Modern Standard Arabic mood changing and depression dataset

This paper presents Modern Standard Arabic data for the automatic estimation of the risk of depression for online personas based on their daily Arabic tweets. The data were collected from 1-1-2020 to 1-1-2021 using automatically collected samples of depression and non-depression tweets. The data contain 1229 records. These data can be used to develop machine-learning tools to identify the risk of an individual being depressed and to build recommender systems that monitor depression.


Specifications
Data science Specific-subject area Data science Type of data .csv file How data were acquired Using TWINT tool from python Data format Raw data (.csv) Description of data collection The raw data of Twitter posts were downloaded using the TWINT tool from Python. The collection process was conducted from 1-1-2020 to 1-1-2021. An Arabic version of the Patient Health Questionnaire (PHQ9) scale was used to specify the categories of the depression dataset. The categories are: (1) losing interest or pleasure in activities, (2) low mood, (3) sleep disorder, (4) loss of energy, (5) weight disorder, (6) feelings of worthlessness, (7) diminished ability to think or concentrate, (8) psychomotor agitation or retardation, and (9) suicidality. Keywords (terms or phrases) that might fall under each of the categories were created, including terms or phrases a person would use to describe their moods, anxiety, and sleep modules, such as ‫ةنيزح"‬ "," ‫ينيف‬ ‫قرأ‬ " and ‫اي"‬ ‫بر‬ ‫يتايح‬ ‫يهتنت‬ " These keywords were then used to collect relevant tweets from Twitter

Value of the Data
• This dataset contains information about real users of Arabic tweets to acquire knowledge about a person's psychology. It can be used for the early detection of depression types for rapid intervention, which can potentially reduce the suffering from the disorder and the stigma associated with mental illness. • This dataset provides a unique opportunity for researchers, primary care clinics, and psychologists to detect users' depression early (12 different mental health disorders, including depressed mood, loss of interest, change in appetite or weight, feeling guilty or worthless, sleep disorder, psychomotor agitation or impairment, fatigue or loss of energy, poor concentration and persistent thoughts of death or suicidal ideation) based on their last 14 days of tweets. • This dataset can be used to automatically complete the Patient Health Questionnaire (PHQ-9), the most well-known psychometric symptoms report for measuring depression symptoms and mental illnesses [1] . • This dataset can help researchers easily build machine-learning and classification models to detect users' depression types. • This is a language-dependent dataset designed for the Arabic language with languagedependent features. However, researchers can recreate a similar dataset for other languages, such as English, by following the same steps explained in this work.

Data Description
This paper presents Modern Standard Arabic data for the automatic estimation of the risk of depression for online personas based on their daily Arabic tweets. The data were collected from 1-1-2020 to 1-1-2021 using automatically collected samples of depression and non-depression tweets. The final data contain 1229 records. These data can be used to develop machine-learning tools to identify the risk of an individual being depressed and to build recommender systems

Experimental Design and Materials and Methods
An Arabic version of the Patient Health Questionnaire (PHQ-9) scale [1] , a well-validated measurement tool for depression, was used in this manuscript. Although it cannot be used formally as a standalone assessment to diagnose clinical depression, it can be used as the first step to screen and detect depression [2,3] . It is commonly used in depressive symptoms selfadministered identification tools, online design studies, research, and psychophysical clinics during the first appointment of subjects.
Subjects are required by PHQ-9 to respond by one of the following numbers: 0, 1, 2, or 3 to nine psychophysical questions that indicate depression. Each of these numbers indicates a certain degree of depression, as shown in Table 2 . The PHQ9 scale questions are shown in Table 1 . In these questions, users are asked about their psychophysical state in the last two weeks (14 days), emotions (e.g., feeling down, tired, tired, anxious), appetite, the occurrence of any sleeping disorder, slowness in movement, and suicidal thoughts [1] .
PHQ-9 scores have been shown in previous studies to be strongly correlated with the results of psychologist depression diagnosis in the psychophysical clinic [4] . Generally, subjects with scores ranging from 0 to 9 are considered healthy, subjects with scores ranging from 10 to 19 are considered to have mild depression, and subjects with scores of 20 points or above are considered severely depressed.
Although depression detection is the first step in treating it [2] , more than a quarter of most depression patients remain undiagnosed [3] . The main issue is that most patients dislike traditional depression detection methods, such as the paper version of the PHQ-9 scale. In summary, subjects find the questionnaire intrusive, cumbersome, and feared [5] .
Twitter data were collected in this work to deal with this issue. It used to automatically answer the nine questions of PHQ-9 for depression screening and detection. Researchers can    Several days An indefinite small quantity that is more than two but less than many.
More than 1/2 the days A time period that includes one-half of the total number of days plus at least one more. use this dataset to develop an automatic mental health assessment methodology in which the depression levels of the subjects are detected. This work began by creating categories for each of these questions, as shown in Table 3 . We then chose keywords (terms or phrases) that could fall under each of the categories. The keywords included terms or phrases a person would use to describe their moods, anxiety, and sleep modules, as shown in Table 4 . The individual to be diagnosed with depression must experience five or more symptoms during the last two-week (14 days) period of the following nine categories: losing interest or pleasure in activities, low mood, sleep disorder, weight disorder, loss of energy, feelings of worthlessness, diminished ability to think or concentrate, psychomotor agitation or retardation, and suicidality [1] . These keywords were used to collect relevant tweets from Twitter using the TWINT library as follows [6] : • Python programming language with the TWINT library was used to collect relevant tweets.

Labeling the dataset
Each record is labeled by one category name (there are nine categories, as shown in Table 3 ).

Cleaning the dataset
Cleaning is an essential step in almost any Natural Language Processing (NLP) task. It aims to eliminate incomplete, noisy, and inconsistent data. The dataset was analyzed before cleaning using Pandas Profile Report tools Fig. 4 . shows the results of this analysis. According to this figure, the biggest problem with this dataset is missing cells (missing values) at 34.9%. Therefore, two cleaning steps were followed.

Cleaning the dataset file:
• Removing empty columns: removed 10 empty columns that did not contain any value.
• Removing unwanted columns: removed 24 unwanted columns that did not contain any important value. • Removing duplicated records: removed duplicated rows (0.04%).      Cleaning tweets: • Removing URLs: removed Tweets links that did not contribute to a depression classification.
• Removing duplicated letters: replaced any letter that appeared consecutively more than two times in a word with one letter. • Removing punctuation: removed punctuation, such as full stop, comma, and brackets. • Removing stop words: removed the most common words in the Arabic language (articles, prepositions, pronouns, conjunctions, etc.) that did not add much information to the text. • Removing emojis: removed all emojis from tweets. • Removing English words: removed all English words. • Text normalization: Transformed a text to a unified form; removed Al-tashkil and elongation.

Reviewing the dataset
Following the cleaning step, each record in the dataset was manually reviewed. Any records that did not relate to depression (e.g., records not describing personal moods, anxiety, movement, sleep mode, suicidal thoughts, or records not associated with the PHQ-9 questions, as shown in Table 1 ) were disregarded. The number of examples for each category/class label was also balanced to achieve the right accuracy when using machine-learning classification algorithms. The final dataset contains 1229 records of 928 Saudi Twitter users as showing in Fig. 1 .

Ethics Statement
Our data does not provide any personally identifiable information and only the tweet IDs and human annotated stance labels are shared. Thus, all data are fully anonymized and were collected and distributed under Twitter's Developer Policy 2021 [7] .
The PHQ-9 is available to healthcare providers completely free of charge and the legal copyright holder, explicitly states that "no permission [is] required to reproduce, translate, display or distribute [the PHQ-9]" [8] .

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships which have or could be perceived to have influenced the work reported in this article.

Data Availability
Modern Standard Arabic mood changing and depression dataset (Original data) (Mendeley Data).