Journal of Intelligent Information Systems最新文献

Dynamic networks are ubiquitous in many domains for modelling evolving graph-structured data and detecting changes allows us to understand the dynamic of the domain represented. A category of computational solutions is represented by the pattern-based change detectors (PBCDs), which are non-parametric unsupervised change detection methods based on observed changes in sets of frequent patterns over time. Patterns have the ability to depict the structural information of the sub-graphs, becoming a useful tool in the interpretation of the changes. Existing PBCDs often rely on exhaustive mining, which corresponds to the worst-case exponential time complexity, making this category of algorithms inefficient in practice. In fact, in such a case, the pattern mining process is even more time-consuming and inefficient due to the combinatorial explosion of the sub-graph pattern space caused by the inherent complexity of the graph structure. Non-exhaustive search strategies can represent a possible approach to this problem, also because not all the possible frequent patterns contribute to changes in the time-evolving data. In this paper, we investigate the viability of different heuristic approaches which prevent the complete exploration of the search space, by returning a concise set of sub-graph patterns (compared to the exhaustive case). The heuristics differ on the criterion used to select representative patterns. The results obtained on real-world and synthetic dynamic networks show that these solutions are effective, when mining patterns, and even more accurate when detecting changes.

Social recommender systems (SRS) have garnered adequate attention due to the supplementary information provided by social network, which aids in making recommendations. However, social network information contains noise, which can be detrimental to recommendation performance. Current social recommendation models are deficient in feature validation and extraction of social data. To fill that gap, we propose a novel model called Social View Explorer Collaborative Filtering (SVE-CF) which aims to extract significant consistent signals from the noisy social network. First, SVE-CF correlates users’ social and interaction behaviors, creating follow, joint, and interaction views to represent all interaction patterns. Second, it samples unlabeled examples from users to assess consistency across the three views, assigning pseudo-labels as evidence of social homophily. Third, it selects top-k pseudo-labels to amplify significant consistent signals and minimize noise through tri-view joint learning. Extensive experiments are conducted to demonstrate the effectiveness of the proposed model over the commonly used state-of-the-art (SOTA) methods.

The need for effective and efficient search and rescue operations is more important than ever as the frequency and severity of disasters increase due to the escalating effects of climate change. Recognizing the value of personal knowledge and past experiences of experts, in this paper, we present findings of an investigation of how past knowledge and experts’ experiences can be effectively integrated with current search and rescue practices to improve rescue planning and resource allocation. A special focus is on investigating and demonstrating the potential associated with integrating knowledge graphs and case-based reasoning as a viable approach for search and rescue decision support. As part of our investigation, we have implemented a demonstrator system using a Norwegian search and rescue dataset and case-based and concept-based similarity retrieval. The main contribution of the paper is insight into how case-based and concept-based retrieval services can be designed to improve the effectiveness of search and rescue planning. To evaluate the validity of ranked cases in terms of how they align with the existing knowledge and insights of search and rescue experts, we use evaluation measures such as precision and recall. In our evaluation, we observed that attributes, such as the rescue operation type, have high precision, while the precision associated with the objects involved is relatively low. Central findings from our evaluation process are that knowledge-based creation, as well as case- and concept-based similarity retrieval services, can be beneficial in optimizing search and rescue planning time and allocating appropriate resources according to search and rescue incident descriptions.

In the study of large-scale complex knowledge graphs, due to the incompleteness of knowledge and the existence of low-frequency knowledge samples, existing knowledge graph complementation methods are often limited by the amount of data and ignore the complex semantic information. To solve this problem, this paper proposes a knowledge graph completion method CGAML based on the combination of Conditional Generative Adversarial Network and Meta-Learning, which utilizes the hierarchical background knowledge as the basis and introduces conditional variables in the Generative Adversarial Network to represent the required semantic information to constrain the semantic attributes of the generated knowledge. In addition, we design a meta-learning multi-task framework to embed Conditional Generative Adversarial Networks into the meta-learning process and propose local constraints and global gradient optimization strategies to quickly adapt to new tasks and improve computational efficiency. Empirically, our method demonstrates superior performance in realizing few-shot link prediction when compared to existing representative methods.

Every day, thousands of questions are asked on the Community Question Answering network, making these questions and answers extremely valuable for information seekers around the world. However, a significant proportion of these questions do not elicit proper answers. There are several reasons for this, with the lack of clarity in questions being one of the most crucial factors. In this study, our primary focus is on enhancing the clarity of unclear questions in Community Question Answering networks. In the first step, DistilBERT, which uses Siamese and triplet network structures for meaningful sentence embeddings, is combined with HDBSCAN, effective in diverse noise datasets and less sensitive to density variations, to extract unique features from each question. Questions were then categorized as clear or unclear using an Extremely Randomized Trees ensemble model, known for its robust resistance to class imbalance, with more than 90% accuracy. Next, efforts were made to extract information that could enhance the clarity of unclear questions by comparing them with similar, clearer questions using Dynamic Time Warping, a versatile technique suitable for time series analyses in information systems and applicable across various domains. Finally, the extracted information was incorporated into the feature vector of unclear questions based on histogram-coverage methods to enhance their clarity. When a question is made clearer, the missing information and its importance are shown to the questioner. This enables the questioner to be aware of the missing information and facilitates them in clarifying the question.

Although Knowledge Graphs (KGs) provide great value in many applications, they are often incomplete with many missing facts. KG Completion (KGC) is a popular technique for knowledge supplement. However, there are two fundamental challenges for KGC. One challenge is that few entity pairs are often available for most relations, and the other is that there exists complex relations, including one-to-many (1-N), many-to-one (N-1), and many-to-many (N-N). In this paper, we propose a new model to accomplish Few-shot KG Completion (FKGC) under complex relations, which is called Relation representation based on Private and Shared features for Adaptive few-shot link prediction (RPSA). In this model, we utilize the hierarchical attention mechanism for extracting the essential and crucial hidden information regarding the entity’s neighborhood so as to improve its representation. To enhance the representation of few-shot relations, we extract the private features (i.e., unique feature of each entity pair that represents the few-shot relation) and shared features (i.e., one or more commonalities among a few entity pairs that represent the few-shot relation). Specifically, a private feature extractor is used to extract the private semantic feature of the few-shot relation in the entity pair. After that, we design a shared feature extractor to extract the shared semantic features among a few reference entity pairs in the few-shot relation. Moreover, an adaptive aggregator aggregates several representations of the few-shot relation about the query. We conduct experiments on three datasets, including NELL-One, CoDEx-S-One and CoDEx-M-One datasets. According to the experimental results, the RPSA’s performance is better than that of the existing FKGC models. In addition, the RPSA model can also handle complex relations well, even in the few-shot scenario.

Entity Resolution (ER) is a crucial process in the field of data management and integration. The primary goal of ER is to identify different profiles (or records) that refer to the same real-world entity across databases. The challenging problem is that labeling a large sample of profiles can be very expensive and time-consuming. Active Machine Learning (ActiveML) addresses this issue by selecting the most representative or informative profiles pairs to be labeled. The informativeness is determined by the capacity to diminish the uncertainty of the model. Conversely, representativeness evaluates whether a selected instance effectively reflects the overall input patterns of unlabeled data. Traditional ActiveML techniques typically rely on one strategy, Which may severely restrict the performance of the ActiveML process and lead to slow convergence. Especially in ER problems with a lack of initial training data. In this paper, we overcame this issue by inventing an approach for balancing the two above strategies. The implemented solution named EBEES (Epsilon-based Balancing Exploration and Exploitation Strategy), Which contains two variations: Adaptive-(epsilon ) and (epsilon )-decreasing. We evaluated the EBEES on twelve datasets. Comparing the EBEES strategy against the state-of-the-art methods, without an initial training data, showed an enhanced performance in terms of F1-score, model stability, and rapid convergence.

Abstract

Recommender systems are indispensable because they influence our day-to-day behavior and decisions by giving us personalized suggestions. Services like Kindle, YouTube, and Netflix depend heavily on the performance of their recommender systems to ensure that their users have a good experience and to increase revenues. Despite their popularity, it has been shown that recommender systems reproduce and amplify the bias present in the real world. The resulting feedback creates a self-perpetuating loop that deteriorates the user experience and results in homogenizing recommendations over time. Further, biased recommendations can also reinforce stereotypes based on gender or ethnicity, thus reinforcing the filter bubbles that we live in. In this paper, we address the problem of gender bias in recommender systems with explicit feedback. We propose a model to quantify the gender bias present in book rating datasets and in the recommendations produced by the recommender systems. Our main contribution is to provide a principled approach to mitigate the bias being produced in the recommendations. We theoretically show that the proposed approach provides unbiased recommendations despite biased data. Through empirical evaluation of publicly available book rating datasets, we further show that the proposed model can significantly reduce bias without significant impact on accuracy and outperforms the existing model in terms of bias. Our method is model-agnostic and can be applied to any recommender system. To demonstrate the performance of our model, we present the results on four recommender algorithms, two from the K-nearest neighbors family, UserKNN and ItemKNN, and the other two from the matrix factorization family, Alternating Least Square and Singular Value Decomposition. The extensive simulations of various recommender algorithms show the generality of the proposed approach.

Nowadays, various applications across industries, healthcare, and security have begun adopting automatic sentiment analysis and emotion detection in short texts, such as posts from social media. Twitter stands out as one of the most popular online social media platforms due to its easy, unique, and advanced accessibility using the API. On the other hand, supervised learning is the most widely used paradigm for tasks involving sentiment polarity and fine-grained emotion detection in short and informal texts, such as Twitter posts. However, supervised learning models are data-hungry and heavily reliant on abundant labeled data, which remains a challenge. This study aims to address this challenge by creating a large-scale real-world dataset of 17.5 million tweets. A distant supervision approach relying on emojis available in tweets is applied to label tweets corresponding to Ekman’s six basic emotions. Additionally, we conducted a series of experiments using various conventional machine learning models and deep learning, including transformer-based models, on our dataset to establish baseline results. The experimental results and an extensive ablation analysis on the dataset showed that BiLSTM with FastText and an attention mechanism outperforms other models in both classification tasks, achieving an F1-score of 70.92% for sentiment classification and 54.85% for emotion detection.

Commonsense Question Answering (CQA) aims to select the correct answers to common knowledge questions. Most existing approaches focus on integrating external knowledge graph (KG) representations with question context representations to facilitate reasoning. However, the approaches cannot effectively select the correct answer due to (i) the incomplete reasoning chains when using knowledge graphs as external knowledge, and (ii) the insufficient understanding of semantic information of the question during the reasoning process. Here we propose a novel model, KA-AGN. First, we utilize a joint representation of dependency parse trees and language models to describe QA pairs. Next, we introduce question semantic information as nodes into a knowledge subgraph and compute the correlations between nodes using adaptive graph networks. Finally, bidirectional attention and graph pruning are employed to update the question representation and the knowledge subgraph representation. To evaluate the performance of our method, we conducted experiments on two widely used benchmark datasets: CommonsenseQA and OpenBookQA. The ablation experiment results demonstrate the effectiveness of the adaptive graph network in enhancing reasoning chains, while showing the ability of the joint representation of dependency parse trees and language models to correctly understand question semantics. Our code is publicly available at https://github.com/agfsghfdhg/KAAGN-main.