Descriptive Variables Research Articles

ASSOCIATION OF PECTORALIS MINOR MUSCLE LENGTH AND SHOULDER RANGE OF MOTION AMONG INDIVIDUALS WITH AND WITH OUT SHOULDER PAIN

Background and Aim: To evaluate the association of pectoralis minor muscle length and the shoulder range of motion with and without shoulder pain. Methodology: A sample of 214 participants with and without shoulder pain were enrolled in an analytical cross sectional study at Institute of physical medicine and rehabilitation, Dow University of health sciences, Karachi. Questionnaire was provided to all participants after taking consent. Individuals were categorized into two equal groups i.e. one with and the other without pain). Shoulder active ranges were measured with universal goniometer and pectoralis minor length with measuring tape. Statistical Package of Social Sciences version 21 was used for data analysis. The descriptive variables were assessed for frequencies and percentages. Continuous variables were shown with mean and standard deviations and were correlated with bivariate correlation test. Considered significant was 0.05 p value. Results: Females were 176(82.2%) and males were 38 (17.8%). Mean ± SD of age, weight, height, and BMI were 26.82 ±7.50, 58.45 ±12.11, 160.59 ± 12.43, and 22.18 ±3.78 respectively. The pain intensity negatively correlated with shoulder range of motions (rs = -0.307 to -0.775, p< 0.05) except medial rotation. Significant difference (p< 0.05) is found for length of pectoralis minor and range of motion between groups. There was also weak positive correlation between pectoralis minor index and shoulder lateral rotation (rs =0.215; p = 0.003). Conclusion: The shoulder pain affects shoulder joint range of motion and pectoralis minor length. Decreased pectoralis minor muscle length accompanies limited shoulder range of motion except, medial rotation.

Growth and Production Performance of Sugar Cane (Saccharum officinarum L.) Clon SB 01, SB 04, sb 19, SB 20 in the Village Curahmalang, Jombang Regency

Sugar productivity decreased in 2015-2016 by 11.76% and reached the lowest productivity in 2017 of 4,985 kg/ha White Crystal Sugar (GKP). The reason could be due to the presence of several types of sugarcane that experienced a decrease in yield during cultivation, such as nutrient deficiencies which resulted in divergent sugarcane plant height. Then the low sugarcane juice contained in sugarcane plants also affected production results. This study aimed to determine the growth and yield performance of sugarcane (Saccharum officinarum L.) clones SB01, SB04, SB19, SB20. This research was conducted in the garden of the Center for Sugarcane Research and Development (P3T) Faculty of Agriculture, University of Muhammadiyah Gresik, PG GEMPOLKREP and PT Perkebunan Nusantara X (PTPN X). Sumobito District, Jombang Regency from March to June 2021. Materials for observations used clones SB 01, clones SB 04, clones SB 19, and clones SB 20. The variables observed were sugarcane shoots, sugar cane stalks, and sugarcane leaves. Data analysis used descriptive analytical, variability and heritability.

Quality and Reliability of Spanish-language YouTube Videos on Breast Cancer

As the number of Spanish-speaking patients with cancer in the USA grows, it is vital that reliable online breast cancer information is available in Spanish. The main objective of our study was to determine the quality and reliability of online videos in Spanish regarding breast cancer. We conducted a YouTube search using the phrase “cancer de mama” (breast cancer) and graded their quality and reliability using the Journal of the American Medical Association (JAMA) Benchmark Criteria, Global Quality Score (GQS), and the Modified DISCERN Questionnaire. Videos were characterized by descriptive variables. Videos were primarily from Spain (36.4%) and Mexico (17.9%). ‘professional’ videos had the highest JAMA mean rank (99.53), the highest GQS mean rank (102.86) and the and second-highest DISCERN (101.10) mean rank. ‘Personal’ videos, by contrast, had the lowest mean rank values for GQS (72.28) and DISCERN (68.83) scores. A significant majority of online breast cancer videos in Spanish were of poor quality and reliability. ‘Professional’ videos are of higher quality and reliability than videos of other typologies; however, they have a limited range of content. In addition, a small percentage of videos were uploaded from the USA.

Evaluation of the impacted maxillary canines position, determination of the possible indicators of the impaction difficulty and the risk factors for adjacent teeth root resorption

Background/Aim. An impacted tooth is a tooth that could not erupt and take its place in the dental row. Impacted maxillary canines are a very frequent problem in orthodontic practices. They are the second most impacted teeth, right after the third molars. The aim of this cross-sectional study was to evaluate the two-dimensional and three-dimensional position of the impacted maxillary canines and make a descriptive study analysis of possible indicators of the impaction difficulty and risk factors for adjacent tooth root resorption. Methods. In this study, 94 subjects with 116 diagnosed maxillary canine impaction were included. The two-dimensional position of the impacted canines (IC) was evaluated on the panoramic projection of the cone beam computed tomography (CBCT) images. Canine position (CP) to the X, Y, and Z-axis was measured and scored using the novel classification system that incorporates three-dimensional information of CBCT imaging ? the KPG index. Based on the KPG index value, impaction was defined as easy, moderate, difficult, and extremely difficult. In order to determine indicators of impaction difficulty and risk factors of root resorption of adjacent teeth, qualitative variables (gender, side of impaction, etc.) and quantitative variables [age, canine distance to the occlusal plane (OccP), etc.] were tested with multiple logistic regression. Results. The mean age of the sub-jects was 19.8 ? 5.2 years. The impactions were twice as prevalent in females compared to males. Most (71.4%) impactions were unilateral in palatal position of the canines. Adjacent root resorption was present in 27.3% of cases of impaction, and the central incisor was mostly affected. More than half of the impactions were of moderate level. In univariable analysis, canine distance to sagittal medial line (SML), angle between canine and SML, OccP, and first premolar, respectively, were significant in impaction difficulty prediction. Multivariable analysis showed that angle between canine and the first premolar could be a risk factor for adjacent root resorption. Conclusion. Most impacted maxillary canines were in the palatal position, with a KPG index value of moderate difficulty. Besides the position to the OccP and SML, angle between canine and the first premolar should be estimated, as a part of diagnostic procedures, to evaluate the risk of adjacent root resorption, prevent resorption, and decide on the treatment plan.

Theoretical models of modulated nematic phases.

Novel modulated nematic phases, such as twist-bend nematics, splay-bend nematics and splay nematics, are an important subject of research in the field of liquid crystals. In this article fundamental information about the discovery, structure and properties of these phases is presented. Various theoretical models of elastic properties are compared, especially the proposed formulae for the free energy density of modulated nematic phases and the conditions for their stability. The emphasis is put on the variety of material parameters and variables in the mathematical description of the structures. The elastic models are classified according to a few criteria. Flexopolarisation is indicated as a main phenomenon responsible for the formation of modulated nematic phases. The elastic models are used for analysing the deformations of the twist-bend nematic structure in external fields. Dielectric, flexoelectric, ferroelectric and magnetic effects are considered. Two types of distortions are distinguished: microscopic (connected with the deformation of the director distribution) and macroscopic (related to the change of the optic axis direction). This review can be a starting point for further studies, for example computer simulations of modulated phases and design of liquid crystalline devices.

INCIDENCE AND EPIDEMIOLOGY OF ADHESIVE CAPSULITIS DURING THE COVID-19 PANDEMIC.

To evaluate a possible increase of adhesive capsulitis incidence during the COVID-19 pandemic. A total of 1,983 patients with shoulder disorders were retrospectively analyzed regarding gender, age, development of adhesive capsulitis and comorbidities (systemic arterial hypertension, diabetes mellitus, dyslipidemia, hypothyroidism, hyperthyroidism, depression, and anxiety) in two different periods: from March 2019 to February 2020 and from March 2020 to February 2021. Descriptive and quantitative variables were statistically analyzed. The program used for the calculations was SPSS 17.0 for Windows. During the pandemic, there was a 2.41-fold increase (p < 0.001) in cases of adhesive capsulitis (compared to the previous year). Patients with depression and anxiety had a significantly increased risk by 8.8 (p < 0.001) and 14 (p < 0.001) times, respectively, of developing frozen shoulder (regarding the two periods studied). A significant increase in the incidence of frozen shoulder was observed after the onset of the COVID-19 pandemic in addition to a simultaneous increase of psychosomatic disorders. Prospective studies would help to ratify the idea contained in this research. Level of Evidence III, Observational Cross-Sectional Study.

Corpus-based Development of Syntactic Complexity in EFL Writing

Syntactic complexity has long been taken as a significant factor in determining writing quality by EFL learners. And researchers in recent years have made a lot of efforts to devise and verify a wide range of factors or indicators for the purpose of measuring syntactic complexity or predicting EFL writing quality. The present study discussed bases itself on a self-built learner corpus with data collected over three years, with the aim of determining the applicable indices for predicting beginner writing quality. Based on previous research, the present study takes into account such factors as unit length, verb-VAC complexity, and clausal complexity. The results of pairwise comparisons by year indicate that there are significant differences for some indices but not for others. In terms of unit length, the three indicators of mean length of sentence, mean length of Tuni, and mean length of clause can serve as the main descriptive variables for the development of the language of beginners; for clausal complexity, seven indices: coordinate phrases per T-unit, verb phrases per T-unit, clauses per T-unit, coordinate phrases per clause, complex nominals per Tunit, complex nominals per clause and T-units per sentence, are the reliable indicators for beginner writing development. But indices for noun phrase complexity and verb-VAC complexity show no significant difference in the Kruskal-Wallis tests. The findings provide proof for the conclusion that knowledge and skills of modification, coordination, and subordination form the real difficulties for EFL beginners. It provides implications for coursebook design and classroom teaching where beginners are supposed to focus more on breeding awareness and skills in these aspects.

Psychoacoustic parameters in noise assessment – A research within the BMDV Network of Experts

This paper is dedicated to the applicability of psychoacoustic parameters for the evaluation of traffic noise. To characterize road traffic noises in detail and to demonstrate the perceptual impact, knowledge of the underlying dimensions that influence the evaluation of traffic sounds in everyday situations is required. In a study, stimulus-dyad comparisons were used to investigate underlying descriptive variables and attributes describing the latent perceptual dimensions of road traffic sounds. The dyads covered the feature space of road traffic sounds and systematically took into account factors such as traffic volume, traffic flow, traffic composition and road surface. The paper explains the experimental methodology, shows the derived attributes, and outlines the development of a questionnaire as a measurement tool to determine the perception of road traffic sounds in detail.

DISTURBANCES IN SERUM CALCIUM AND PHOSPHATE LEVELS IN PRE-DIABETES AND TYPE 2 DIABETES MELLITUS

Objective: To evaluate serum calcium and phosphate levels in Pre-diabetic, Diabetic &amp; non-diabetic subjects. Material and Methods: This comparative cross-sectional study was conducted at the Department of Chemical Pathology and Endocrinology, Armed Forces Institute of Pathology, from July to December 2021. The study used a non-probability consecutive sample to include 300 people. Those people have been divided into three groups; diabetics, pre-diabetics and non-diabetics with a hundred people in every group. Blood specimens had been taken for the estimation of calcium, phosphate, HbA1c, plasma glucose and insulin in fasting in their respective tubes. SPSS version 24 was used to inspect the data. All descriptive variables had their means and standard deviations calculated. To examine descriptive characteristics within and between all groups, one way ANOVA and post hoc LSD were performed, and p-values 0. 05 were considered significant. Results: Calcium and Phosphorus levels were lower in Diabetic group (Ca= 1.97±0.16mmol/l, PO4 = 0.76±0.07 mmol/l) as compared to pre-diabetic group (Ca= 2.15±0.18 mmol/l, PO4 = 0.93±0.06 mmol/l) and higher in non-diabetic group (Ca= 2.33±0.10mmol/l, PO4 = 1.19±0.20 mmol/l) than both other groups. A one-way ANOVA and Post Hoc LSD revealed that the difference between and within groups was statistically significant with a p value of 0.01. Conclusion: Our research has shown that as glycemic control slowly deteriorates, blood calcium and phosphate levels fall. Keywords: Calcium, Phosphorus, Pre-diabetes, Type 2 diabetes mellitus.

보육교사의 성격 5요인과 동료 관계가 소진에 미치는 영향

Objectives This study intends to improve daycare quality by examining the impact of personality types and coworker relationship on burnout among daycare teachers, as well as psychological characteristics that can help prevent burnout among daycare teachers.&#x0D; Methods The participants in this study were 198 daycare teachers who worked in daycare centers in Seoul and Chungbuk. In addition, In addition, Pearson's timely correlation coefficient was calculated to analyze the correlation between descriptive statistics and variables to find out the general trend using the SPSS 21.0 statistical program. Multiple regression analysis was also conducted to find out the influence between burnout.&#x0D; Results First, emotional exhaustion was favorably connected with neurotic characteristics, while sincerity, extroversion, affinity, and openness were negatively correlated when the five personality variables of daycare teachers, coworker relationships, and burnout were investigated. Extroversion, affinity, and honesty were negatively connected with dehumanization, while neurotic factors were positively correlated. Sincerity, extroversion, affinity, and flexibility were inversely connected with a drop in sense of achievement, while neurotic factors were positively correlated. The relationship between coworker relationships and burnout sub-factors was all negative, indicating that the better the coworker relationship, the less dehumanization, loss of achievement, and emotional exhaustion. Second, when looking at the five personality factors of daycare teachers and the effect of relationships with coworkers on burnout, it was observed that neurosis, which is a sub-factor of burnout, had a substantial effect among personality factors, while other personality factors had no effect. Dehumanization had a strong impact on personality traits such as neuroticism and coworker relationships, as well as sincerity, affinity, and extroversion, which had a major impact on achievement loss.&#x0D; Conclusions As a result, efforts should be made to aggressively assist daycare teachers with aspects that are prone to burnout among the teacher's personality features, and diverse program support is needed not only for individuals but also for teachers to have a mutual interaction.

Post COVID-19 Pandemic Generalized Anxiety Status of Health Professional undergraduate students

Aim: To assess the psychological impact of COVID-19 on mental status of undergraduate medical students, after reopening of educational institutions. Methods: This descriptive, online cross-sectional study was conducted on medical students of Combined Military Hospital Lahore Medical College &amp; Institute of Dentistry, Lahore Pakistan for 6 months from 1st June to 30th November, 2021. Generalized Anxiety Disorder Scale (GAD-7) was used as the study tool. Data was analyzed by using statistical software SPSS - 23. Frequencies and percentages were used for descriptive variables. A univariate analysis was utilized to identify the noticeable associations between traits of sample and the anxiety level in current scenario of COVID-19 pandemic. Finally multivariate logistic regression analyses, odds ratio (OR), and Spearman's correlation coefficient, r, was used to evaluate the association between COVID-19-related stressors and anxiety level. Results: Of the 324 respondents, majority were females residing in urban areas with their parents. Severe anxiety was experienced by 23.8% of students. Female respondents were found to have more anxiety as compared to males (OR = 1.81; 95% CI = 1.173 – 2.815). Moreover, respondents having a relative or an acquaintance infected with COVID-19 reported to have more anxiety (OR = 3.007, 95% CI = 2.377 - 3.804). Conclusion: A significant number of students are experiencing anxiety in the post COVID-19 phase, especially those that had an acquaintance infected with COVID-19. There is a need to look after student's mental health and to implement appropriate psychological strategies and interventions to deal with this level of psychological distress in the students at this phase of pandemic for optimum training of future health professionals. Keywords: Anxiety, COVID-19, healthcare professionals, mental health, medical students, pandemic

Perception of Professional Competence of Medical Doctors Trained Using Innovative and Traditional Learning Methods

Background; Innovative learning also known as Problem Based Learning (PBL) is student centered whereas Traditional teaching is tutor centered. PBL is a main method of acquiring and applying knowledge and thus develop life-long learning skills. Moi University school of Medicine uses PBL whereas University of Nairobi uses traditional learning methods. Objectives; To evaluate the perceptions of professional competence among medical doctors using innovative learning methods (PBL) as contrasted to traditional teaching methods in their medical education. Methods; This was a cross-sectional study carried out on the medical doctors from Moi University and University of Nairobi. Moi University uses problem-based learning whereas University of Nairobi uses traditional methods as a learning model. A total of 92 post internship medical doctors were recruited. After the obtaining an informed consent, data was collected from the sampled respondents. Data was analyzed using STATA version 10 to obtain descriptive variables. Results; From Moi University there were 37(40.2%) and from University of Nairobi were 55(59.8%). There were 38 females (41.3%) and 54 male (58.7%). Majority of the respondents (79.3%) were in the age range of 25-29years. A total of eight competencies were assessed. In all the eight assessed competencies, the P-value for variable rating in two competencies was less than 0.05. These were; Leadership skills and Planning, organization and working independently. Conclusion; PBL curricula graduates felt that they were better prepared in leadership skills as compared to counterparts who were trained using traditional teaching methods.

Applicability of Multivariate Linear Regression in Building Energy Demand Estimation

The vision of the research project is to find an energy optimal building configuration, suitable for specified requirements and restrictions. The first step on this way is to create a measure to compare building configurations, faster than explicit energetic simulations. The current study examines the applicability of multivariate linear regression to support the solution of building optimization problems. During the study, multivariate linear regression models were created to estimate the expected annual heating energy demand of building configurations and examined their accuracy Between examinations, the models were modified so that the complexity was increased only to such an extent that the approximation was still sufficiently accurate. The result was a multivariate linear model that estimated the expected output for unknown descriptive variables with a 0% relative error and a 1.6% standard deviation. The R2 point of the estimates was 0.9884. Based on these, the model was considered applicable in the search space defined by the training patterns.

Dodging justice: characteristics of men with multiple victims who evade detection for long periods

ABSTRACT Little is known about men who sexually abuse multiple children and evade detection for long periods. This group is responsible for inflicting significant harm due to the severity, and longevity of their offending. The aim of the study was to identify men who perpetrate persistent sexual offences against children and differentiate them by those with long detection lags (LDLs) (n = 35) and short detection lags (SDLs) (n = 35) using police databases and case files. The men were compared on demographic characteristics, lifestyle factors, victimisation, psychopathic traits, antisocial orientation, atypical sexual interest, and descriptive sexual offence variables. The LDL participants had more psychopathic traits, more pre-pubescent child victims, more often be in professional employment and have less antisocial and criminogenic indicators than the comparative SDL group. These findings have significant practical implications for reduced victimisation through early identification, forensic risk assessment, treatment and earlier prevention strategies for CSA. Practice impact statement The current findings have important and practical implications within forensic risk assessment, especially Structured Professional Judgement (SPJ) when assessing specific groups of offenders. In addition, research informed SPJ can better prioritise policing investigative resources, both retrospectively (e.g. identifying victims) and prospectively (e.g. reducing detection lags and victimisation) while defensively informing the criminal justice system of the most appropriate supervision intensity and intervention modalities for men with clinical and forensic indicators of potential high victimisation and longer detection lags.

Lung ultrasonography performed by nephrologist: COVID-19 as an opportunity to reveal ultrasound's full potential and usefulness in the dialysis room.

Interest in point-of-care ultrasound (POCUS) and lung ultrasound (LUS) is growing in the nephrology and dialysis field, and the number of nephrologists skilled in what is proving to be the "5th pillar of bedside physical examination" is increasing. Patients on hemodialysis (HD) are at high risk of contracting severe acute respiratory syndrome coronavirus 2 (SARS-COV-2) and developing coronavirus disease 2019 (COVID-19) serious complications. Despite this, to our knowledge there are no studies to date that show the role of LUS in this setting, while there are many in the emergency room, where LUS proved to be an important tool, providing risk stratification and guiding management strategies and resource allocation. Therefore, it is not clear whether the usefulness and cut-offs of LUS highlighted in studies in the general population are reliable in dialysis, or whether variations, precautions and adjustments to this specific situation are necessary. This was a 1-year monocentric prospective observational cohort study of 56 HD patients with COVID-19. Patients underwent a monitoring protocol that included at first evaluation bedside LUS, using a 12-scan scoring system, by the same nephrologist. All data were prospectively and systematically collected. Outcomes. hospitalization rate, combined outcome [non-invasive ventilation (NIV+death)], mortality. Descriptive variables are presented as medians (interquartile range), or percentage. Univariate and multivariate analysis, as well as Kaplan-Meier (K-M) survival curves, were carried out. P was fixed at .05. Median age was 78 years, 90% had at least one comorbidity (46% diabetics), 55% were hospitalized and 23% deaths. Median duration of disease was 23 days (14-34). A LUS score ≥11 represented a 13-fold risk of hospitalization, a 16.5-fold risk of combined outcome (NIV+death) vs risk factors such as age [odds ratio (OR) 1.6], diabetes (OR 1.2), male sex (OR 1.3) and obesity (OR 1.25), and a 7.7-fold risk of mortality. In the logistic regression, LUS score ≥11 is associated with the combined outcome with a hazard ratio (HR) of 6.1 vs inflammations indices such as CRP ≥9mg/dL (HR 5.5) and interleukin-6 (IL-6) ≥62 pg/mL (HR 5.4). In K-M curves, survival drops significantly with LUS score above 11. In our experience of COVID-19 HD patients, LUS appeared to be an effective and easy tool, predicting the need for NIV and mortality better than "classic" known COVID-19 risk factors such as age, diabetes, male sex and obesity, and even better than inflammations indices such as CRP and IL-6. These results are consistent with those of the studies in the emergency room setting, but with a lower LUS score cut-off (11 vs 16-18). This is probably due to the higher global frailty and peculiarity of HD population, and emphasizes how nephrologists should themselves use LUS and POCUS as a part of their everyday clinical practice, adapting it to the peculiarity of the HD ward.

A kinetic description of the body size distribution of species

In this paper, by resorting to classical methods of statistical mechanics, we build a kinetic model able to reproduce the observed statistical weight distribution of many diverse species. The kinetic description of the time variations of the weight distribution is based on elementary interactions that describe in a qualitative and quantitative way successive evolutionary updates, and determine explicit equilibrium distributions. Numerical fittings on mammalian eutherians of the order Chiroptera population illustrates the effectiveness of the approach.

Classifications of Odontoid Process Fractures: A Systematic Review and Proposal of a New Simplified Classification System Based on Embryology.

Odontoid fractures are the most common cervical spine fractures in the elderly. Although many classification systems have been developed for them, the ambiguity in various definitions can potentially lead to misunderstandings. This paper aims to review the terminologies and current classification systems of odontoid fractures and propose a new, simplified anatomical classification. Given the descriptive variability of odontoid fractures in current classifications, we systematically reviewed the literature using PRISMA guidelines querying the National Library of Medicine PubMed database. The initial literature search yielded 175 publications. A total of seven reports met the inclusion criteria and were ultimately included for a full review. The classification systems previously used to categorize fractures of the odontoid process often need to be more transparent, imprecise, and incongruous. To simplify them, a new embryologically accurate system is proposed. A new embryological and anatomically-based system, combining the former systems' specific attributes, allows a more straightforward and adaptable classification of odontoid fractures.

Do artificial neural networks love sex? How the combination of artificial neural networks with evolutionary algorithms may help to identify gender influence in rheumatic diseases.

Although medical research has been performed predominantly on men both in preclinical and clinical studies, continuous efforts have been made to overcome this gender bias. Examining retrospectively 21 data sets containing sex as one of the descriptive variables, it was possible to verify how many times our AI protocol decided to keep gender information in the predictive model. The data sets pertained a vast array of diseases such as dyspeptic syndrome, atrophic gastritis, venous thrombosis, gastroesophageal reflux disease, irritable bowel syndrome, Alzheimer diseases and mild cognitive impairment, myocardial infarction, gastrointestinal bleeding, gastric cancer, hypercortisolism, AIDS, Covid diagnosis, extracorporeal membrane oxygenation in intensive therapy, among others. The sample size of these data sets ranged between 80 and 3147 (average 600). The number of variables ranged from 19 to 101 (average 41). Gender resulted to be part of the heuristic predictive model 19 out of 21 times. This means that also for highly adaptive and potent tools like Artificial Neural Networks, information on sex carries a specific value. The result of this study confirms the importance of gender information in building high performance predictive model in the field of Artificial Intelligence (AI). Therefore, also for AI, sex counts.

Earth's Paleomagnetic Field and its Changes Through Time

Paleomagnetic data is the only way to access the behavior of the geomagnetic field of internal origin through geological time. Therefore, the complete description of the paleosecular variation (PSV) is fundamental for unraveling the evolution of the Earth's core. Modeling of PSV relies on observing the fluctuations in the direction and intensity of the field on the surface, the evaluation of the paleomagnetic data dispersion in different times and situations of low and high reversal rates, the kinematics of the virtual poles during reversals, and other aspects. Magneto-cyclostratigraphy is a powerful tool to constrain the age of sedimentary formations, and the recognition of patterns of magnetic variation represents a breakthrough in the development of cyclostratigraphy. This paper reviews some of the contributions to the subject of the Laboratory of Paleomagnetism at Instituto de Astronomia, Geofísica e Ciências Atmosféricas the University of São Paulo.

Learning Explainable Disentangled Representations of E-Commerce Data by Aligning Their Visual and Textual Attributes

Understanding multimedia content remains a challenging problem in e-commerce search and recommendation applications. It is difficult to obtain item representations that capture the relevant product attributes since these product attributes are fine-grained and scattered across product images with huge visual variations and product descriptions that are noisy and incomplete. In addition, the interpretability and explainability of item representations have become more important in order to make e-commerce applications more intelligible to humans. Multimodal disentangled representation learning, where the independent generative factors of multimodal data are identified and encoded in separate subsets of features in the feature space, is an interesting research area to explore in an e-commerce context given the benefits of the resulting disentangled representations such as generalizability, robustness and interpretability. However, the characteristics of real-word e-commerce data, such as the extensive visual variation, noisy and incomplete product descriptions, and complex cross-modal relations of vision and language, together with the lack of an automatic interpretation method to explain the contents of disentangled representations, means that current approaches for multimodal disentangled representation learning do not suffice for e-commerce data. Therefore, in this work, we design an explainable variational autoencoder framework (E-VAE) which leverages visual and textual item data to obtain disentangled item representations by jointly learning to disentangle the visual item data and to infer a two-level alignment of the visual and textual item data in a multimodal disentangled space. As such, E-VAE tackles the main challenges in disentangling multimodal e-commerce data. Firstly, with the weak supervision of the two-level alignment our E-VAE learns to steer the disentanglement process towards discovering the relevant factors of variations in the multimodal data and to ignore irrelevant visual variations which are abundant in e-commerce data. Secondly, to the best of our knowledge our E-VAE is the first VAE-based framework that has an automatic interpretation mechanism that allows to explain the components of the disentangled item representations with text. With our textual explanations we provide insight in the quality of the disentanglement. Furthermore, we demonstrate that with our explainable disentangled item representations we achieve state-of-the-art outfit recommendation results on the Polyvore Outfits dataset and report new state-of-the-art cross-modal search results on the Amazon Dresses dataset.