Aim/Purpose.
This paper addresses the challenge of emotional analysis in Hebrew texts, specifically focusing on enhancing machine learning techniques with psychological feature lexicons to improve classification accuracy in identifying depression.

Background.
Emotional analysis in Hebrew texts presents unique challenges due to the language's intricate morphology and rich derivation system. This paper seeks to leverage advanced machine learning methods augmented with carefully crafted psychological feature lexicons to address these challenges and improve the identification of depression from online discourse.

Methodology.
The study involves scraping and analyzing a dataset consisting of over 350K posts from 25K users on the "Camoni" health-related social network spanning 2010-2021. Various machine learning models, including SVM, Random Forest, Logistic Regression, and Multi-Layer Perceptron, were employed alongside ensemble methods such as Bagging, Boosting, and Stacking. Features were selected using TF-IDF, incorporating both word and character n-grams (Aisopos et al., 2016; HaCohen-Kerner et al., 2018). Pre-processing steps, including punctuation removal, stop word elimination, and lemmatization, were applied, to handle the challenges in Hebrew as a reach morphological language (Amram et al., 2018; Tsarfaty et al., 2019). Then hyperparameter tuning was conducted to optimize model performance across different languages. Following this, the models were enriched with features extracted from sentiment lexicons conducted by professional psychologists. (Shapira et al., 2021).

Contribution.
This paper contributes to the field by demonstrating the efficacy of integrating psychological feature lexicons into machine-learning models for emotional analysis in Hebrew texts. Addressing the unique linguistic challenges, it advances the understanding of depression detection in online communities and informs the development of more effective preventive measures and treatments.

Findings.
Through experimentation, it was discovered that enriching the models with features from sentiment lexicons significantly improved classification accuracy. Among the sentiment lexicons tested, six were identified as particularly enchanting: Negative emojis, positive emojis, neutral emojis, Hostile words, Anxiety words, and No-Trust words. The coverage and the quality of a feature lexicon are and may contribute to the success of various tasks like opinion mining and sentiment analysis (Feldman, 2013; Liu, 2012; Yang et al., 2020).

Recommendations for Practitioners.
Practitioners in mental health and social work should prioritize enriching machine learning models with sentiment lexicons to enhance the accuracy and effectiveness of depression detection in online discourse. By incorporating lexicons capturing emotional nuances, practitioners can improve the sensitivity of their screening processes.

Recommendations for Researchers.
Future research endeavors should focus on further refining machine learning models by enriching them with sentiment lexicons. Additionally, exploring the integration of sentiment lexicons into deep learning models could provide further insights into the classification of emotional content in textual data.

Impact on Society.
The findings have significant implications for the development of more accurate and efficient methods for detecting depression in online Hebrew discourse. By leveraging advanced machine learning techniques augmented with psychological feature lexicons, this research contributes to enhancing mental health interventions and promoting well-being in online communities.

Future Research.
Future research should not only continue exploring the integration of sentiment lexicons into machine learning models but also extend this investigation to deep learning architectures. Investigating the effectiveness of sentiment lexicons in enhancing the performance of deep learning models could advance our understanding of emotional analysis in textual data and improve depression detection algorithms.