Similar Tasks Research Articles

The Effect of an Expository Intervention on Strategy Use and Oral Expression of Informational Texts for Adolescents With Learning Disabilities.

This study investigated the effect of an expository strategy intervention called Sketch and Speak on strategy use and oral reporting of informational texts for students with language-learning disabilities (LLD). Four adolescents with LLD participated in a single-case multiple-baseline-across-participants treatment experiment. Ten individual treatment sessions involved shared reading of an informational article, identifying important or interesting ideas to remember, making pictographic or bulleted notes paired with oral sentence formulation and rehearsal, and orally rehearsing the final full report. Following each baseline and treatment session, participants had an opportunity to review their notes and then gave a free-recall oral report and answered content and strategy awareness questions. Pre/post measures of independent strategy use and oral reporting were also administered. All participants learned pictography and improved their written notes, strategy awareness, and quality of oral reports compared to baseline. Three participants improved their independent note-taking, oral reports, and strategy awareness on proximal tasks. One participant showed independent oral rehearsal within treatment and on the proximal transfer task. In the distal independence task, all the participants showed some improvement in planning notes format but none for explanations of a familiar sport/game. One participant used pictography for any task in which there was a choice of notation format. Sketch and Speak provides an effective set of teaching strategies to improve informational oral reporting for older students with LLD. Students may generalize improved note-taking as a learning strategy to similar tasks, but independent oral rehearsal is more difficult to obtain.

Comparing color qualia structures through a similarity task in young children versus adults

Examination of the subjective qualitative aspects of an experience, or "qualia" in short, is a fundamental and core aspect of consciousness research. How can we characterize the particular quality of redness, i.e. a red quale? Based on a recent proposal of the structural characterization of qualia, which did not rely on verbal descriptions, we developed a task that obtained pairwise similarity judgments at four graded levels, with easy and intuitive visual interfaces designed to engage young children. We examined color qualia structures in children (3 to 12-y-old in Japan and 6 to 8-y-old in China) and compared these with those of Japanese adults. Approximately half of the assessed 3-y-old children completed the experiment via our touch panel device version of the task and had reliable responses. Despite known developmental and/or cultural effects of color term usage, we found that color qualia structures were quite similar across the age groups and cultures. Our finding supports the view that color qualia structures emerge early. We also observed age-related differences in the evaluations of some color pairs, which implied subtle changes in the structures behind color experience.

Age-related differences in the relationship between confidence and false memory in a mnemonic discrimination task

In addition to episodic memory loss there is an increase in false remembering in ageing especially when the discrimination between studied and new items is difficult in a recognition memory task. The aim of this study was to identify the underlying psychological mechanisms of this behavior, specifically, the possible role of false recollection. We used the Mnemonic Similarity Task, a widely used task in neuroscience research developed to assess the behavioral manifestation of hippocampal computations, pattern separation and pattern completion. First, older and young adults (n = 39 and 44, respectively) were presented with images of everyday objects. Then, on a surprise recognition test, they saw old (studied) and new (non-studied) items as well as visually similar lures of the images seen in the study phase. Instead of using the original Old/New test format, we asked participants to make confidence judgments. Our response frequency and ROC (receiver operating characteristics) analyses revealed overconfidence in false memories for the lures in the group of older adults suggesting false recollection. Such overconfidence was not observed for the completely new stimuli. Our results imply that older adults tend not to acknowledge some memory problems as a consequence of very high confidence in false memories.

Learning reshapes the hippocampal representation hierarchy

A key feature of biological and artificial neural networks is the progressive refinement of their neural representations with experience. In neuroscience, this fact has inspired several recent studies in sensory and motor systems. However, less is known about how higher associational cortical areas, such as the hippocampus, modify representations throughout the learning of complex tasks. Here, we focus on associative learning, a process that requires forming a connection between the representations of different variables for appropriate behavioral response. We trained rats in a space-context associative task and monitored hippocampal neural activity throughout the entire learning period, over several days. This allowed us to assess changes in the representations of context, movement direction, and position, as well as their relationship to behavior. We identified a hierarchical representational structure in the encoding of these three task variables that was preserved throughout learning. Nevertheless, we also observed changes at the lower levels of the hierarchy where context was encoded. These changes were local in neural activity space and restricted to physical positions where context identification was necessary for correct decision-making, supporting better context decoding and contextual code compression. Our results demonstrate that the hippocampal code not only accommodates hierarchical relationships between different variables but also enables efficient learning through minimal changes in neural activity space. Beyond the hippocampus, our work reveals a representation learning mechanism that might be implemented in other biological and artificial networks performing similar tasks.

V ex IR2V ec : An Architecture-Neutral Embedding Framework for Binary Similarity

Binary similarity involves determining whether two binary programs exhibit similar functionality with applications in vulnerability detection, malware analysis, and copyright detection. However, variations in compiler settings, target architectures, and deliberate code obfuscations significantly complicate the similarity measurement by effectively altering the syntax, semantics, and structure of the underlying binary. To address these challenges, we propose V ex IR2V ec , a robust, architecture-neutral approach based on VEX-IR to solve binary similarity tasks. V ex IR2V ec consists of three key components: a peephole extractor, a normalization engine (V ex INE), and an embedding model (V ex N et ). The process to build program embeddings starts with the extraction of sequences of basic blocks, or peepholes , from control-flow graphs via random walks, capturing structural information. These generated peepholes are then normalized using V ex INE, which applies compiler-inspired transformations to reduce architectural and compiler-induced variations. Embeddings of peepholes are generated using representation learning techniques, avoiding Out-Of-Vocabulary (OOV) issues. These embeddings are then fine-tuned with V ex N et , a feed-forward Siamese network that maps functions into a high dimensional space for diffing and searching tasks in an application-independent manner. We evaluate V ex IR2V ec against five baselines — BinDiff, DeepBinDiff, SAFE, BinFinder, and histograms of opcodes — on a dataset comprising \(2.7M\) functions and \(15.5K\) binaries from \(7\) projects compiled across \(12\) compilers targeting x86 and ARM architectures. The experiments span four adversarial settings — cross-optimization, cross-compilation, cross-architecture, and obfuscations — that are typically exploited by malware and vulnerabilities. In diffing experiments, V ex IR2V ec outperforms the nearest baseline in these four scenarios by \(40\%\) , \(18\%\) , \(21\%\) , and \(60\%\) , respectively. In the searching experiment, V ex IR2V ec achieves a mean average precision of \(0.76\) , the nearest baseline, by \(46\%\) . Our framework is highly scalable and is built as a lightweight, multi-threaded, parallel library using only open-source tools. V ex IR2V ec is \(\approx 3.1\) – \(3.5\times\) faster than the closest baselines and orders-of-magnitude faster than other tools.

Learning of the mean, but not variance, of color distributions cues target location probability

Humans are good at picking up statistical regularities in the environment. Probability cueing paradigms have demonstrated that the location of a target can be predicted based on spatial regularities. This is assumed to rely on flexible spatial priority maps that are influenced by visual context. We investigated whether stimulus features such as color distributions differing in mean and variance can cue location regularities. In experiment 1, participants searched for an oddly colored target diamond in a 6 × 6 set. On each trial, the distractors were drawn from one of two color distributions centered on different color averages. Each distribution was associated with different target location probabilities, one distribution where the target had an 80% chance to appear on the left (the rich location), while the rich location would be on the right for the other distribution. Participants were significantly faster at locating the target when it appeared in the rich location for both distributions, demonstrating learning of the relationship between color average and location probability. In experiments 2 and 3, observers performed a similar search task, but the distributions had different variances with the same average color. There was no evidence that search became faster when the target appeared in a rich location, suggesting that contingencies between target probabilities and color variance were not learned. These results demonstrate how statistical location learning is flexible, with different visual contexts leading to different spatial priority maps, but they also reveal important limits to such learning.

Multi-Omics Deep-Learning Prediction of Homologous Recombination Deficiency-Like Phenotype Improved Risk Stratification and Guided Therapeutic Decisions in Gynecological Cancers.

Homologous recombination deficiency (HRD) is a well-recognized important biomarker in determining the clinical benefits of platinum-based chemotherapy and PARP inhibitor therapy for patients diagnosed with gynecologic cancers. Accurate prediction of HRD phenotype remains challenging. Here, we proposed a novel Multi-Omics integrative Deep-learning framework named MODeepHRD for detecting HRD-positive phenotype. MODeepHRD utilizes a convolutional attention autoencoder that effectively leverages omics-specific and cross-omics complementary knowledge learning. We trained MODeepHRD on 351 ovarian cancer (OV) patients using transcriptomic, DNA methylation and mutation data, and validated it in 2133 OV samples of 22 datasets. The predicted HRD-positive tumors were significantly associated with improved survival (HR = 0.68; 95% CI, 0.60-0.77; log-rank p < 0.001 for meta-cohort; HR = 0.5; 95% CI, 0.29-0.86; log-rank p = 0.01 for ICGC-OV cohort) and higher response to platinum-based chemotherapy compared to predicted HRD-negative tumors. The translational potential of MODeepHRDs was further validated in multicenter breast and endometrial cancer cohorts. Furthermore, MODeepHRD outperforms conventional machine-learning methods and other similar task approaches. In conclusion, our study demonstrates the promising value of deep learning as a solution for HRD testing in the clinical setting. MODeepHRD holds potential clinical applicability in guiding patient risk stratification and therapeutic decisions, providing valuable insights for precision oncology and personalized treatment strategies.

TSO-HA*-Net: A Hybrid Global Path Planner for the Inspection Vehicles Used in Caged Poultry Houses

Traditional track-based inspection schemes for caged poultry houses face issues with vulnerable tracks and cumbersome maintenance, while existing rail-less alternatives lack robust, reliable path planners. This study proposes TSO-HA*-Net, a hybrid global path planner that combines TSO-HA* with topological planning, which allows the inspection vehicle to continuously traverse a predetermined trackless route within each poultry house and conduct house-to-house inspections. Initially, the spatiotemporally optimized Hybrid A* (TSO-HA*) is employed as the lower-level planner to efficiently construct a semi-structured topological network by integrating predefined inspection rules into the global grid map of the poultry houses. Subsequently, the Dijkstra’s algorithm is adopted to plan a smooth inspection route that aligns with the starting and ending poses, conforming to the network. TSO-HA* retains the smoothness of HA* paths while reducing both time and computational overhead, thereby enhancing speed and efficiency in network generation. Experimental results show that compared to LDP-MAP and A*-dis, utilizing the distance reference tree (DRT) for h2 calculation, the total planning time of the TSO-HA* algorithm is reduced by 66.6% and 96.4%, respectively, and the stored nodes are reduced by 99.7% and 97.4%, respectively. The application of the collision template in TSO-HA* results in a minimum reduction of 4.0% in front-end planning time, and the prior collision detection further decreases planning time by an average of 19.1%. The TSO-HA*-Net algorithm achieves global topological planning in a mere 546.6 ms, thereby addressing the critical deficiency of a viable global planner for inspection vehicles in poultry houses. This study provides valuable case studies and algorithmic insights for similar inspection task.

Development of Prompt Templates for Large Language Model-Driven Screening in Systematic Reviews.

Systematic reviews (SRs) are hindered by the initial rigorous article screen, which delays access to reliable information synthesis. To develop generic prompt templates for large language model (LLM)-driven abstract and full-text screening that can be adapted to different reviews. Diagnostic test accuracy. 48 425 citations were tested for abstract screening across 10 SRs. Full-text screening evaluated all 12 690 freely available articles from the original search. Prompt development used the GPT4-0125-preview model (OpenAI). None. Large language models were prompted to include or exclude articles based on SR eligibility criteria. Model outputs were compared with original SR author decisions after full-text screening to evaluate performance (accuracy, sensitivity, and specificity). Optimized prompts using GPT4-0125-preview achieved a weighted sensitivity of 97.7% (range, 86.7% to 100%) and specificity of 85.2% (range, 68.3% to 95.9%) in abstract screening and weighted sensitivity of 96.5% (range, 89.7% to 100.0%) and specificity of 91.2% (range, 80.7% to 100%) in full-text screening across 10 SRs. In contrast, zero-shot prompts had poor sensitivity (49.0% abstract, 49.1% full-text). Across LLMs, Claude-3.5 (Anthropic) and GPT4 variants had similar performance, whereas Gemini Pro (Google) and GPT3.5 (OpenAI) models underperformed. Direct screening costs for 10 000 citations differed substantially: Where single human abstract screening was estimated to require more than 83 hours and $1666.67 USD, our LLM-based approach completed screening in under 1 day for $157.02 USD. Further prompt optimizations may exist. Retrospective study. Convenience sample of SRs. Full-text screening evaluations were limited to free PubMed Central full-text articles. A generic prompt for abstract and full-text screening achieving high sensitivity and specificity that can be adapted to other SRs and LLMs was developed. Our prompting innovations may have value to SR investigators and researchers conducting similar criteria-based tasks across the medical sciences. None.

Transfer of balance performance depends on the specificity of balance training.

This study investigated whether a four-week training on an easy-level unstable board (EL) could induce a transfer of balance performance in a hard-level unstable board (HL) and in an unexpected perturbation-based task. Non-linear center of pressure (CoP) analysis investigated whether training could induce postural control adaptations in trained and untrained tasks. Thirty-four subjects were divided into a training (TR, N=17) and a control (CTRL, N=17) group. Balance was assessed before (T0) and after (T1) a balance training under static and dynamic conditions (EL, HL, and perturbation-based task). A force platform allowed the calculation of CoP displacement while balance performance based on the angular displacement of the unstable boards was assessed with an inertial sensor. From the angular displacement, we calculated three parameters of balance performance: Full Balance (FB), Fine Balance (FiB), and Gross Balance (GB). Stabilogram diffusion analysis (SDA) and sample entropy (SampEn) indirectly assessed neuromuscular control mechanisms. Results showed improvements in the TR from T0 to T1 in balance performance for FB (p<0.001), FiB (p<0.05), and GB (p<0.01) on EL and HL boards. In the perturbation-based task, the earliest CoP response consequent to perturbation improved after training (p<0.01). SampEn and SDA revealed increased automaticity (p<0.05) and efficiency (p<0.05) of balance control in the EL and HL tasks after training. Balance training led to highly task-specific adaptations and improvements that can be transferred between functionally similar balance tasks. Postural strategies learned during training seemed barely transferable to a different balance task, as the unexpected perturbation of the base of support.

Effective Hard Negative Mining for Contrastive Learning-Based Code Search

Background . Code search aims to find the most relevant code snippet in a large codebase based on a given natural language query. An accurate code search engine can increase code reuse and improve programming efficiency. The focus of code search is how to represent the semantic similarity of code and query. With the development of code pre-trained models, the pattern of using numeric feature vectors (embeddings) to represent code semantics and using vector distance to represent semantic similarity has replaced traditional string matching methods. The quality of semantic representations is critical to the effectiveness of downstream tasks such as code search. Currently, the state-of-the-art (SOTA) learning method uses the contrastive learning paradigm. The objective of contrastive learning is to maximize the similarity between matching code and query (positive samples) and minimize the similarity between mismatched pairs (negative samples). To increase the reusing of negative samples, prior contrastive learning approaches use a large queue (memory bank) to store embeddings. Problem . However, there is still a lot of room for improvement in using negative examples for code search: ① Due to the random selection of negative samples, semantic representations learned by existing models cannot distinguish similar codes well. ② Since semantic vectors in the memory bank are reused from previous inference results and then directly used for loss function calculation without gradient descent, the model cannot effectively learn the negative sample semantic information. Method . To solve the above problems, we propose a contrastive learning code search model with hard negative mining called CoCoHaNeRe: ❶ To enable the model to distinguish similar codes, we introduce hard negative examples into contrastive training, which are negative examples in the codebase that are most similar to positive examples. As a result, hard negative examples are most likely to make the model make mistakes. ❷ To improve the learning efficiency of negative samples during training, we add all hard negative examples to the model's gradient descent process. Result . To verify the effectiveness of CoCoHaNeRe, we conducted experiments on large code search datasets with six programming languages, as well as similar retrieval tasks code clone detection and code question answering. Experimental results show that our model achieves SOTA performance. In the code search task, the average MRR score of CoCoHaNeRe exceeds CodeBERT, GraphCodeBERT, and UniXcoder by 11.25%, 8.13%, and 7.38%, respectively. It has also made great progress in code clone detection and code question answering. In addition, our method performs well in different programming languages and code pre-training models. Furthermore, qualitative analysis shows that our model effectively distinguishes high-order semantic differences between similar codes.

Complexity and Time

Abstract A large literature shows that people’s valuation of delayed financial rewards violates exponential discounting, exhibiting a hyperbolic pattern: high short-run impatience that strongly decreases in the length of the delay. We test the hypothesis that the hyperbolic pattern in measured discount rates reflects mistakes driven by the complexity of evaluating delayed payoffs. We document that hyperbolicity (i) is strongly associated with choice inconsistency and cognitive uncertainty, (ii) increases in overt complexity manipulations and (iii) arises nearly identically in computationally similar tasks that involve no actual payoff delays. Our results suggest that even if people had exponential discount functions, complexity-driven mistakes would cause them to make hyperbolic choices. We examine which experimental techniques to estimate present bias are (not) confounded by information-processing constraints.

Middle school mathematics teachers’ proportional reasoning and its relation to their content and pedagogical content knowledge

ABSTRACT Proportional reasoning is an important but challenging skill for students and teachers. This article, presents findings from two studies investigating whether four categories of reasoning identified by Copur-Gencturk et al. (2022, A Closer Look at Teachers’ Proportional Reasoning. International Journal of Science and Mathematics Education, 21(1), 113–129) apply to a national sample of U.S. middle school mathematics teachers (N = 1,320) and can be captured consistently by similar tasks. We explored the meaningfulness of these reasoning categories with a new sample of middle school mathematics teachers (N = 224) investigating the role of reasoning categories in teachers’ content and pedagogical content knowledge. Both studies suggest that teachers’ proportional reasoning could be grouped into four categories: incorrect, additive, relative, and proportional. Additionally, the way teachers reason in novel proportional situations provides insights into their overall understanding of the content and their pedagogical content knowledge of ratios and proportional relationships.

Becoming Similar, but Drifting Apart: Partnerships Between Universities and Public Research Organizations

Abstract Many national research and innovation systems include higher education institutions and public research organizations (PRO) with different mandates and tasks. This paper investigates what happens to the relationship between a university and a PRO when they are increasingly pushed towards fulfilling similar tasks and functions. We investigate this through a historical case study of the relationship between a university and a PRO within the field of science and technology and draw on concepts from the institutional logics and institutional complexity literatures to frame the study. We find that in the early phase of the relationship, institutional complexity was handled through a strategy of “structural differentiation” where the university outsourced the commercial logic to the PRO, but in practice the two operated as integrated organizations. In the later phases, growing external demands and internal developments led to a “blending” strategy where the university reincorporated the commercial science logic. This had a fundamental impact on the relationship between the two organizations in terms of increased competition and blurring of roles, which led to increased structural and operational separation. Our study underlines the importance of considering wider, and often unintended, ramifications of policies on the research and innovation system as a whole.

DeepInterAware: Deep Interaction Interface-Aware Network for Improving Antigen-Antibody Interaction Prediction from Sequence Data.

Identifying interactions between candidate antibodies and target antigens is a key step in developing effective human therapeutics. The antigen-antibody interaction (AAI)occurs at the structural level, but the limited structure data poses a significant challenge. However, recent studies revealed that structural information can be learned from the vast amount of sequence data, indicating that the interaction prediction can benefit from the abundance of antigen and antibody sequences. In this study, DeepInterAware (deep interaction interface-aware network) is proposed, a framework dynamically incorporating interaction interface information directly learned from sequence data, along with the inherent specificity information of the sequences. Experimental results in interaction prediction demonstrate that DeepInterAware outperforms existing methods and exhibits promising inductive capabilities for predicting interactions involving unseen antigens or antibodies, and transfer capabilities for similar tasks. More notably, DeepInterAware has unique advantages that existing methods lack. First, DeepInterAware can dive into the underlying mechanisms of AAIs, offering the ability to identify potential binding sites. Second, it is proficient in detecting mutations within antigens or antibodies, and can be extended for precise predictions of the binding free energy changes upon mutations. The HER2-targeting antibody screening experiment further underscores DeepInterAware's exceptional capability in identifying binding antibodies for target antigens, establishing it as an important tool for antibodyscreening.

CigFacies: a massive-scale benchmark dataset of seismic facies and its application

Abstract. Seismic facies classification is crucial for seismic stratigraphic interpretation and hydrocarbon reservoir characterization but remains a tedious and time-consuming task that requires significant manual effort. Data-driven deep-learning approaches are highly promising for automating the seismic facies classification with high efficiency and accuracy, as they have already achieved significant success in similar image classification tasks within the field of computer vision (CV). However, unlike the CV domain, the field of seismic exploration lacks a comprehensive benchmark dataset for seismic facies, severely limiting the development, application, and evaluation of deep-learning approaches in seismic facies classification. To address this gap, we propose a comprehensive workflow to construct a massive-scale benchmark dataset of seismic facies and evaluate its effectiveness in training a deep-learning model. Specifically, we first develop a knowledge graph of seismic facies based on geological concepts and seismic reflection configurations. Guided by the graph, we then implement the three strategies of field seismic data curation, knowledge-guided synthesization, and generative adversarial network (GAN)-based generation to construct a benchmark dataset of 8000 diverse samples for five common seismic facies. Finally, we use the benchmark dataset to train a network and then apply it to two 3-D seismic data for automatic seismic facies classification. The predictions are highly consistent with expert interpretation results, demonstrating that the diversity and representativeness of our benchmark dataset are sufficient to train a network that can be generalized well in seismic facies classification across field data. We have made this dataset (https://doi.org/10.5281/zenodo.10777460, Gao et al., 2024a), the trained model, and the associated codes (https://doi.org/10.5281/zenodo.13150879, Gao et al., 2024b) publicly available for further research and validation of intelligent seismic facies classification.

Investigating Sensitivity to Shared Information and Personal Experience in Children's Use of Majority Information.

Children and adults alike rely on others to learn about the world, but also need to be able to determine the strength of both their own evidence as well as the evidence that other people provide, particularly when different sources of information disagree. For example, if two informants agree on a belief but share the same evidence, their testimony is statistically dependent on each other, and may be weaker evidence for that belief than two informants who draw on different pieces of evidence to support that belief. Across three experiments (total N = 492), we examine how 4- and 5-year-old children evaluate statistical dependency on a task where they must determine which of two jars that toys were drawn from. A majority of informants, whose testimony could draw from the same evidence or different evidence, always endorsed one jar. Then, children were presented with a dissenting informant or their own personal data that was consistent with the other jar. Children showed no sensitivity to statistical dependency, choosing the majority with equal probability regardless of the independence of their testimony, but also systematically overweighted their own personal data, endorsing the jar consistent with their own evidence more often than would be predicted by an optimal Bayesian model. In contrast, children made choices consistent with this model on a similar task in which the data was presented to children without testimony. Our findings suggest that young children treat majorities as broadly informative, but that the challenges of inferring others' experiences may lead them to rely on concrete, visible evidence when it is available.

Social Media Apps: A Paradigm for Examining Usability of Mobile Apps for Working Age Adults with Mild-Moderate Cognitive Disabilities

An estimated 79% of the US population used some form of social media regularly in 2019, highlighting a powerful platform for fostering connections with others. This aspect is particularly beneficial for people with cognitive disabilities (PwCDs) who may experience extreme social isolation. Studies have linked social media use amongst PwCDs to increased mental well-being and social capital, and PwCDs have reported corresponding increases in self-esteem and social connection. However, research also indicates that usability issues with social media user interfaces (UIs) often create barriers that deter participation of PwCDs. In this study, we investigate the usability of five social media apps for PwCDs, with the goal of identifying key UI features that facilitate/impede user performance and experience. The five apps selected for testing were: Facebook, Instagram, Pinterest, Snapchat, and Tumblr. Six to eight basic tasks were defined for each app, establishing a usability testing framework designed to compare similar tasks across apps. 19 working-age participants diagnosed with mild/moderate cognitive disabilities by their care provider were recruited and tested with each app across five different sessions. Mixed-methods data analysis and source of error analysis were performed for the resultant data. Snapchat and Pinterest were found to have the highest and lowest rates of task failure, respectively, consistent with the participant preference data on satisfaction and perceived usability. The fundamental task types with the highest rate of participant failure across all apps entailed either selecting or creating media, consistent with the identified qualitative theme that PwCDs tend to avoid using more active features in mobile apps due to their higher task complexity. Ten key types of usability issues were identified across all five apps, with the top issues identified as poor icon design and misleading icons, unclear user interface hierarchy, lack of user interface element separation and contrast, and similar task approaches with very different outcomes.

Contrastive Learning for Enhanced Retrieval on Online Medical Platforms

Online medical platforms have rapidly developed in recent years. Due to search inaccuracies and incomplete information provided by doctors, some suitable doctors are excluded during the first-stage retrieval. To enhance doctor retrieval and reduce user search costs, obtaining disease word embeddings is essential. Existing methods primarily rely on external corpora and sentence context, which may not align with downstream tasks, and many languages lack standardized medical corpora. Therefore, we integrate contrastive learning with medical knowledge to generate and augment data using simulations and propose a simple network structure for training on the constructed samples. We created two task datasets based on the platform’s data. Experimental results demonstrate that our framework achieves superior outcomes with lower embedding dimensions. In the similarity task, our framework attains an accuracy of 0.854, and in the retrieval task, it achieves an F1 score of 0.853, surpassing the current best results. Our framework has been successfully implemented on Baidu Health, one of the China’s largest online medical platforms, serving over 10 million users. This framework effectively simulates a doctor retrieval system, optimizing the process to ensure more accurate and comprehensive retrieval of suitable doctors.

A versatile machine learning workflow for high-throughput analysis of supported metal catalyst particles.

A versatile machine learning workflow for high-throughput analysis of supported metal catalyst particles.