Data mining, machine learning (ML), and artificial intelligence (AI) are widely presented as the response to cyber threats that signature-based defenses can no longer contain. This structured, integrative review examines a less-discussed consequence of that shift: the techniques that support intrusion, malware, phishing, insider, and fraud detection also enable offensive misuse and privacy loss, because defensive and offensive systems often draw on shared computational primitives. We organise the recent literature (2015–2026) around one question: how do shared AI and data-mining capabilities produce coupled defensive, offensive, and privacy consequences in cyber threat detection? The review is selective rather than exhaustive. Its contribution is to connect three literatures usually treated separately: AI-based detection; adversarial ML and generative-AI misuse; and privacy leakage from trained models. We then describe a technical ethics in which privacy and security obligations are expressed through specific mechanisms, including differential privacy, federated learning, homomorphic encryption, secure multi-party computation, anonymisation, and explainable models, together with their limits. Available evidence, much of it industry-reported, suggests an operational asymmetry favouring generation in selected generative-phishing and social-engineering settings, rather than a general advantage of offence over defence. As a conceptual interpretation rather than a synthesised empirical result, we argue that detection capability and privacy protection are better treated as a coupled, multi-objective design problem than as competing goals.
This paper investigates the connotation of new quality productivity, integrating existing research findings to construct a measurement index system tailored for our country, and conducts an in-depth analysis of its impact on economic growth. The study utilizes provincial panel data from Chinese provinces spanning 2012 to 2022. The results indicate that, in contrast to prior studies, new quality productivity influences economic growth not only through supply-side mechanisms but also plays a pivotal role in demand-side transmission; notably, the upgrading of consumption structures and the expansion of investment domains exert both direct and indirect effects. Accordingly, this paper proposes relevant policy recommendations. The innovative aspects of this study are as follows: 1. Measuring new quality productivity across multiple dimensions by incorporating environmental and ecological factors while exploring its transmission mechanism from the demand side; 2. Introducing dynamic quantitative analysis to simulate the operational mechanisms of new quality productivity on consumption and investment; 3. Empirical tests demonstrate that after addressing issues related to heterogeneity, endogeneity, and robustness, the transmission mechanism of new quality productivity remains significant and robust.
Panel data are the main defence of empirical economics against unobserved confounding, and two-way fixed effects regression has long been the default estimator. A review of why that default fails, and of what has replaced it, is provided. When treatment timing varies and treatment effects are heterogeneous, the two-way fixed effects coefficient is a weighted average of causal contrasts whose weights need not be positive, so that the estimand can fall outside the range of the effects it aggregates. Heterogeneity-robust difference-indifferences estimators repair the aggregation but leave identification untouched, and sensitivity analysis rather than pre-trend testing is the appropriate response to doubt about parallel trends. When trends are not parallel, factor-model and matrix completion counterfactuals are needed, and these are shown to depend critically on the length of the pre-treatment window. Eight Monte Carlo experiments quantify the negative weighting, the cost of imposing additivity, the failure of the two-way regression under continuous and switching treatments, and the fact that conditioning on a passed pre-trend test makes the bias larger rather than smaller.
12 October 2026 in Digital Economy and Sustainable Development
Transactions in a digital economy generate data that firms accumulate but national accounts do not price. Within the model developed here, that omission implies measured GDP understates output. We adapt the recursive data-economy framework of Farboodi and Veldkamp (2026) to China and construct a bottom-up measure of digital investment from the itemized asset disclosures of 5,120 listed firms. The model does not identify a unique asset perimeter, so we carry two measurement boundaries throughout rather than designating one as the headline. For 2024 the adjustment is 1.60% of GDP, or 2.16 trillion RMB, under a broad boundary that includes telecommunications network assets, and 0.60%, or 0.82 trillion RMB, under a narrow boundary that excludes them. These are alternative measurement boundaries, not statistical bounds. On listed firms alone, before any extrapolation to unlisted firms, the broad-boundary figure is 0.91%. Because the knowledge stock is not identified without forecast-error data, the broad-boundary figure carries a calibrated scenario range of 0.77–4.07%, which reflects alternative calibrations rather than sampling uncertainty. Every figure is conditional on a normalization of the mapping between measured digital investment and the value of data. A validation audit of the classifier uncovers measurement defects that are not specific to China.
18 August 2026 in Journal of King Saud University Computer and Information Sciences
Cross-domain recommendation over heterogeneous e-commerce networks faces three unresolved technical failures: existing GNN recommenders apply a single shared attention vector across all neighbor pairs, which can introduce a type-conflation bias governed by the inter-type variance of attention coefficients; meta-path importance weights learned in a source domain become systematically miscalibrated under the relational distribution shift of cross-domain transfer; and contrastive learning methods apply type-agnostic perturbations that assign equal dropout intensity to semantically distinct edge types, degrading representation quality in heterogeneous graph settings. We propose HGAT-Rec, which incorporates a heterogeneity-aware contrastive learning (HCL) objective that grounds view construction and sample selection in the typed relational structure of a cross-domain heterogeneous graph: type-stratified edge dropout preserves high-signal interaction channels proportionally to their attention weight, type-conditioned positive samples are drawn from meta-path neighborhoods sharing genuine relational content, and a cross-domain InfoNCE term enforces relative ordering constraints across the full overlapping user population. To support HCL, HGAT-Rec further provides (i) a type-triple-indexed graph attention mechanism parameterized by the joint combination of source node type, target node type, and edge type; (ii) a domain-conditioned meta-path aggregation layer with separate per-domain path importance distributions sharing a domain-invariant query vector; and (iii) a TransR-based knowledge gated fusion module decoupling item semantics from interaction sparsity. Theoretical grounding is provided by two formal propositions characterizing type-conflation bias and contrastive robustness under sparsity. Experiments on the Amazon Product Co-purchasing benchmark show that HGAT-Rec outperforms eleven baselines, achieving 11.20% and 8.70% improvements in NDCG@10 and HR@10 over the strongest competitor, a 31.80% gain for cold-start users, and a sparsity degradation rate of 35.2% under 80% interaction removal versus 49.0% for the next-best baseline.