Bias Adjustment Research Articles

Canadian climate data portals: A comparative analysis from a user perspective

Climate data portals are essential tools for climate change adaptation. This study analyses differences between two Canadian portals providing bias-adjusted CMIP6 simulations: Climate Data Canada and Portraits Climatiques. The study evaluates three core variables (daily maximum temperature, daily minimum temperature and precipitation) as well as assesses five case studies, taken from the agriculture, transport and health sectors, that relied on climate indicators available through the portals. The underlying datasets vary in multiple ways (bias-adjustment methodology, climate of reference, ensemble composition, emissions scenarios) and, in general, the climatology of variables and indicators tends to be statistically different between portals towards the end of the century. Differences are significantly reduced when comparing projected changes with respect to present climate conditions, highlighting the important role played by the dataset used as a reference for the bias-adjustment procedure. When considered from the point of view of practical applications, the discrepancies between the portals are generally, although not always, sufficiently small that they do not impact the resulting decisions. Finally, indicators based on a fixed threshold were found to be strongly influenced by the reference used for the bias adjustment.

Which Error Components in TRMM-Based Multisatellite Precipitation Estimates Reduce over Chinese Mainland after Official Bias Adjustments: Systematic or Random?

Abstract Climatological calibration algorithm (CCA) and satellite–gauge combination (SG) are two official bias adjustments for satellite precipitation estimates (SPE) in the Tropical Rainfall Measuring Mission (TRMM) Multisatellite Precipitation Analysis (TMPA). The CCA is designed for the near-real-time SPEs, while the SG procedure is a final step to merge pure SPEs with gauge observations. This study explored the impacts of CCA and SG on the systematic and random errors of TMPA SPEs. The errors of TMPA version-7 near-real-time products before and after CCA (RT_UC, RT_C), and the research product TMPA 3B42 (V7), were decomposed into systematic and random components, benchmarked by the China Gauge-based Daily Precipitation Analysis (CGDPA). After being calibrated by CCA, RT_C reduced the systematic errors relative to RT_UC over the Chinese mainland, except in the Tibetan Plateau and Tianshan Mountains. However, CCA did not aid in reducing random errors; instead, it even exacerbated the random errors. On the other hand, the SG merging is more effective in reducing systematic errors of SPEs than CCA calibration because of the direct inclusion of simultaneous gauge data from the Global Precipitation Climatology Centre (GPCC). We also found that SG merging reduced the random errors of pure SPEs over regions with relatively higher elevations. Despite lower random errors in V7 over the complex terrain region, the SG unfavorably increased the random errors over southeastern China. The results reported here may offer valuable insights into the effects of CCA and SG techniques drawn from TMPA, with the potential to advance the development of bias-adjusting algorithms for SPEs in the future.

Predictive factors of clot propagation in patients with superficial venous thrombosis towards deep venous thrombosis and pulmonary embolism: a systematic review and meta-analysis

ObjectiveA subset of patients with superficial venous thrombosis (SVT) experiences clot propagation towards deep venous thrombosis (DVT) and/or pulmonary embolism (PE). The aim of this systematic review is to identify...

Benzodiazepines and risk of dementia – Is there a reason for alarm?

IntroductionThe population ageing is a reality associated with an increase in prevalence of Dementia. The use of benzodiazepines is often postulated as a risk factor in these syndromes.Contrary to recommendations for its short-time use, long-term and chronic use are common, with an estimated 8,7% of elderly people in the US taking benzodiazepines.ObjectivesTo clarify the most recent evidence on the use of benzodiazepines and the risk of developing dementia.MethodsNon-systematic review of literature, using PubMed as database and filtering the results for meta-analysis.ResultsFour articles were included in this review.Zhong G et al. concluded that risk of dementia increased in consumers of benzodiazepines and it was associated with higher doses.In turn, AlDawasari A et al., when trying to clarify the use of different sedative-hypnotic drugs, found and increased risk with the consumption of benzodiazepines. After exclusion of articles with confounders and adjustment for protopathic bias, the risk was not maintained.Lucchetta RC et al. concluded that the risk exists but without inferring differences between doses or duration of action.Finally, Penninkilampi R e Eslick GD investigated this association, after controlling for the protopathic bias, concluding, contrary to AlDawasari et al., that the association benzodiazepines consumption and dementia do not result from this bias.ConclusionsWe cannot draw robust and concrete conclusions between benzodiazepines consumption and the pathogenesis of dementia because not only is the literature limited, but results are also heterogeneous. However, these prescriptions must be carried out cautiously, especially in the elderly, due to the known adverse effects associated with them.Disclosure of InterestNone Declared

315 A Propensity Score-matched Comparative Study of Balloon Guide Catheters Versus Conventional Guide Catheters for Mechanical Thrombectomy With Stenting in Tandem Strokes: Comparison of 1st-pass Effect, Symptomatic ICH, and 90-day Functional Outcomes

INTRODUCTION: Extensive clot burden in tandem strokes accounts for poor mechanical thrombectomy (MT) outcomes. Several studies have shown benefit of balloon guide catheters (BGC) in MT and carotid artery stenting. METHODS: Tandem stroke patients identified from our endovascular database were dichotomized into groups treated with BGCs versus conventional guide catheters. One-to-one PSM adjustment for baseline demographics and treatment selection bias using nearest-neighbor matching was performed. Patient demographics, presentation characteristics, and procedural details were recorded. Outcomes assessed were final modified thrombolysis in cerebral infarction (mTICI) grade, periprocedural symptomatic intracranial hemorrhage (sICH) rate, in-hospital mortality, and 90-day modified Rankin Scale (mRS) score. Mann-Whitney U-test and multivariate logistic regression were performed to compare procedural parameters and clinical outcomes. RESULTS: Concurrent carotid revascularization (stenting with/without angioplasty) and MT was performed in 125 cases (BGC: 85; no BGC: 40). After PSM (40 patients/group), BGC group had a significantly shorter procedure duration (77.9 vs. 61.5 min; odds ratio [OR] = 0.996; p = 0.006), lower discharge National Institutes of Health Stroke Scale score (8.0 vs. 11.0; OR = 0.987; p = 0.042), and higher odds of 90-day mRS 0-2 score (52.3% vs. 27.5%; OR = 0.34; p = 0.040). On multivariate regression, BGC group had a significantly higher first-pass effect rate (mTICI: 2B-3) (OR = 1.115, 95% confidence interval [CI] = 1.015-1.432; p = 0.013) and lower periprocedural sICH rate (OR = 0.615; 95% CI = 0.406-0.932; p = 0.025). No difference in in-hospital mortality was observed (OR = 1.591; 95% CI = 0.976-2.593; p = 0.067). CONCLUSIONS: BGCs used for concurrent MT–carotid revascularization with flow arrest were safe and resulted in superior clinical and angiographic outcomes in tandem stroke patients.

Global Projection of Flood Risk With a Bivariate Framework Under 1.5–3.0°C Warming Levels

AbstractGlobal warming increases the atmospheric water‐holding capacity, consequently altering the frequency, and intensity of extreme hydrological events. River floods characterized by large peak flow or prolonged duration can amplify the risk of social disruption and affect ecosystem stability. However, previous studies have mostly focused on univariate flood magnitude characteristics, such as flood peak or volume, and there is still limited understanding of how these joint flood characteristics (i.e., magnitude and duration) might co‐evolve under different warming levels. Here, we develop a systematical bivariate framework to project future flood risk in 11,528 catchments across the globe. By constructing the joint distribution of flood peak and duration with copulas, we examine global flood risk with a bivariate framework under varying levels of global warming (i.e., within a range of 1.5–3.0°C above pre‐industrial levels). The flood projections are produced by driving five calibrated lumped hydrological models (HMs) using the simulations with bias adjustment of five global climate models (GCMs) under three shared socioeconomic pathways (SSP126, SSP370, and SSP585). On average, global warming from 1.5 to 3.0°C tends to amplify flood peak and lengthen flood duration across almost all continents, but changes are not unidirectional and vary regionally around the globe. The joint return period (JRP) of the historical (1985–2014) 50‐year flood event is projected to decrease to a median with 36 years under a medium emission pathway at the 1.5°C warming level. Finally, we evaluate the drivers of these JRP changes in the future climate and quantify the uncertainty arising from the different GCMs, SSPs, and HMs. Our findings highlight the importance of limiting greenhouse gas emission to slow down global warming and developing climate adaptation strategies to address future flood hazards.

Lead-Time Corrected Effect on Breast Cancer Survival in Germany by Mode of Detection.

(1) Background: Screen-detected breast cancer patients tend to have better survival than patients diagnosed with symptomatic cancer. The main driver of improved survival in screen-detected cancer is detection at earlier stage. An important bias is introduced by lead time, i.e., the time span by which the diagnosis has been advanced by screening. We examine whether there is a remaining survival difference that could be attributable to mode of detection, for example, because of higher quality of care. (2) Methods: Women with a breast cancer (BC) diagnosis in 2000-2022 were included from a population-based cancer registry from Schleswig-Holstein, Germany, which also registers the mode of cancer detection. Mammography screening was available from 2005 onwards. We compared the survival for BC detected by screening with symptomatic BC detection using Kaplan-Meier, unadjusted Cox regressions, and Cox regressions adjusted for age, grading, and UICC stage. Correction for lead time bias was carried out by assuming an exponential distribution of the period during which the tumor is asymptomatic but screen-detectable (sojourn time). We used a common estimate and two recently published estimates of sojourn times. (3) Results: The analysis included 32,169 women. Survival for symptomatic BC was lower than for screen-detected BC (hazard ratio (HR): 0.23, 95% confidence interval (CI): 0.21-0.25). Adjustment for prognostic factors and lead time bias with the commonly used sojourn time resulted in an HR of 0.84 (CI: 0.75-0.94). Using different sojourn times resulted in an HR of 0.73 to 0.90. (4) Conclusions: Survival for symptomatic BC was only one quarter of screen-detected tumors, which is obviously biased. After adjustment for lead-time bias and prognostic variables, including UICC stage, survival was 27% to 10% better for screen-detected BC, which might be attributed to BC screening. Although this result fits quite well with published results for other countries with BC screening, further sources for residual confounding (e.g., self-selection) cannot be ruled out.

The improvement of device linearity in AlGaN/GaN HEMTs at millimeter-wave frequencies using dual-gate configuration

The linearity performance of the AlGaN/GaN high-electron-mobility-transistors (HEMTs) is critical for circuit applications especially at millimeter-wave frequencies. In this paper, we propose a new dual-gate (DG) configuration for linearity improvement at millimeter-wave frequencies. Using 2D simulation, we observed the improvement in device linearity by placing the additional DC gate in the source-gate access region. Adjusting the bias voltage of the DC gate modulates the horizontal electric field in the channel and controls the acceleration behavior of the channel electrons. Furthermore, such bias adjustment enhances the electron density in the channel to compensate for the transconductance (Gm) degradation. Both of these contribute to the flattening of the Gm profile for linearity improvement. The proposed DG configuration improves the C/I ratio by 10 dB compared to the conventional single gate (SG) device at 38 GHz. Additionally, a 60 % improvement in the cutoff frequency (fT) and maximum oscillation frequency (fmax) is obtained compared to the traditional DG configuration where the second gate is placed in the gate-drain drift region. This is mainly due to the reduction in gate capacitances from the redistribution of the electrons in the device. Such superior characteristics have shown great potential of the proposed device for high linearity applications at millimeter-wave frequencies.

Strategies to deal with genetic analyzer-specific DNA methylation measurements.

Targeted bisulfite sequencing using single-base extension (SBE)can be used to measure DNA methylation via capillary electrophoresis on genetic analyzers in forensic labs. Several accurate age prediction models have been reported using this method. However, using different genetic analyzers with different software settings can generate different methylation values, leading to significant errors in age prediction. To address this issue, the study proposes and compares four methods as follows: (1) adjusting methylation values using numerous actual body fluid DNA samples, (2) adjusting methylation values using control DNAs with varying methylation ratios, (3) constructing new age prediction models for each genetic analyzer type, and (4) constructing new age prediction models that could be applied to all types of genetic analyzers. To test the methods for adjusting values using actual body fluid DNA samples, previously reported adjusting equations were used for blood/saliva DNA age prediction markers (ELOVL2, FHL2, KLF14, MIR29B2CHG/C1orf132, and TRIM59). New equations were generated for semen DNA age prediction markers (TTC7B, LOC401324/cg12837463, and LOC729960/NOX4) by drawing polynomial regression lines between the results of the three types of genetic analyzers (3130, 3500, and SeqStudio). The same method was applied to obtain adjustment equations using 11 control DNA samples. To develop new age prediction models for each genetic analyzer type, linear regression analysis was conducted using DNA methylation data from 150 blood, 150 saliva, and 62 semen samples. For the genetic analyzer-independent models, control DNAs were used to formulate equations for calibrating the bias of the data from each genetic analyzer, and linear regression analysis was performed using calibrated body fluid DNA data. In the comparison results, the genetic analyzer-specific models showed the highest accuracy. However, genetic analyzer-independent models through bias adjustment also provided accurate age prediction results, suggesting its use as an alternative in situations with multiple constraints.

Projected changes of rain, sleet, and snowfall in Norway

ABSTRACT Tailored products based on climate projections are in demand for climate adaptation planning in different industries. To meet the needs of the tourism industry, the authors applied available datatsets to calculate projections for the distribution of precipitation as rain, sleet, and snow in Norway, using daily average temperature to classify the precipitation phases. Amounts and number of days with precipitation in the different phases were calculated. The projections were based on bias-adjusted output from 10 EURO-CORDEX models under two emission scenarios. In general, total precipitation, as well as temperature, was projected to increase, while the number of days with precipitation was not projected to change significantly. The proportion of rainfall was projected to increase while that of snow was expected to decrease. Sleet ratio was projected to decrease in low lying coastal areas, and to increase in mountainous and inland areas. The results were presented for several tourist destinations. However, the authors found that the bias adjustment method applied in the input dataset led to a bias towards too much rain and too little snow, which should be considered when interpreting the results. They concluded that projections of rain, sleet and snow days were considered less affected by that flaw.

Comparison of bias adjustment in meta-analysis using data-based and opinion-based methods.

Several methods exist for bias adjustment of meta-analysis results, but there has been no comprehensive comparison with unadjusted methods. We compare 6 bias-adjustment methods with 2 unadjusted methods to examine how these different methods perform. We re-analyzed a meta-analysis that included 10 randomized controlled trials. Two data-based methods (Welton's data-based approach and Doi's quality effects model) and 4 opinion-informed methods (opinion-based approach, opinion-based distributions combined statistically with data-based distributions, numerical opinions informed by data-based distributions, and opinions obtained by selecting areas from data-based distributions) were used to incorporate methodological quality information into the meta-analytical estimates. The results of these 6 methods were compared with 2 unadjusted models: the DerSimonian-Laird random effects model and Doi's inverse variance heterogeneity model. The 4 opinion-based methods returned the random effects model estimates with wider uncertainty. The data-based and quality effects methods returned different results and aligned with the inverse variance heterogeneity method with some minor downward bias adjustment. Opinion-based methods seem to only add uncertainty rather than bias adjust.

Impacts of improved horizontal resolutions in the simulations of mean and extreme precipitation using CMIP6 HighResMIP models over West Africa.

We conducted an analysis of 16 historical simulations from the High-Resolution Model Intercomparison Project (HighResMIP) as part of the Coupled Model Intercomparison Project (CMIP) phase 6 (CMIP6). These simulations encompass both high- and low-resolution models and aim to investigate the impact of improved horizontal resolution on mean and extreme precipitation in West Africa between 1985 and 2014. Six Expert Team on Climate Change Detection and Indices (ETCCDI) were used to charactererize extreme indices. Bias adjustment was used to detect and adjust the biases in the models. Our observations indicate that the southeastern and southwestern regions of West Africa experience the most significant precipitation, which aligns with the simulations from HighResMIP. The enhanced horizontal resolution notably influences the simulation of orographically induced rainfall in elevated areas and intensifies precipitation in various aspects. When examining the highest 1-day precipitation, our observations reveal that most of the Guinea Coast region had 1-day rainfall exceeding 100mm. However, this was overestimated and in some simulations underestimated by HighResMIP simulations. Furthermore, an increase in horizontal resolution appears to enhance the ability of high-resolution models to replicate the observed patterns of heavy precipitation (R10mm) and very heavy rainfall (R20mm) days. Spatial and temporal analysis suggests that uncertainty exists in the simulation of extreme precipitation in both high- and low-resolution simulations over West Africa. Also, bias adjustment shows a significant bias in the simulations. To address this issue, we employed a bias adjustment approach.

Simulation models for aggregated data meta-analysis: Evaluation of pooling effect sizes and publication biases.

Simulation studies are commonly used to evaluate the performance of newly developed meta-analysis methods. For methodology that is developed for an aggregated data meta-analysis, researchers often resort to simulation of the aggregated data directly, instead of simulating individual participant data from which the aggregated data would be calculated in reality. Clearly, distributional characteristics of the aggregated data statistics may be derived from distributional assumptions of the underlying individual data, but they are often not made explicit in publications. This article provides the distribution of the aggregated data statistics that were derived from a heteroscedastic mixed effects model for continuous individual data and a procedure for directly simulating the aggregated data statistics. We also compare our simulation approach with other simulation approaches used in literature. We describe their theoretical differences and conduct a simulation study for three meta-analysis methods: DerSimonian and Laird method for pooling aggregated study effect sizes and the Trim & Fill and precision-effect test and precision-effect estimate with standard errors method for adjustment of publication bias. We demonstrate that the choice of simulation model for aggregated data may have an impact on (the conclusions of) the performance of the meta-analysis method. We recommend the use of multiple aggregated data simulation models to investigate the sensitivity in the performance of the meta-analysis method. Additionally, we recommend that researchers try to make the individual participant data model explicit and derive from this model the distributional consequences of the aggregated statistics to help select appropriate aggregated data simulation models.

National, regional, and global estimates of low birthweight in 2020, with trends from 2000: a systematic analysis

Low birthweight (LBW; <2500 g) is an important predictor of health outcomes throughout the life course. We aimed to update country, regional, and global estimates of LBW prevalence for 2020, with trends from 2000, to assess progress towards global targets to reduce LBW by 30% by 2030. For this systematic analysis, we searched population-based, nationally representative data on LBW from Jan 1, 2000, to Dec 31, 2020. Using 2042 administrative and survey datapoints from 158 countries and areas, we developed a Bayesian hierarchical regression model incorporating country-specific intercepts, time-varying covariates, non-linear time trends, and bias adjustments based on data quality. We also provided novel estimates by birthweight subgroups. An estimated 19·8 million (95% credible interval 18·4-21·7 million) or 14·7% (13·7-16·1) of liveborn newborns were LBW worldwide in 2020, compared with 22·1 million (20·7-23·9 million) and 16·6% (15·5-17·9) in 2000-an absolute reduction of 1·9 percentage points between 2000 and 2020. Using 2012 as the baseline, as this is when the Global Nutrition Target began, the estimated average annual rate of reduction from 2012 to 2020 was 0·3% worldwide, 0·85% in southern Asia, and 0·59% in sub-Saharan Africa. Nearly three-quarters of LBW births in 2020 occurred in these two regions: of 19 833 900 estimated LBW births worldwide, 8 817 000 (44·5%) were in southern Asia and 5 381 300 (27·1%) were in sub-Saharan Africa. Of 945 300 estimated LBW births in northern America, Australia and New Zealand, central Asia, and Europe, approximately 35·0% (323 700) weighed less than 2000 g: 5·8% (95% CI 5·2-6·4; 54 800 [95% CI 49 400-60 800]) weighed less than 1000 g, 9·0% (8·7-9·4; 85 400 [82 000-88 900]) weighed between 1000 g and 1499 g, and 19·4% (19·0-19·8; 183 500 [180 000-187 000]) weighed between 1500 g and 1999 g. Insufficient progress has occurred over the past two decades to meet the Global Nutrition Target of a 30% reduction in LBW between 2012 and 2030. Accelerating progress requires investments throughout the lifecycle focused on primary prevention, especially for adolescent girls and women living in the most affected countries. With increasing numbers of births in facilities and advancing electronic information systems, improvements in the quality and availability of administrative LBW data are also achievable. The Children's Investment Fund Foundation; the UNDP-UNFPA-UNICEF-WHO World Bank Special Programme of Research, Development and Research Training in Human Reproduction; and the Bill & Melinda Gates Foundation.

Resonant Tunneling-Enhanced Photoresponsivity in a Twisted Graphene van der Waals Heterostructure.

Leveraging its ultrahigh carrier mobility, zero-bandgap linear dispersion, and extremely short response time, graphene exhibits remarkable potential in ultrafast broad-band photodetection. Nonetheless, the inherently low responsivity of graphene photodetectors, due to the low photogenerated carrier density, significantly impedes the development of practical devices. In this study, we present an improved photoresponse within a graphene-hexagonal boron nitride-graphene vertical tunnel junction device, where the crystallographic orientation of the two graphene electrodes is aligned. Through meticulous device structure design and the adjustment of bias and gate voltages, we observe a 2 orders of magnitude increase in tunneling photocurrent, which is attributed to the momentum-conserving resonant electron tunneling. The enhanced external photoresponsivity is evident across a wide temperature and spectral range and achieves 0.7 A/W for visible light excitation. This characteristic, coupled with the device's negative differential conductance, suggests a novel avenue for highly efficient photodetection and high-frequency, logic-based optoelectronics using van der Waals heterostructures.

Ibicus: a new open-source Python package and comprehensive interface for statistical bias adjustment and evaluation in climate modelling (v1.0.1)

Abstract. Statistical bias adjustment is commonly applied to climate models before using their results in impact studies. However, different methods based on a distributional mapping between observational and model data can change the simulated trends as well as the spatiotemporal and inter-variable consistency of the model, and are prone to misuse if not evaluated thoroughly. Despite the importance of these fundamental issues, researchers who apply bias adjustment currently do not have the tools at hand to compare different methods or evaluate the results sufficiently to detect possible distortions. Because of this, widespread practice in statistical bias adjustment is not aligned with recommendations from the academic literature. To address the practical issues impeding this, we introduce ibicus, an open-source Python package for the implementation of eight different peer-reviewed and widely used bias adjustment methods in a common framework and their comprehensive evaluation. The evaluation framework introduced in ibicus allows the user to analyse changes to the marginal, spatiotemporal and inter-variable structure of user-defined climate indices and distributional properties as well as any alteration of the climate change trend simulated in the model. Applying ibicus in a case study over the Mediterranean region using seven CMIP6 global circulation models, this study finds that the most appropriate bias adjustment method depends on the variable and impact studied, and that even methods that aim to preserve the climate change trend can modify it. These findings highlight the importance of use-case-specific selection of the method and the need for a rigorous evaluation of results when applying statistical bias adjustment.

Adjustment of selection bias for clinical trials: a simulation study

Clinical trial selection bias is a common issue, as patients are typically not selected randomly from a target population. Various statistical approaches have been proposed to adjust for this bias, including IPW (inverse probability weights), SPS (subclassification with propensity scores), and EVB (external validity bias). However, there has been very little statistical research to compare the performance of these methods in clinical trials. To bridge this gap, we conducted a simulation study using a patient population with seven covariates and a true treatment effect size of 0.5 (Cohen’s d). Next, we assessed the efficacy of the three statistical methods on nonrandom clinical trial samples with varying sizes and covariates. Based on our simulation results, EVB is the most effective method for adjusting clinical trial selection bias when there are seven covariates. SPS is the most effective method for adjusting clinical trial selection bias when there are three and five covariates. However, we observed that IPW's performance was inadequate, indicating that it may not be a suitable option for selection bias adjustment in clinical trials. In summary, our study sheds light on the effectiveness of various statistical methods in mitigating selection bias in clinical trials.

Quantile Regression for Longitudinal Functional Data with Application to Feed Intake of Lactating Sows

AbstractThis article focuses on the study of lactating sows, where the main interest is the influence of temperature, measured throughout the day, on the lower quantiles of the daily feed intake. We outline a model framework and estimation methodology for quantile regression in scenarios with longitudinal data and functional covariates. The quantile regression model uses a time-varying regression coefficient function to quantify the association between covariates and the quantile level of interest, and it includes subject-specific intercepts to incorporate within-subject dependence. Estimation relies on spline representations of the unknown coefficient functions and can be carried out with existing software. We introduce bootstrap procedures for bias adjustment and computation of standard errors. Analysis of the lactation data indicates, among others, that the influence of temperature increases during the lactation period.Supplementary materials accompanying this paper appear on-line.

Factor Model Comparisons with Conditioning Information

Abstract We develop methods for testing factor models when the weights in portfolios of factors and test assets can vary with lagged information. We derive and evaluate consistent standard errors and finite sample bias adjustments for unconditional maximum squared Sharpe ratios and their differences. Bias adjustment using a second-order approximation performs well. We derive optimal zero-beta rates for models with dynamically trading portfolios. Factor models’ Sharpe ratios are larger but standard test asset portfolios’ maximum Sharpe ratios are larger still when there is dynamic trading. As a result, most of the popular factor models are rejected.

LSAB: User Behavioral Pattern Modeling in Sequential Recommendation by Learning Self-Attention Bias

Since the weight of a self-attention model is not affected by the sequence interval, it can more accurately and completely describe the user interests, so it is widely used in processing sequential recommendation. However, the mainstream self-attention model focuses on the similarity between items when calculating the attention weight of user behavioral patterns but fails to reflect the impact of user sudden drift decisions on the model in time. In this article, we introduce a bias strategy in the self-attention module, referred to as Learning Self-Attention Bias (LSAB) to more accurately learn the fast-changing user behavioral patterns. The introduction of LSAB allows for the adjustment of bias resulting from self-attention weights, leading to enhanced prediction performance in sequential recommendation. In addition, this article designs four attention-weight bias types catering to diverse user behavior preferences. After testing on the benchmark datasets, each bias strategy in LSAB is useful for state-of-the-art and can improve the performance of the models by nearly 5% on average. The source code listing is publicly available at https://gitee.com/kyle-liao/lsab .