Journal of Intelligent Information Systems最新文献

The cold-start problem is a long-standing problem in recommender systems, i.e., lack of historical interaction information hinders effective recommendations for new users and items. Existing methods typically incorporate attribute information of users and items to address the strict cold-start problem. Most existing recommendation methods overlook the sparsity of user attributes in cold start recommendation systems. In this paper, we develop a novel framework, Graph Attention Networks with Adaptive Neighbor Graph Aggregation for cold-start Recommendation (A-GAR), which utilizes the user/item relationship information in cold-start recommendation systems to alleviate the sparsity of attributes. we can achieve more accurate recommendations in cold-start scenarios by fully exploring the complex relations between users/items using graph structures. Specifically, to learn the complex relationships between user/item attributes, we utilize SENet (Squeeze and Excitation Network) and MLP (Multilayer Perceptron) networks to adaptively fuse the embeddings of user/item and their second-order interaction vectors, achieving high-order feature aggregation. To address the issue of lacking preference information in cold-start recommendations, we extend the variational autoencoder to reconstruct missing user preferences (item characteristics) from higher-order attribute features of users/items. In order to learn the potential semantic relationships of nodes in the neighbor graph structure, an attribute graph attention network is used to aggregate the neighbor information of users and the interaction information between neighbors. In this way, the high-order relationships between nodes and the potential semantics of adjacent graphs can be fully explored. Extensive experiments on three real-word datasets with various cold-start scenarios demonstrate that A-GAR yields significant improvements for strict cold-start recommendations.

Predictive Process Monitoring (PPM) is a field of Process Mining that aims at predicting how an ongoing execution of a business process will develop in the future using past process executions recorded in event logs. The recent stream of publications in this field shows the need for tools able to support researchers and users in comparing and selecting the techniques that are the most suitable for them. In this paper, we present Nirdizati , a dedicated tool for supporting users in building, comparing and explaining the PPM models that can then be used to perform predictions on the future of an ongoing case. Nirdizati has been constructed by carefully considering the necessary capabilities of a PPM tool and by implementing them in a client-server architecture able to support modularity and scalability. The features of Nirdizati support researchers and practitioners within the entire pipeline for constructing reliable PPM models. The assessment using reactive design patterns and load tests provides an evaluation of the interaction among the architectural elements, and of the scalability with multiple users accessing the prototype in a concurrent manner, respectively. By providing a rich set of different state-of-the-art approaches, Nirdizati offers to Process Mining researchers and practitioners a useful and flexible instrument for comparing and selecting PPM techniques.

Group recommendation systems are widely applied in social media, e-commerce, and diverse platforms. These systems face challenges associated with data privacy constraints and protection regulations, impeding the sharing of user data for model improvement. To address the issue of data silos, federated learning emerges as a viable solution. However, difficulties arise due to the non-independent and non-identically distributed nature of data across different platforms, affecting performance. Furthermore, conventional federated learning often overlooks individual differences among stakeholders. In response to these challenges, we propose a pioneering cross-platform federated group recommendation system named FedGR. FedGR integrates hypergraph convolution, attention aggregation, and fully connected fusion components with federated learning to ensure exceptional model performance while preserving the confidentiality of private data. Additionally, we introduce a novel federated model aggregation strategy that prioritizes models with high training effectiveness, thereby improving overall model performance. To address individual differences, we design a temporal personalization update strategy for updating item representations, allowing local models to focus more on their individual characteristics. To evaluate FedGR, we apply our approach to three real-world datasets, demonstrating the robust capabilities of our cross-platform group recommendation system.

The COVID-19 Numerical Claims Open Research Dataset (CONCORD) is a comprehensive, open-source dataset that extracts numerical claims from academic papers on COVID-19 research. A weak-supervision model is employed for this extraction, taking advantage of its white-box, explainable nature and reduced computational and annotation costs compared to transformer-based models. This model uses labelling functions such as pattern matching, external knowledge bases, phrase matching, and third-party models to generate labels, with an aggregator function handling contradictory labels. Evaluated against established baselines, the model achieved a weighted F1-score of 0.932 and a micro F1-score of 0.930. While transformer-based models achieve comparable results, the explainability of weak-supervision offers distinct advantages. Additionally, generative LLMs were tested to understand their effectiveness in extracting numerical claims, highlighting the impact of prompt engineering on performance. CONCORD contains approximately 200,000 numerical claims from over 57,000 COVID-19 research articles, serving as a valuable resource for tracking developments in COVID-19 research. This dataset, coupled with the weak-supervision approach, provides researchers with a significant tool for advancing COVID-19 research and showcases the potential of these methodologies in the broader biomedical field.

Pre-training-based methods are considered some of the most advanced techniques in natural language processing tasks, particularly in text classification. However, these methods often overlook global semantic information. In contrast, traditional graph learning methods focus solely on structured information from text to graph, neglecting the hidden local information within the syntactic structure of the text. When combined, these approaches may introduce new noise and training biases. To tackle these challenges, we introduce DA-BAG, a novel approach that co-trains BERT and graph convolution models. Utilizing a self-domain adversarial training method on a single dataset, DA-BAG extracts multi-domain distribution features across multiple models, enabling self-adversarial domain adaptation training without the need for additional data, thereby enhancing model generalization and robustness. Furthermore, by incorporating an attention mechanism in multiple models, DA-BAG effectively combines the structural semantics of the graph with the token-level semantics of the pre-trained model, leveraging hidden information within the text’s syntactic structure. Additionally, a sequential multi-layer graph convolutional neural(GCN) connection structure based on a residual pre-activation variant is employed to stabilize the feature distribution of graph data and adjust the graph data structure accordingly. Extensive evaluations on 5 datasets(20NG, R8, R52, Ohsumed, MR) demonstrate that DA-BAG achieves state-of-the-art performance across a diverse range of datasets.

Recommendation systems play a major role in modern music streaming platforms, assisting consumers in finding new music that suits their tastes. However, a significant challenge persists when it comes to recommending new songs that lack historical data. This study introduces a Content based Attentive Sequential Recommendation Model (CASRM) that deals with item cold start issue and recommends relevant and fresh music using Gated Graph Neural Networks (GNNs). Music metadata such as artists, albums, genres, and tags are included in the content information, along with context data incorporating user behaviour such as sessions, listening logs, and music playing sequences. By representing the music data as a graph, we can effectively capture the intricate relationships between songs and users. To capture users’ music preferences, we analyse their interactions with songs within the sessions. We incorporate content-based item embeddings for newly added items, enabling personalized recommendations for new songs based on their characteristics and similarities to the songs listened by users in the past. Specifically, we examined the proposed model on three distinct datasets, and the experimental outcomes show its efficacy in predicting music ratings for new songs. Compared to other baseline methods, the CASRM model achieves superior performance in providing accurate and diverse music recommendations in cold-start scenarios.

Bipolar disorder is a disorder in which a person expresses manic and depressed emotions repeatedly. Diagnosing bipolar disorder accurately can be difficult because other mood disorders or even regular mood changes may have similar symptoms. Therefore, psychiatrists need to spend a long time observing and interviewing clients to make the diagnosis. Recent studies have trained machine learning models for detecting bipolar disorder on social media. However, most of these studies focused on increasing the accuracy of the model without explaining the classification results. Although the posts of a bipolar disorder user can be observed manually, doing so is not practical since a user can have many posts which may not depict any signs of bipolar disorder. Without any explanations, the trustworthiness of the model decreases. We propose a deep learning model that not only detects and classifies bipolar disorder users but also explains how the model generates the classification results. The posts are first grouped using Latent Dirichlet Allocation, a method commonly used to classify the topic of a text. These groups are then input into the model, and attention mechanisms are utilized to determine which groups have more attention weights and are considered more heavily. Finally, an explanation of the classification results is obtained by visualizing the attention weights. Several case studies are presented to demonstrate the explanations generated through our proposed model. Our model is also compared to other models, achieving the best performance with an F1-Score of 0.92.

Forest tree dieback inventory has a crucial role in improving forest management strategies. This inventory is traditionally performed by forests through laborious and time-consuming human assessment of individual trees. On the other hand, the large amount of Earth satellite data that are publicly available with the Copernicus program and can be processed through advanced deep learning techniques has recently been established as an alternative to field surveys for forest tree dieback tasks. However, to realize its full potential, deep learning requires a deep understanding of satellite data since the data collection and preparation steps are essential as the model development step. In this study, we explore the performance of a data-centric semantic segmentation approach to detect forest tree dieback events due to bark beetle infestation in satellite images. The proposed approach prepares a multisensor data set collected using both the SAR Sentinel-1 sensor and the optical Sentinel-2 sensor and uses this dataset to train a multisensor semantic segmentation model. The evaluation shows the effectiveness of the proposed approach in a real inventory case study that regards non-overlapping forest scenes from the Northeast of France acquired in October 2018. The selected scenes host bark beetle infestation hotspots of different sizes, which originate from the mass reproduction of the bark beetle in the 2018 infestation.

Crowdfunding has evolved into a formidable mechanism for collective financing, challenging traditional funding sources such as bank loans, venture capital, and private equity with its global reach and versatile applications across various sectors. This paper explores the complex dynamics of crowdfunding platforms, particularly focusing on investor behaviour and investment patterns within equity and lending campaigns in Italy. By leveraging advanced machine learning techniques, including XGBoost and LSTM networks, we develop predictive models that dynamically analyze real-time and historical data to accurately forecast the success or failure of crowdfunding campaigns. To address the existing gaps in crowdfunding analysis tools, we introduce two novel datasets—one for equity crowdfunding and another for lending. Moreover, our approach extends beyond traditional binary success metrics, proposing novel measures. The insights gained from this study could support crowdfunding strategies, significantly improving project selection and promotional tactics on platforms. By enhancing decision-making processes and providing forward-looking guidance to investors, our computational model aims to empower both campaign creators and platform administrators, ultimately improving the overall efficacy and sustainability of crowdfunding as a financing tool.

Lexical Simplification (LS) is the task of substituting complex words within a sentence for simpler alternatives while maintaining the sentence’s original meaning. LS is the lexical component of Text Simplification (TS) systems with the aim of improving accessibility to various target populations such as individuals with low literacy or reading disabilities. Prior surveys have been published several years before the introduction of transformers, transformer-based large language models (LLMs), and prompt learning that have drastically changed the field of NLP. The high performance of these models has sparked renewed interest in LS. To reflect these recent advances, we present a comprehensive survey of papers published since 2017 on LS and its sub-tasks focusing on deep learning. Finally, we describe available benchmark datasets for the future development of LS systems.