Terms Of Frequency Research Articles

텍스트 마이닝을 활용한 성인 내담자의 온라인 진로상담 호소문제 분석: 커리어넷 온라인 진로상담 사례를 중심으로

Objectives The aim of this study is to present the basic data for efficient intervention of online career counseling by analyzing the career concerns from adult clients in online career counseling, using text mining. Methods A corpus consisting of 71 bulletin board text of adult clients aged 31 to 58 was gathered from CareerNet's open online career counseling from July 2019 to June 2021. Following data preprocessing, the text underwent analysis through word frequency, term frequency-inverse document frequency (TF-IDF), associated keyword network analysis, and topic modeling using latent Dirichlet allocation (LDA). Results First, it was analyzed what kind of career concerns are from the adult clients in online career counseling. The top keywords that appeared most often were ‘thinking, job, study, what, people, employment, company, concerns, university, preparation’, and in terms of word importance, words such as ‘career counseling, confirmation, lifelong learning, employment information’. The co-occurring keywords were analyzed, and keywords like ‘what’, ‘study’, ‘think’ appeared as words with high connection centrality and mediation centrality. In the LDA model analysis, they were categorized into five topics. Second, it was examined how they appeared differently by gender. The top frequency words were similar, but the importance of the words showed differences by gender. In the centrality model, connection centrality and mediation centrality were similar, but in the topic modeling, men were classified into four topics and women into three topics. Conclusions While the career concerns expressed in online counseling were similar to those expressed in face-to-face counseling, there were also words that indicated psychological difficulties. Awareness of the need for career information and assistance in solving career problems was found, especially in relation to life-long learning. Key concerns in early adulthood emerged, implied that career development is influenced by gender role stereotypes and that delayed career development affects lifespan development.

Mapping human security strategies to sustainable development goals: a machine learning approach

This paper introduces a machine learning (ML) based approach for integrating Human Security (HS) and Sustainable Development Goals (SDGs). Originating in the 1990s, HS focuses on strategic, people-centric interventions for ensuring comprehensive welfare and resilience. It closely aligns with the SDGs, together forming the foundation for global sustainable development initiatives. Our methodology involves a structured analysis of the relationship between 44 HS reports and the SDGs using advanced text analytics and machine learning methods, each report was evaluated based on its thematic relevance to HS aspects and SDG goals and sub-targets using a keyword-based matching approach. The data retrieved from the HS-related reports was transformed into a numerical format using the Term Frequency-Inverse Document Frequency (TF-IDF) method and classified through the Random Forest Classifier. Advanced models were utilized for exploration, namely BERT (Devlin et al. in Polit Sci, 2018) DistilBERT (Sanh et al. in ArXiv preprint arXiv: 1910.01108, 2019), and ELECTRA (Clark et al. in Polit Sci, 2020). This process contributed to the development of a web-based SDG mapping tool, specifically tailored for elaborating on the HS-SDG nexus. enabling the analysis of 13 new reports and their connections to the SDGs. This web application provides a streamlined interface for analyzing documents in PDF form and determining their relevance to the SDGs, allowing users to upload PDF files via an HTML form. Following expert consultation, our results outperform currently available similar online tools and assist in establishing strong links between the reports and global objectives, offering a nuanced understanding of the interplay between HS and sustainable development. This research provides a scalable framework to explore the relationship between HS and the global sustainability agenda, offering a practical, efficient resource for scholars and policymakers.

Security detection algorithm using CNN: Anomaly detection for API call sequence

This study proposes a security detection algorithm based on convolutional neural networks (CNNs) to enhance anomaly detection in API call sequences, addressing the challenges of capturing complex temporal relationships and nonlinear features in high-dimensional sparse API data. The proposed algorithm includes several preprocessing steps, such as deduplication to reduce redundancy, feature extraction using the TF-IDF (term frequency-inverse document frequency) algorithm, and logarithmic transformation to mitigate the impact of high-frequency APIs. An importance scoring mechanism is introduced to quantify the role of each API in anomaly detection. A customized TextCNN architecture is designed for API sequences, incorporating input layers, word embedding, multi-size convolution and pooling layers, attention layers, and fully connected layers. The attention layer is particularly applied to enhance the detection efficiency of evasion features. The model is trained using the Sigmoid activation function, CrossEntropyLoss loss function, and optimized via the Adam algorithm. The Softmax function is utilized to transform the feature vector into a probability distribution, with a threshold of 0.5 for anomaly detection. Clustering and auxiliary information are integrated to further improve classification accuracy and guide security strategy formulation. Experimental results demonstrate that the optimized TextCNN anomaly detection algorithm achieves an average accuracy of 95.88%, a recall rate of 91.23%, a false positive rate of 2.34%, and a false negative rate of 1.78%. These findings highlight the algorithm’s ability to enhance feature extraction accuracy, improve high-dimensional data processing, and provide an effective solution for real-time security monitoring, thus strengthening the security of the development environment.

Using Natural Language Processing (NLP) Methods to Predict Topics Included in 2019 Ohio Syphilis Disease Intervention Specialist (DIS) Records.

Free-text notes in disease intervention specialist (DIS) records may contain relevant information for STI control. In their current form, the notes are not analyzable without manual reading, which is labor-intensive and prone to error. We used natural language processing (NLP) methods to analyze 2019 Ohio DIS syphilis records with non-missing notes (n = 1,987). We identified 21 topics relevant for transmission and case investigations. We manually coded these records to create "gold standard" labels for each topic (0 = topic not present, 1 = topic present), then trained machine learning models to identify the topics in the text. For models to analyze text data, the text must be converted to numbers. We explored two approaches to numerically represent words: (1) term frequency, inverse document frequency (TF-IDF), which measures importance of words based on how many times they appear in a record and in the dataset as a whole, and (2) GloVe embeddings, which are numerical vectors that were developed by researchers for each word in the English language to encode its semantic meaning. We explored three types of statistical models (naïve Bayes, support vector machine [SVM], and logistic regression) using TF-IDF, and one type of neural network model (long short-term memory [LSTM] model) using GloVe. All models were used for binary prediction (i.e., topic not present, topic present). For most topics, the LSTM model performed the best overall in identifying topics, and the SVM model performed the best among the statistical models. For example, the LSTM model predicted the topic "substance use" with high accuracy (97%), sensitivity (92%), and specificity (98%). No model performed well for uncommon topics (e.g., "alcohol use" or "delays in care"). Machine learning models performed well in identifying some topics in 2019 Ohio syphilis records. This analysis is a first step in applying NLP methods to making DIS notes more accessible for analysis.

Word-of-Mouth Evaluation of Ancient Towns in Southern China Using Web Comments

With the rapid development of digital networks and communication technologies, traditional word-of-mouth (WOM) has transformed into electronic word-of-mouth (eWOM), which plays a pivotal role in improving the management and service quality of ancient town tourism. This study uses Python web scraping techniques to gather eWOM data from the top ten ancient towns in southern China. Using IPA analysis, the analytic hierarchy process (AHP), Term Frequency–Inverse Document Frequency (TF-IDF), and cluster analysis, we developed a comprehensive eWOM evaluation framework. This framework was employed to perform word frequency analysis, sentiment analysis, topic modeling, and rating analysis, providing deeper insights into tourists’ perceptions. The results reveal several key findings: (1) Transportation infrastructure varies significantly across the towns. Heshun and Huangyao suffer from poor accessibility, while the remaining towns benefit from the developed transportation network of the Yangtze River Delta. (2) The volume of eWOM is strongly influenced by seasonal patterns and was notably impacted by the COVID-19 pandemic. (3) The majority of tourists express positive sentiments toward the ancient towns, with a focus on the available facilities. Their highest levels of satisfaction, however, are associated with the scenic landscapes. (4) A comprehensive eWOM analysis suggests that Wuzhen and Xidi–Hongcun are the most popular tourist destinations, while Zhujiajiao, Huangyao, Zhouzhuang, and Nanxun exhibit lower levels of both attention and visitor satisfaction.

Machine Learning Base Spam Comments Detection on YouTube

The project proposes a machine learning- based approach for detecting spam comments on YouTube using Natural Language Processing (NLP) algorithms. As user-generated content continues to grow exponentially, spam comments pose a significant threat to user experience and platform integrity. Leveraging a labeled dataset of comments, the system employs various NLP preprocessing techniques, including tokenization, stop word removal, and stemming, to clean and prepare the text data. Advanced feature extraction methods such as Term Frequency-Inverse Document Frequency (TF- IDF) and word embeddings are used to capture semantic nuances in the comments. Machine learning classifiers like Logistic Regression (LR), Support Vector Machines (SVM), and Random Forest (RF) are then applied to classify comments as spam or legitimate. Accuracy, precision, recall, and F1-score are employed to determine the models in order to ensure a good balance between false positives and negatives. This research enhances content moderation capabilities by filtering spam and providing insights into spam trends, contributing to improved online content management and a better user experience on social media platforms. Index Terms- Machine Learning, Spam Detection, YouTube, Natural Language Processing, NLP, Classification, Text Processing, Feature Extraction.

Hybrid Machine Learning and Deep Learning Approaches for Insult Detection in Roman Urdu Text

Thisstudy introduces a new model for detecting insults in Roman Urdu, filling an important gap in natural language processing (NLP) for low-resource languages. The transliterated nature of Roman Urdu also poses specific challenges from a computational linguistics perspective, including non-standardized grammar, variation in spellings for the same word, and high levels of code-mixing with English, which together make automated insult detection for Roman Urdu a highly complex problem. To address these problems, we created a large-scale dataset with 46,045 labeled comments from social media websites such as Twitter, Facebook, and YouTube. This is the first dataset for insult detection for Roman Urdu that was created and annotated with insulting and non-insulting content. Advanced preprocessing methods such as text cleaning, text normalization, and tokenization are used in the study, as well as feature extraction using TF–IDF through unigram (Uni), bigram (Bi), trigram (Tri), and their unions: Uni+Bi+Trigram. We compared ten machine learning algorithms (logistic regression, support vector machines, random forest, gradient boosting, AdaBoost, and XGBoost) and three deep learning topologies (CNN, LSTM, and Bi-LSTM). Different models were compared, and ensemble ones were proven to give the highest F1-scores, reaching 97.79%, 97.78%, and 95.25%, respectively, for AdaBoost, decision tree, TF–IDF, and Uni+Bi+Trigram configurations. Deeper learning models also performed on par, with CNN achieving an F1-score of 97.01%. Overall, the results highlight the utility of n-gram features and the combination of robust classifiers in detecting insults. This study makes strides in improving NLP for Roman Urdu, yet further research has established the foundation of pre-trained transformers and hybrid approaches; this could overcome existing systems and platform limitations. This study has conscious implications, mainly on the construction of automated moderation tools to achieve safer online spaces, especially for South Asian social media websites.

Paul’s Style and the Problem of the Pastoral Letters: Assessing Statistical Models of Description and Inference

Abstract In this study, we combine the core competencies of a New Testament scholar and a mathematician to conduct a stylometric analysis of the Pauline and Pastoral Letters based on 18 epistolary modes of discourse, which we argue are at least as stylistically relevant as traditional features such as term frequencies. By focusing on features of this type, we apply discrete probability models – namely, the binomial, Poisson, and negative binomial distributions – to determine which ones are the best fit for each feature. With our approach, we have found a way to utilize Paul’s authentic letters as a standard of measurement by which the style(s) of the Pastorals can be compared, an essential first step to statistical inference that is generally lacking in Pauline stylometry. With few exceptions, we find that the Pastoral Letters do not contain stylistic differences with the authentic Paulines at a statistically significant level, a finding that disputes many previous studies (especially in the qualitative tradition) that assume stylistic difference, but which is in line with an emerging trend in computational analyses of the Pauline and Pastoral Letters.

Fraud-BERT: transformer based context aware online recruitment fraud detection

Online recruitment facilitates the automatic hiring process for recruiters and provides convenience to job seekers via online job platforms. Parallelly, it has given rise to malicious use of such platforms by fraudsters who post fake jobs and steal money and personal information from innocent job seekers. It is difficult to detect fake jobs manually, as these are meticulously crafted to mimic legitimate ones. Previously, various machine learning approaches have employed Bag-of-Words (BoW) and Term Frequency-Inverse Document Frequency (TF-IDF) feature extraction methods for this task. However, these methods are non-contextual and show skewness in results due to imbalances in data distribution. This paper presents Fraud-BERT, a transformer-based contextual framework leveraging Bidirectional Encoder Representations from Transformers (BERT) via transfer learning approach and evaluates it on a highly imbalanced fake job dataset, which is popularly named as Employment Scam Aegean Dataset (EMSCAD). The dataset is available on Kaggle website. The superiority of the proposed method is demonstrated by a comparative analysis with conventional methods. The results of the study conclude that the proposed method is more robust in tackling imbalanced data and, it has significantly out-performed existing state-of-the-art studies with F1 score of 0.93 and 99% accuracy.

Account of Spatio-Temporal Characteristics in Customs Anti-Smuggling Intelligence Acquisition

The related information on smuggling crimes exists extensively in various media, with multiple data sources. Anti-smuggling intelligence faces the contradiction between the explosive growth of data size and high-efficiency intelligence judgment. Considering the current characteristics of smuggling activities, it is urgent to obtain knowledge from multi-source case data. Aiming to explore a smuggling knowledge acquisition algorithm based on deep learning, this study proposed an anti-smuggling knowledge representation model with both temporal and spatial characteristics and a knowledge-driven anti-smuggling intelligent judgment method. By combining two means, data, information, knowledge, and intelligence were effectively fused via the Term Frequency-Inverse Document Frequency (TF-IDF) technique and Levenshtein distance algorithms, promoting deep mining and application of anti-smuggling big-data resources and enhancing both automation and intelligence levels in anti-smuggling intelligence judgment.

Genetic Transfer in Action: Uncovering DNA Flow in an Extremophilic Microbial Community.

Horizontal genetic transfer (HGT) is a significant driver of genomic novelty in all domains of life. HGT has been investigated in many studies however, the focus has been on conspicuous protein-coding DNA transfers that often prove to be adaptive in recipient organisms and are therefore fixed longer-term in lineages. These results comprise a subclass of HGTs and do not represent exhaustive (coding and non-coding) DNA transfer and its impact on ecology. Uncovering exhaustive HGT can provide key insights into the connectivity of genomes in communities and how these transfers may occur. In this study, we use the term frequency-inverse document frequency (TF-IDF) technique, that has been used successfully to mine DNA transfers within real and simulated high-quality prokaryote genomes, to search for exhaustive HGTs within an extremophilic microbial community. We establish a pipeline for validating transfers identified using this approach. We find that most DNA transfers are within-domain and involve non-coding DNA. A relatively high proportion of the predicted protein-coding HGTs appear to encode transposase activity, restriction-modification system components, and biofilm formation functions. Our study demonstrates the utility of the TF-IDF approach for HGT detection and provides insights into the mechanisms of recent DNA transfer.

Evaluating the effectiveness of machine learning methods for keyword coverage using semantic data analysis

This article presents a comprehensive comparative analysis of two advanced hybrid machine learning approaches for keyword extraction: bidirectional encoder representations from transformers (BERT) combined with autoencoder (AE) and term frequency-inverse document frequency (TF-IDF) combined with autoencoder. The research targets the task of semantic analysis in text data to evaluate the effectiveness of these methods in ensuring adequate keyword coverage across diverse text corpora. The study delves into the architecture and operational principles of each method, with a particular focus on the integration with autoencoders to enhance the semantic integrity and relevance of the extracted keywords. The experimental section provides a detailed performance analysis of both methods on various text datasets, highlighting how the structure and semantic richness of the source data influence the outcomes. The evaluation methodology includes precision, recall, and F1-score metrics. The paper discusses the advantages and disadvantages of each approach and their suitability for specific keyword extraction tasks. The findings offer valuable insights for the scientific community, aiding in the selection of the most appropriate text processing method for applications requiring deep semantic understanding and high accuracy in information extraction.

Arabic fake news detection using hybrid contextual features

Technology has advanced and social media users have grown dramatically in the last decade. Because social media makes information easily accessible, some people or organizations distribute false news for political or commercial gain. This news may influence elections and attitudes. Even though English fake news is widely detected and limited, Arabic fake news is hard to recognize owing to a lack of study and data collection. Wara Arabic bidirectional encoder representations from transformers (WaraBERT), a hybrid feature extraction approach, combines word level tokenization with two Arabic bidirectional encoder representations from transformers (AraBERT) variants to provide varied features. The study also discusses eliminating stopwords, punctuations, and tanween markings from Arabic data. This study employed two datasets. The first, Arabic fake news dataset (AFND), has 606,912 records. Second dataset Arabic news (AraNews) has 123,219 entries. WaraBERT-V1 obtained 93.83% AFND accuracy using the bidirectional long short-term memory (BiLSTM) model. However, the WaraBERT-V2 technique obtained 81.25%, improving detection accuracy above previous researchers for the AraNews dataset. These findings show that WaraBERT outperforms word list techniques (WLT), term frequency-inverse document frequency (TF-IDF), and AraBERT on both datasets.

Analysis of Suicide-related Tweets During the COVID-19 Pandemic

Background: The COVID-19 virus started in 2019 and badly affected the different sectors of many countries around the world. Based on this, financial difficulties, loss of loved ones, sudden anger, relationships, family disputes, and psychological distress increased, and individuals were stalled from carrying out their lifestyle in a normal way, and some individuals were even motivated to commit suicide. Objective: It is important to reduce the number of suicides and identify the reasons for this situation. Through this research, the focus is on identifying the main topics discussed relevant to suicides during the COVID-19 pandemic. Methods: Individuals use Twitter, a social media platform, to share their ideas freely and publically. We collected 9750 primary data through Twitter API (Application Programming Interface). After preprocessing and feature extraction by TF-IDF (Term Frequency-Inverse Document Frequency), we applied the LDA (Latent Dirichlet Allocation) and Probabilistic Latent Semantic Analysis (PLSA) topic modeling algorithms to identify topics. Results: Based on the LDA results, we extracted ten different topics under the three themes, such as the impact of COVID-19, human feelings, getting support, and having awareness. Intertopic Distance Map, Most Salient Terms, and Word Clouds Visualization are used to check the results. The coherence score and perplexing value are used to measure how interpretable the extracted topics are to humans. PLSA also extracted 25 topics with their probabilities, and Kullback–Leibler (KL) divergence was used to check the results. Conclusion: We were able to gain insight into human emotions and the main motivations behind suicide attempts using the topics we extracted. Expert feedback proved that LDA results were better than PLSA. Based on that, we found the main impact of COVID-19 on human lives, how human feelings were changed positively and negatively during that period, what supporting and awareness methods people used, and what they preferred. The required measures can then be taken by those responsible authorities and individuals to prevent, reduce, and get ready for this kind of suicidal incident in the future.

Unlocking Potential Score Insights of Sentimental Analysis with a Deep Learning Revolutionizes

Online hate has emerged as a rapidly growing issue worldwide, often stemming from differences in opinion. It is crucial to use appropriate language and words on social media platforms, as inappropriate communication can negatively impact others. Consequently, detecting hate speech is of significant importance. While manual methods are commonly employed to identify hate and offensive content on social media, they are time-consuming, labor-intensive, and prone to errors. Therefore, AI-based approaches are increasingly being adopted for the effective classification of hate and offensive speech. The proposed model incorporates various text preprocessing techniques, such as removing extraneous elements like URLs, emojis, and blank spaces. Following preprocessing, tokenization is applied to break down the text into smaller components or tokens. The tokenization technique utilized in this study is TF-IDF (Term Frequency–Inverse Document Frequency). After tokenization, the model performs the classification of hate and offensive speech using the proposed BiLSTM-based SM-CJ (Scalable Multi-Channel Joint) framework. The BiLSTM-based SM-CJ model is particularly effective in detecting hate, offensive, and neutral tweets due to its ability to capture both forward and backward contexts within a given text. Detecting hate speech requires a comprehensive understanding of the text and the identification of patterns spanning across multiple words or phrases. To achieve this, the LSTM component of the BiLSTM model is designed to capture long-term dependencies by utilizing information from earlier parts of the text. The proposed SM-CJ framework further aligns the input sequence lengths fetched from the input layer, enabling the model to focus on specific segments of the input sequence that are most relevant for hate speech detection. This approach allows the model to accurately capture derogatory language, and subtle nuances present in hate speech. Finally, the performance of the proposed framework is evaluated using various metrics, including accuracy, recall, F1-score, and precision. The results are compared with state-of-the-art approaches, demonstrating the effectiveness of the proposed model. Doi: 10.28991/ESJ-2025-09-01-03 Full Text: PDF

Computational analysis of virus-host protein-protein interactions using gene ontology and natural language processing

The role of in-silico computational methods in identifying protein-protein interactions (PPIs) between target and host proteins is crucial for developing effective infection treatments. These methods are essential for quickly determining high-quality and accurate PPIs, predicting protein pairs with the highest likelihood of physical interaction from a large pool, and reducing the need for experimental confirmation or prioritizing pairs for experiments. This study proposes using gene ontology and natural language processing (NLP) approaches to extract and quantify features from protein sequences. In the first step, proteins were represented using gene ontology terms, and a set of features was generated. In the second step, NLP techniques treated gene ontology terms as a word dictionary, creating numerical vectors using the bag of words (BoW), count vector, term frequency-inverse document frequency (TF-IDF), and information content methods. In the third step, different machine learning methods, including Decision Tree, Random Forest, Bagging-RepTree, Bagging-RF, BayesNet, Deep Neural Network (DNN), Logistic Regression, Support Vector Machine (SVM), and VotedPerceptron, were employed to predict protein interactions in the datasets. In the fourth step, the Max-Min Parents and Children (MMPC) feature selection algorithm was applied to improve predictions using fewer features. The performance of the developed method was tested on the SARS-CoV-2 protein interaction dataset. The MMPC algorithm reduced the feature count by over 99%, enhancing protein interaction prediction. After feature selection, the DNN method achieved the highest predictive performance, with an AUC of 0.878 and an F-Measure of 0.793. Sequence-based protein encoding methods AAC, APAAC, CKSAAPP, CTriad, DC, and PAAC were applied to proteins in the SARS-CoV-2 interaction dataset and their performance was compared with GO-NLP. The performance of the relevant methods was measured separately and combined. The highest performance was obtained from the combined dataset with an AUC value of 0.888. This study demonstrates that the proposed gene ontology and NLP approach can successfully predict protein-protein interactions for antiviral drug design with significantly fewer features using the MMPC-DNN model.

A Combined Smoothing Method of Ensemble Pulsar Timescale and Application for Pulsar Atomic Clock Combined Timescale

Abstract Clock difference between Ensemble pulsar timescale (PT) and the International Atomic Time (TAI) PT-TAI derived from the International Pulsar Timing Array (IPTA) data set indicates very similar variation trend with the Terrestrial Time TT(BIPMXXXX)-TAI but PT has larger measurement error. In this paper we discuss smoothing method of PT using combined smoothing filter and compare the results with that from other filters. The clock difference sequence between PT-TAI and the first time derivative series of the TT(BIPMXXXX)-TAI can be combined by combined smoothing filter to yield two smooth curves tied by the constraints assuring that the latter is the derivative of the former. The ensemble pulsar time IPTA2016 with respect to TAI published by G. Hobbs et al and first time derivative series of the TT(BIPM2017)-TAI with quadratic polynomial terms removed are processed by combined smoothing filter in order to demonstrate property of the smoothed results. How to correctly estimate two smoothing coefficients is described and the output results of combined smoothing filter are analyzed. The results show that the combined smoothing method efficiently removed high frequency noises of two input data series and the smoothed data of the PT-TAI combined long term fractional frequency stability of the pulsar time and frequency accuracy of the terrestrial time. Fractional frequency stability analysis indicates that both short and medium time interval stability of the smoothed PT-TAI is improved while keeping its original long term frequency stability level. The combined smoothing filter is more suitable for smoothing observational pulsar timescale data than any filter that only make smoothing of single pulsar time series. The smoothed pulsar time by combined smoothing filter is a pulsar atomic time combined time scale. This kind of combined time scale can also be used as terrestrial time.

Ofsted inspection reports in early childhood education settings narrowly focussed: A corpus and sentiment analysis

High-quality early childhood education is key for children’s positive and long-term outcomes. Countries differ on how they quality assure provision. In England, Ofsted is the independent inspection body, often the source of controversy for lack of transparency in ratings. However, there is a dearth of evidence on how inspectors’ judgements are made, particularly in the early years sector. This study examines the consistency and nature of individual judgements in Ofsted reports, through an in-depth corpus analysis of inspection reports for non-domestic early years providers (nurseries and preschools) in England. It is the first to analyse the extent to which inspections focus on well-regarded understandings of high-quality early childhood pedagogy and provides an appraisal of the affective tone of the reports, using sentiment analysis. Results show that inspectors are fairly consistent, with similar high frequent terms present across reports, but the inspection focus may be too limited in scope. There is an over-emphasis on school readiness based on outcomes of older children, without clear references to the participation and engagement of younger children and process variables. A more positive tone in the reports is more frequent in high-quality rated settings, except for the section describing children’s outcomes, the shortest across reports.

Enhancing Mental Health Disorder Detection: A Hybrid Classifier System Using Soft Voting and Ensemble Methods

Early detection and diagnosis of mental health disorders based on patient-reported symptoms is vital in mental health care. This research proposes an ensemble learning pipeline for predicting mental health disorders using symptoms and demographic data. The pipeline incorporates preprocessing techniques such as standardization, one-hot encoding, and Term Frequency - Inverse Document Frequency transformation, and addresses class imbalance with the Synthetic Minority Over-sampling Technique to enhance predictions of rare disorders. Multiple classifiers, including Random Forest, Gradient Boosting, XGBoost, CatBoost, and LightGBM, are combined in a voting classifier, with each model optimized using Grid Search. The pipeline's performance is evaluated on its ability to predict mental health disorders, showing potential for supporting clinical diagnosis. CatBoost proves effective in certain models, while LightGBM delivers strong performance across most. The models achieve around 95% accuracy, demonstrating strong discrimination between disorders and other conditions.

P-1891. What’s the Reason for the Consult? Identifying Clusters of Infectious Disease Consultations in an Academic Health System

Abstract Background In many institutions, infectious diseases (ID) consultations are requested via electronic order entry and the reason for consultation is included as free text in the order. We employed an unsupervised clustering algorithm to determine whether the consult order text can be used to ascertain clinically meaningful groups among patients for whom ID consultation is requested. Figure 1. Example Word Cloud ‘Clusters’ of Free-Text Reasons for Infectious Disease Consultation and Corresponding Subject Matter Expert Interpretation Methods We examined the initial ID consultation for all Duke University Health System inpatients receiving consultation from 1/1/2014 - 12/31/2023. The free text “Reason for Consult?” field in the consultation order was extracted, transformed into a weighted vector format using Term-Frequency Inverse-Document Frequency, and categorized using an unsupervised K-means clustering algorithm; Algorithm parameters were adjusted to optimize clinical relevance of the clusters. Within each patient cluster, we evaluated mortality and hospital length of stay, with patients censored at their time of last contact with the health system.Figure 2.Kaplan-Meier Survival Estimates for Each Cluster Results 37,848 unique patients with an ID consultation during the study period were evaluated, and 35,197 had free-text reasons associated with the electronic consult order. A combination of the K-means Elbow Method, silhouette score, and subject matter expert evaluation identified that approximately 10 clusters were optimal to categorize the dataset. Word clouds were generated for each cluster (Figure 1) and expert assessment identified the central theme of each cluster (Table 1). We identified unique patient characteristics or syndromes in nine of the patient clusters, while the final cluster was characterized as a ‘generic’ ID consult. One-year mortality varied according to cluster (Figure 2) with the highest mortality seen in the ‘Hematologic Malignancy’ category, and lowest mortality in the ‘Complicated Wound Infection’ and ‘HIV Management’ categories. Conclusion An unsupervised clustering algorithm allowed for the identification of clinically meaningful groups of ID consultations. If validated in other settings, this approach could present an accessible way to understand qualitative and prognostic trends in ID clinical practice.Table 1.Cluster Characteristics*Institutional policy at the end of study period requires Infectious Disease consultation for patients with HIV, new lung transplant recipients, patients with candidemia or Staphylococcus aureus bacteremia, and patients for whom restricted antibiotics are requested.**Infinite median survival times indicate that the Kaplan-Meier survival curve did not cross the 50% threshold through the evaluated period. Disclosures Kristen Dicks, MD, UpToDate: Advisor/Consultant Nicholas A. Turner, MD, MHSc, PDI: Research contract|Purio Labs: Research contract