Detection of suicide-related posts in Twitter data streams
Citations
271 citations
139 citations
Cites background or methods from "Detection of suicide-related posts ..."
...In addition, the classification results here are just an initial exploration of the problem; for example, we plan to follow Vioulès et al. (2018) in exploring hierarchical rather than four-way classification, which yielded substantial improvements, and we are exploring the role of hierarchical…...
[...]
...In work similar to the work we report here, Vioulès et al. (2018) applied a similar data collection approach to Coppersmith et al., searching Twitter for tweets containing key phrases based on risk factors and warning signs identified by the American Psychiatric Association and the American…...
[...]
123 citations
104 citations
Cites background or methods from "Detection of suicide-related posts ..."
...O’Dea et al. [11] collected tweets using the public API and developed automatic suicide detection by 7 applying logistic regression and SVM on TF-IDF features....
[...]
...collected 5,446 tweets using Twitter streaming API [4], of which 2,381 and 3,065 tweets from the distressed users and normal users, respectively....
[...]
...[4] conducted usercentric and post-centric behavior analysis and applied a martingale framework to detect sudden emotional changes in Twitter data stream for monitoring suicide warning signs....
[...]
...Vioulès et al. collected 5,446 tweets using Twitter streaming API [4], of which 2,381 and 3,065 tweets from the distressed users and normal users, respectively....
[...]
...suicide factors fall under under three categories: health factors, environment factors, and historical factors [4]....
[...]
73 citations
References
15,478 citations
6,539 citations
3,830 citations
"Detection of suicide-related posts ..." refers background in this paper
...However, unlike retrospective detection settings [23, 24], which focus on batch processing, here we are interested in the setting where the data arrives as a stream in real-time....
[...]
2,975 citations
"Detection of suicide-related posts ..." refers background in this paper
...However, the social stigma surrounding mental illnesses means that at-risk individuals may avoid professional assistance [4]....
[...]
2,570 citations
"Detection of suicide-related posts ..." refers background in this paper
...The character limitation (140 maximum) of tweets lends itself to a choice of shorter n-grams, particularly uni-grams and bigrams [36, 37]....
[...]
Related Papers (5)
Frequently Asked Questions (13)
Q1. What contributions have the authors mentioned in the paper "Detection of suicide-related posts in twitter data streams" ?
In this paper, the authors present a new approach that uses the social media platform Twitter to quantify suicide-warning signs for individuals and to detect posts containing suicide-related content. Experiments show that their text-scoring approach effectively captures warning signs in text compared to traditional machine learning classifiers.
Q2. What are the future works mentioned in the paper "Detection of suicide-related posts in twitter data streams" ?
For future research, the authors plan to further explore the impact of martingale parameters on the change detection effectiveness. However, overall, the authors believe their initial work presents an innovative approach to detecting suicide-related content in a text stream setting.
Q3. Why did the authors use the Twitter streaming API?
Due to the absence of publicly available datasets for the evaluation of suicide detection in social media, the authors used the Twitter streaming API to collect tweets.
Q4. What are the two groups of behavioral features that the authors used to identify the mental state of a?
To identify online behaviors that may reflect the mental state of a Twitter user, the authors established two groups of behavioral features: user-centric and post-centric features [11, 28].
Q5. What is the first approach to classifying a post?
The first approach is a natural language processing (NLP) method that combines features generated from the text based on an ensemble of lexicons.
Q6. What is the cost of annotating the text?
Although machine learning is commonly used to classify text, the supervised algorithms require annotated datasets, which may be costly in terms of time and potential annotator error.
Q7. What was the main challenge when implementing the martingale framework?
To detect changes in emotional well-being, the authors considered a Twitter user’s activity as a stream of observations and applied a martingale framework to detect change points within that stream.
Q8. How many tweets were used to evaluate distress?
To evaluate lexicon-based NLP approach, the authors used the cross-sectional set of 500 tweets and looked at the average, maximum, and minimum score given by each distress class.
Q9. What is the simplest way to determine whether there is an abrupt change in the user behavior?
Q10. What is the effect of the -values on the incoming data points?
Q11. What are the features that can quantify an individual's interaction with their online community?
The friends and followers features can quantify an individual's interaction with their online community, such as a sudden decrease in communication.
Q12. What is the correlation between the two spikes?
Upon further investigation, the authors found that these two spikes are linked to negative SPA scores (positive emotion) corresponding to birthday wishes that Frank received from other users.
Q13. What does the martingale framework need to be adjusted to interpret?
These peaks highlight that their martingale framework will need to be adjusted to interpret negative SPA text scores as a positive behavioral change that do not require an alarm to sound.