Depression Language
Finally we define two specialized features focused on characterizing the topical language of individuals detected positively with depression. While our previous measure focused on the linguistic style of depressive language, we are also interested in analyzing what people talk about.
(a) Depression lexicon. The first feature measures the usage of depression-related terms, defined broadly, in Twitter posts. For this purpose, we built a lexicon of terms that are likely to appear in postings from individuals discussing depression or its symptoms in online settings. We mined a 10% sample of a snapshot of the “Mental Health” category of Yahoo! Answers. In addition to already being categorized as relevant to depression, these posts are separated in-to questions and answers and are relatively short, making them well-aligned to the construction of a depression lexi-con that can eventually be deployed on Twitter.
We extracted all questions and the best answer for each of question, resulting in 900,000 question/answer pairs. Af-ter tokenizing the question/answer texts, we calculated for each word in the corpus its association with the regex “de-press*” using pointwise mutual information (PMI) and log likelihood ratio (LLR). We created the union of top 1% of terms in terms of LLR and PMI. To remove extremely fre-quent terms, we calculated the tf.idf for these terms in Wikipedia and used the top 1000 words with high tf.idf. Thereafter we deployed this lexicon to determine frequen-cy of use of depression terms that appear in the Twitter posts of each user, on a given day.
(b) Antidepressant usage. The next feature measures the degree of use of names of antidepressants popular in the treatment of clinical depression (any possible overlap with the above lexicon was eliminated). Individuals with de-pression condition are likely to use these names in their posts, possibly to receive feedback on their effects during the course of treatment (Ramirez-Esparza et al., 2008). We used the Wikipedia page on “list of antidepressants” in or-der to construct a lexicon of drug names
Depression LanguageFinally we define two specialized features focused on characterizing the topical language of individuals detected positively with depression. While our previous measure focused on the linguistic style of depressive language, we are also interested in analyzing what people talk about.(a) Depression lexicon. The first feature measures the usage of depression-related terms, defined broadly, in Twitter posts. For this purpose, we built a lexicon of terms that are likely to appear in postings from individuals discussing depression or its symptoms in online settings. We mined a 10% sample of a snapshot of the “Mental Health” category of Yahoo! Answers. In addition to already being categorized as relevant to depression, these posts are separated in-to questions and answers and are relatively short, making them well-aligned to the construction of a depression lexi-con that can eventually be deployed on Twitter.We extracted all questions and the best answer for each of question, resulting in 900,000 question/answer pairs. Af-ter tokenizing the question/answer texts, we calculated for each word in the corpus its association with the regex “de-press*” using pointwise mutual information (PMI) and log likelihood ratio (LLR). We created the union of top 1% of terms in terms of LLR and PMI. To remove extremely fre-quent terms, we calculated the tf.idf for these terms in Wikipedia and used the top 1000 words with high tf.idf. Thereafter we deployed this lexicon to determine frequen-cy of use of depression terms that appear in the Twitter posts of each user, on a given day.(b) Antidepressant usage. The next feature measures the degree of use of names of antidepressants popular in the treatment of clinical depression (any possible overlap with the above lexicon was eliminated). Individuals with de-pression condition are likely to use these names in their posts, possibly to receive feedback on their effects during the course of treatment (Ramirez-Esparza et al., 2008). We used the Wikipedia page on “list of antidepressants” in or-der to construct a lexicon of drug names
การแปล กรุณารอสักครู่..
