RelKD 2024: The Second International Workshop on Resource-Efficient Learning for Knowledge Discovery

It is our great pleasure to welcome you to the 19th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD). The annual ACM SIGKDD conference is the premier international forum for data mining, knowledge discovery and big data. It brings together researchers and practitioners from academia, industry, and government to share their ideas, research results and experiences. KDD-2013 features plenary presentations, paper presentations, poster sessions, workshops, tutorials, exhibits, demonstrations, and the KDD Cup competition. Today, you hear a lot about big data, data science and data intensive computing. The core of this work is extracting knowledge and useful information from data, which for science leads to beautiful insights, and for applications leads to actions, alerts and decisions. The KDD community has always been at the center of this activity and it is clear from this conference that it will continue to drive this broader field of big data. This year there were 726 submissions to the KDD Research Track, and 125 papers were accepted. There were 136 submissions to the KDD Industry and Government Track, and 34 papers were accepted. KDD also has a history of inviting talks that are of broad interest to the KDD community. This year we chose to have 4 plenary talks. A program committee also selected 8 talks to present at the Industry Practice Exposition. A strength of the KDD conference is the number of workshops and tutorials that are co-located with it. This year there were 10 full-day workshops, 5 half-day workshops, and 6 tutorials. We thank all sponsors, who are a very important part of the conference, and the members of the Organizing Committee and our other colleagues who volunteered their time during the past year to make this conference a success. Special thanks goes to the Research Track Co-Chairs and the Industry and Government Track Co-Chairs. Also special thanks are due to the Local Arrangements Chair, the Treasurer, the Proceedings Co-Chairs, and the KDD Cup Committee. We are grateful to the several program committees that provided the advice necessary to put together a quality program - the Research Track Program Committee, the Research Track Senior Program Committee, the Industry and Government Track Program Committee, the Industry Practice Expo Program Committee, the Workshop Program Committee, the Tutorial Program Committee, and the Demo Program Committee. We know that you will find this year's exhibits and demonstrations exciting and remind you that some of the most interesting discussions can be found there. Please join us for KDD-2013 to gain new knowledge and to exchange exciting new research results, leading practices, and high impact applications in big data, knowledge discovery and data mining. We hope that you will find this program interesting and thought-provoking and that the conference will provide you with a valuable opportunity to share ideas with other researchers and practitioners from institutions around the world.

Deep learning on graphs is an upcoming area of study. This chapter provides an introduction to graph neural networks (GNNs), a type of neural network that is designed to process data represented in the form of graphs. First, it summarizes the explanation of deep learning on graphs. The fundamental concepts of graph neural networks, as well as GNN theories, are then explained. In this chapter, different types of graph neural network (GNN) are also explained. At the end, the applications of graph neural network where GNN is used and for what purpose it is going to be used are explained. This also explores the various applications of GNNs in fields such as social network analysis, recommendation systems, drug discovery, computer vision, and natural language processing. With the increasing prevalence of graph data, GNNs are becoming increasingly important and will likely continue to play a significant role in many fields in the future.

Evaluation Of Recommendation System On A Movie Dataset

Knowledge discovery is a compute and data intensive process that allows for finding patterns, trends, and models in large datasets. The Grid can be effectively exploited for deploying knowledge discovery applications because of the high-performance it can offer and its distributed infrastructure. For effective use of Grids in knowledge discovery, the development of middleware is critical to support data management, data transfer, data mining and knowledge representation. To such purpose, we designed the Knowledge Grid, a high-level environment providing for Grid-based knowledge discovery tools and services. Such services allow users to create and manage complex knowledge discovery applications, composed as workflows that integrate data sources and data mining tools provided as distributed Grid services. This chapter describes the Knowledge Grid architecture and describes how its components can be used to design and implement distributed knowledge discovery applications. Then, the chapter describes how the Knowledge Grid services can be made accessible using the Open Grid Services Architecture (OGSA) model.

The subject of the research is machine learning models and methods for generating legal contracts with limited resources and performance evaluation benchmarks. The goal of the work is to analyse approaches of domain-specific Large Language Models development and to find the optimal method of creating independent specialized systems that can generate contracts in different languages and legal systems. The article addresses the following tasks: identification of existing companies and solutions in this area, exploring approaches to create texts in natural language, analysis of evaluation and comparison methods of such systems, inspecting limitations and shortcomings of existing solutions and approaches, finding the optimal method of developing systems with limited resources. The following results were obtained: approaches of natural language generation and their features were investigated; the "Transformer" architecture was defined as a modern standard in the field of text information generation; different model types which are based on this architecture were considered; data sources for training were analysed; methods of adapting models in specialized areas were considered; model evaluating benchmarks for various tasks were reviewed; shortcomings of the existing specialized language models and the incompleteness of existing benchmarks for contract generation task evaluation were revealed. As a result of the analytical experiment, it was determined that the Retrieval-Augmented Generation method is the most optimal for solving the given task under the given conditions. The conducted experiment and its results can be used as a basis for further research of domain-specific language models development with limited resources. Conclusions: the article provides an overview of natural language generation methods using modern machine learning techniques, considers their advantages and disadvantages for small companies and scientific institutions that have limited resources. The work examinates a specialized legal domain and the problem of contract generation and determines the most optimal method to solve it.

Chronic kidney disease (CKD) is a global health burden that affects approximately 10% of the adult population in the world. It is also recognized as the top 20 causes of death worldwide. Unfortunately, there is no cure for CKD however, it is possible to slow down its progression and mollify the damage by early diagnosis of the disease. Due to a limited number of nephrologists, the early diagnosis of CKD is often not possible for most of the people. Therefore the use of modern computer-aided methods is necessary to aid the traditional CKD diagnosis system to be more efficient and accurate. In this research, our primary focus was to apply 3 modern machine learning techniques namely logistic regression, feedforward neural networks and wide & deep learning to diagnose CKD as well as finding the best performing technique by evaluating their diagnosis performance. To evaluate their performance, f1-score, precision, recall and AUC score was used for logistic regression and an additional loss score was considered for the feedforward neural networks and wide & deep model. We found the feedforward neural network as the best performing technique for CKD diagnosis with 0.99 f1-score, 0.97 precision, 0.99 recall and 0.99 AUC score. Logistic regression produced the lowest result among all and the wide & deep learning with a larger number of hidden layers and neurons found to be effective for larger datasets.

A survey of graph neural network based recommendation in social networks

For many applications in knowledge discovery in databases, finding outliers, which are rare events, is of importance. Outliers are observations that deviate significantly from the rest of the data, so they seem to have been generated by another process (Hawkins, 1980). Such outlier objects often contain information about an untypical behaviour of the system.

Increasing time-series remote sensing images provide the information about the evolution processes of ecosystems on multi-spatial scales. Vegetation plays an important role in sustaining the natural environment and supporting human being with goods and ecosystem services. Detection of vegetation degradation has become a hot spot of multi-disciplinary researches recently. In this paper, a case study of spatio-temporal data mining and knowledge discovery for detection of vegetation degradation has been conducted. The special issues focused on the quantitative determination of historical evolutionary trend and furthermore, the sustainability of different trends in the future. Taking the Circum-Bohai-Sea region as the case study area, the Unary Linear Regression Model (ULRM) has been established based on the time-series SPOT-VGT images from 1998 to 2008, and then the Hurst index has been calculated by R/S method on the spatial scales of cell (1km <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> ) and the whole study area. It turned out that, the combined analysis between Slope of ULRM and Hurst index could effectively reveal the characteristics of vegetation changes, which included the degraded areas in the past as well as the risk level of degradation in the future. Overall, the areas of vegetation degradation in the future amount to 38.87 thousand square kilometers, which accounts for 7.55% of the whole study area. In addition, these degraded areas mainly distributed around the metropolitan regions, coastal zone, and so on. The findings will help us with more intelligent strategies of degradation prevention.

With the ubiquity of complex networks in domains as diverse as social network analysis, recommender systems and epidemiology, improvements in link predictions have a potential for far-reaching impact. In machine learning, a well-known approach to improve classification performance is to combine several methods in an ensemble. It is also well established that, diversity or complementarity of the classifiers to be combined, is a prerequisite for a potential performance improvement of the formed ensemble. In this work, we combine link prediction heuristics (e.g. Common Neighbour, Adamic-Adar etc.), with static graph neural networks, discrete dynamic graph neural networks, and continuous dynamic graph neural networks. We analyze the importance of each of the methods and how complementary they are with respect to each other. We also use greedy searches to select better subsets of methods to combine and train the ensemble in a roll-forward manner to make it adapt to changes in the dynamic network over time. In the conducted experiments on a number of challenging link prediction tasks, the ensemble from the greedy search and the roll-forward ensemble shows on average a 55% and 61% improvement respectively, over the best single method.

Network mining plays a pivotal role in many high-impact application domains, including information retrieval, healthcare, social network analysis, security and recommender systems. State-of-the-art offers a wealth of sophisticated network mining algorithms, many of which have been widely adopted in real-world with superior empirical performance. Nonetheless, they often lack effective and efficient ways to characterize how the results of a given mining task relate to the underlying network structure. In this paper, we introduce network derivative mining problem. Given the input network and a specific mining algorithm, network derivative mining finds a derivative network whose edges measure the influence of the corresponding edges of the input network on the mining results. We envision that network derivative mining could be beneficial in a variety of scenarios, ranging from explainable network mining, adversarial network mining, sensitivity analysis on network structure, active learning, learning with side information to counterfactual learning on networks. We propose a generic framework for network derivative mining from the optimization perspective and provide various instantiations for three classic network mining tasks, including ranking, clustering, and matrix completion. For each mining task, we develop effective algorithm for constructing the derivative network based on influence function analysis, with numerous optimizations to ensure a linear complexity in both time and space. Extensive experimental evaluation on real-world datasets demonstrates the efficacy of the proposed framework and algorithms.

Augmented label propagation for seed set expansion

Community detection has been widely studied in the areas of social network analysis and recommendation system. However, most existing research focus on cases where relationships are explicit or depend on simultaneous appearance. In this paper, we propose to study the community detection problem where the relationships are not based on simultaneous appearance, but time-delayed appearances. In other words, we aim to capture the relationship where one individual physically follows another individual. In our attempt to capture such relationships, the major challenge is the presence of spatial homophily, i.e., individuals are attracted to locations due to their popularities and not because of communications. In tackling the community detection problem with spatial homophily and delayed responses, we make the following key contributions: (1) We introduce a four-phase framework, which by way of using quantified impacts excludes homophily. (2) To validate the framework, we generate a synthetic dataset based on a known community structure and then infer that community structure. (3) Finally, we execute this framework on a real-world dataset with more than 6,000 taxis in Singapore. Our results are also compared to those of a baseline approach without homophily-elimination.

With the rapid growth of the social web an increasing number of people started to replicate their off-line preferences and lives in an on-line environment. Consequently, the social web provides an enormous source for social network data, which can be used in both commercial and research applications. However, people often take part in multiple social network sites and, generally, they share only a selected amount of data to the audience of a specific platform. Consequently, the interlinkage of social graphs from different sources getting increasingly important for applications such as social network analysis, personalization, or recommender systems. This paper proposes a novel method to enhance available user re-identification systems for social network data aggregation based on face-recognition algorithms. Furthermore, the method is combined with traditional text-based approaches in order to attempt a counter-balancing of the weaknesses of both methods. Using two samples of real-world social networks (with 1610 and 1690 identities each) we show that even though a pure face-recognition based method gets outperformed by the traditional text-based method (area under the ROC curve 0.986 vs. 0.938) the combined method significantly outperforms both of these (0.998, p = 0.0001) suggesting that the face-based method indeed carries complimentary information to raw text attributes.

We study the prospect of constraining invisible branching ratio of the Higgs boson in the vector boson fusion channel using deep learning techniques. Taking advantage of the differing QCD radiation patterns between signal and background, we find that modern machine learning techniques have the capability of significantly outperforming traditional analyses.