Editorial: Innovations and Perspectives in Data Mining and Knowledge Discovery
Naoki Abe, Huan Liu, Kuansan Wang · Frontiers in Big Data · 2021
In this edited collection under a special research topic, we present four recent articles in the broader areas surrounding data mining and knowledge discovery for big data, which together attest to the diversity and broadness of application domains and associated technical agenda being explored in this fast evolving discipline.In particular, the collection centers around papers affiliated with the ACM SIGKDD International Conference on Data Mining and Knowledge Discovery (KDD 2019) and its satellite workshops (e.g., WSDM: Workshop on Issues of Sentiment Discovery and Opinion Mining, Fragile Earth: Data Science for a Sustainable Planet), which address a varying set of agenda such as epistemology of data mining (e.g., societal and ethical aspects), automation of data mining (e.g., robustness of deep neural architecture search), advances in text mining (e.g.deep learning based embeddings of text data and data collection) and application in climate change mitigation (e.g., data collection eco-systems).It is our hope that this collection will provide the audience with a glimpse into the rich landscape of data mining related research agenda and applications of big data today.The first article is on "Big Data and the Little Big Bang: An Epistemological (R)evolution" by Dominik Balazka and Dario Rodighiero.This is a timely article that looks at the field of big data in a refreshing perspective.The authors try to answer intriguing questions such as "What qualifies as big data?What does big data promise?Is big data a revolution or evolution?".Starting from an analysis of frequently employed definitions of big data, the authors argue that there are intrinsic weaknesses of big data and it is more appropriate to define big data in relational terms.The excessive emphasis on volume and technological aspects of big data, combined with neglected epistemological issues, implies that big data is neutral, omni-comprehensive and theory free.The authors show that this rhetoric contradicts the empirical reality of big data: 1) data collection is not neutral nor objective; 2) big data has more data than before, but more does not mean all, the totality of the data population; and 3) interpretation and knowledge production remain both theoretically informed and subjective.The authors then argue that big data may be interpreted as a methodological revolution carried over by evolutionary processes in technology and epistemology, or a third paradigm.They also point out that big data has promoted a new digital divide between big data rich and big data poor populations, therefore, radically shaping the power dynamics involved in the processes of production and analysis of data.The second article is about "On Robustness of Neural Architecture Search Under Label Noise" by Yi-