors cannot fully account for the prolonged peer review delays for interdisciplinary research. Furthermore, our results underscore a noteworthy observation: referees generally pose more inquiries toward interdisciplinary endeavors. This is consistent with the fact that scientists submitting interdisciplinary manuscripts may inherently require additional time to adequately address the detailed comments and questions posed by referees. Our results have policy implications for funders, journal editors, and institutions seeking to promote and facilitate interdisciplinary research.

The identification of influential simplices is crucial for understanding higher-order network dynamics. Yet, despite relatively mature research on influential nodes (0-simplices) mining, characterizing simplices’ influence and identifying influential simplices remain challenging due to notable discrepancies in vital nodes and vital simplices mining. In this paper, we propose a higher-order graph learning model, named influential simplices mining neural networks (ISMnet), to identify vital simplices in simplicial complexes. ISMnet leverages novel higher-order representations: hierarchical bipartite graphs and higher-order hierarchical (HoH) Laplacians, where target simplices are grouped into a hub set and can interact with other simplices. It also employs learnable graph convolution operators in each HoH Laplacian domain to capture interactions among simplices and can identify influential simplices of arbitrary order by changing the hub set. Notably, ISMnet addresses the limitations inherent in traditional graph neural networks that struggle with higher-order tasks, while seamlessly retaining the capability to exploit network topology and node features concurrently. Numerical results on empirical and synthetic datasets demonstrate that ISMnet significantly outperforms existing methods by at least 12% and 4%, respectively, in ranking 2-simplices. In general, this novel framework promises to serve as a potent tool in higher-order network analysis.

Knowledge base question answering aims to answer complex questions from large-scale knowledge bases. Although existing generative language models that translate questions into SPARQL queries have achieved promising results, there are still generation errors due to redundancies or errors in the knowledge fed to the generative models and difficulties in representing the implicit logic of knowledge as the specific syntax of SPARQL. To address above issues, we propose TrackerQA, a novel self-supervised reasoning framework based on basic graph patterns (BGP) to determine precise paths and enhance SPARQL generation. First, we develop a contrastive learning semantic matching model to reduce the large knowledge searching space. Then, we built a BGP parser that parses the recalled knowledge and constraints into BGP graphs, which can deconstruct complex knowledge into BGP triples and naturally obtain supervision from gold SPARQL. Next, we design a self-supervised BGP graph neural network that encodes knowledge through graph transformation layers with directed message-passing control and employs a question-aware attention mechanism to predict the exact BGP paths. Finally, a SPARQL generator integrates the paths into a pre-trained language model to improve the performance of SPARQL generation. Experiments on the KQA Pro dataset show that our model achieves state-of-the-art answering accuracy scores of 95.32%, being the closest to the human level at 97.5%, and reasons out KB paths with F1 scores of 0.98 for nodes and 0.99 for edges.

Network alignment (NA), discovering anchor nodes that represent the same entities across different networks, plays a fundamental role in information fusion. Most existing embedding-based methods rarely study the alignment module, which learns a global mapping to unify embedding spaces. However, global mapping is a holistic solution for nodes, incapable of optimally projecting each node, thus deteriorating alignment accuracy. To solve this problem, this paper proposes a local mapping framework named MANA, which fine-tunes global mapping by meta-learning to obtain node-level local mappings. One advantage of MANA is that it tailors local mapping adapted to the embedded locality of each node while maintaining the general knowledge of global mapping. The main challenge of applying meta-learning to network alignment is paradigm incompatibility; that is, how to construct effective meta-tasks and support sets for zero-shot NA. Therefore, the paper constructs meta-tasks with similarity-based support sets. A support set is taken from pairs of anchor nodes, the source nodes of which are embedded close to the query node. Our framework can be applied to existing mapping-based NA models. Experimental results show that mapping-based models with MANA improve evaluation scores by 1%–59% relative to their original models, demonstrating the effectiveness of local mapping. Some simple mapping-based models improved by MANA even outperform sophisticated sharing-based NA approaches.

The identification of sexual activities in images can be helpful in detecting the level of content severity and can assist pornography detectors in filtering specific types of content. In this paper, we propose a Deep Learning-based framework, named DeepHSAR, for semi-supervised fine-grained multi-label Human Sexual Activity Recognition (HSAR). To the best of our knowledge, this is the first work to propose an approach to HSAR. We also introduce a new multi-label dataset, named SexualActs-150k, containing 150k images manually labeled with 19 types of sexual activities. DeepHSAR has two multi-label classification streams: one for global image representation and another for fine-grained representation. To perform fine-grained image classification without ground-truth bounding box annotations, we propose a novel semi-supervised approach for multi-label fine-grained recognition, which learns through an iterative clustering and iterative CNN training process. We obtained a significant performance gain by fusing both streams (i.e., overall F1-score of 79.29%), compared to when they work separately. The experiments demonstrate that the proposed framework explicitly outperforms baseline and state-of-the-art approaches. In addition, the proposed framework also obtains state-of-the-art or competitive results in semi-supervised multi-label learning experiments on the NUS-WIDE and MS-COCO datasets with overall F1-scores of 75.98% and 85.17%, respectively. Furthermore, the proposed DeepHSAR has been assessed on the NPDI Pornography-2k video dataset, achieving a new state-of-the-art with 99.85% accuracy.

Network entities have important asset mapping, vulnerability, and service delivery applications. In cyberspace, where the network structure is complex and the number of entities is large, effectively obtaining the relevant attributes of entities is a difficult task. Graph neural network-based approaches focus on target IP node messaging from neighboring nodes; however, the graph learning task ignores the heterophilous relationship of network entity identification (NEI) tasks in the graph structure and fails to effectively message from non-neighboring nodes. To address the limitations of the existing task, we propose a NEI model based on heterophilous graph learning (HpGraphNEI); HpGraphNEI converts heterophilous graphs under the NEI task into homophilous graphs and uses the graph learning mechanism to carry out attribute completion task for incomplete entity attributes. First, the acquired dataset is feature-extracted by network measurement, and the clustering algorithm is employed to divide the target nodes into communities. Second, the network topology graph is constructed to embed the node attribute information and neighborhood structure information into the graph in the form of feature vectors. Then, the global attention in the community is calculated according to the attention results, the edges with strong correlation in the network are filtered, the adjacency matrix is reconstructed, and then the updated node information is aggregated to complete the incomplete attribute completion. Fourth, the updated nodes are categorized to output network entity categories and construct network entity portraits based on the attribute completion nodes. We conducted a 2-month data collection in three real regions and successfully identified 6 types of network entities. Compared with the optimal baseline, all the metrics have significantly improved, with NEI accuracy above 93.74% and up to 96.28%, improved 2.27% to 2.69%.

The personalized bundle generation problem, which aims to create a preferred bundle for user from numerous candidate items, receives increasing attention in recommendation. However, existing works ignore the order-invariant nature of the bundle and adopt sequential modeling methods as the solution, which might introduce inductive bias and cause a large latency in prediction. To address this problem, we propose to perform the bundle generation via non-autoregressive mechanism and design a novel encoder–decoder framework named BundleNAT, which can effectively output the targeted bundle in one-shot without relying on any inherent order. In detail, instead of learning sequential dependency, we propose to adopt pre-training techniques and graph neural network to fully embed user-based preference and item-based compatibility information, and use a self-attention based encoder to further extract global dependency pattern. We then design a permutation-equivariant decoding architecture that is able to directly output the desired bundle in a one-shot manner. Experiments on three real-world datasets from Youshu and Netease show the proposed BundleNAT significantly outperforms the current state-of-the-art methods in average by up to 35.92%, 10.97% and 23.67% absolute improvements in Precision, Precision+, and Recall, respectively.

The task of event relation extraction (ERE) aims to organize multiple events and their relations as a directed graph. However, existing ERE methods exhibit two limitations: (1) Events in a document are typically expressed with merely a trigger word or phrase neglecting the rich semantic structure of event arguments and roles. (2) Various event relations such as coreference, temporal, causal, and subevent relations are correlated with each other, while little work has explored the benefits of extracting all relations in a joint manner. In this work, we investigate an event-relation graph propagation network and a novel document-level semantically-enriched event representation for joint ERE. First, we address the lack of event argument annotations by extracting cross-sentence implicit arguments based on the AMR graph. Then triggers and argument roles are aggregated with a structure-aware encoder. Second, based on the rich event information, event relation interactions are incorporated with a subgraph propagation and aggregation mechanism. During training, we further develop a novel triadic contrastive loss to capture high-order event pair relationships. We conduct experiments based on the public MAVEN-ERE benchmark and the results show that our model achieves 60.7%, 37.4%, 32.9% F1-scores on temporal, causal and subevent relations, and 86.1% MUC-score on coreferences, outperforming the current joint models by a large margin. Further in-depth analysis shows the effectiveness of our model in capturing event-event dependencies in document context. The proposed model can be used for event graph construction and storyline understanding.

This study presents a financial distress prediction model focusing on the linguistic analysis of risk-related sections of corporate annual reports. Here, we introduce a novel methodology that leverages BERT-based contextualized embedding models for nuanced extraction of financial sentiment and topic coherence. This stands in contrast to existing research, which predominantly relies on dictionary-based or non-contextual word embeddings and addresses their limitations in context sensitivity. Furthermore, we apply an innovative financial distress prediction model that combines the robust XGBoost algorithm with unsupervised outlier detection techniques. This hybrid model is specifically designed to tackle the issue of class imbalance, a persistent challenge in financial distress prediction. The efficacy of the proposed model is empirically validated using a comprehensive dataset of 2545 companies listed on major global stock exchanges. Our findings indicate that the introduced model not only significantly outperforms most existing state-of-the-art financial distress prediction models in terms of predictive accuracy, but also significantly outperforms the Loughran & McDonald dictionary-based approach and the Word2Vec model, underlining its potential as a superior analytical tool for financial distress prediction.

Large language models (LLMs) exhibit impressive capabilities across diverse tasks in natural language processing. Nevertheless, challenges arise such as large model parameter size and limited model accessibility through APIs such as ChatGPT and GPT-4, which prohibits the model deployment on mobile devices and domain adaptation or fine-tuning. Moreover, while LLMs excel in general domains, their performance in specialized fields such as GIS may not always align with the expectations of domain experts. This is primarily attributed to the diverse disciplinary origins of the training data, which often lack comprehensive coverage and treatment of knowledge specific to individual disciplines (e.g., GIS). Therefore, there is a crucial need to train and adapt LLMs specifically designed for different professional fields. In this paper, our focus is on the GIS domain, where we introduce BB(BaBy)-GeoGPT, a large language model with GIS-specific knowledge. To achieve this goal, we curated a comprehensive set of resources, comprising model pretraining data (BB-GeoPT, 26,907 documents), supervised fine-tuning data (BB-GeoSFT, 35,876 instructions), and evaluation data (BB-GeoEval, 600 objective questions and 150 subjective questions). BB-GeoGPT is developed by first adapting an open-source general-domain LLM, the LLaMA-2-7B model, to our pretraining data. Subsequently, we use instruction tuning to further fine-tune the model on our BB-GeoSFT. Through extensive experiments on the evaluation dataset, BB-GeoGPT demonstrates improvements ranging from 10.55% to 47.57% for objective questions and from 7.87% to 27.73% for subjective questions, when compared to general LLMs of similar size in terms of accuracy. Moreover, our data collection strategy and the amassed data can serve as a foundation for advancing LLM research in the GIS domain, fostering further development.