Jennifer Rodriguez Trujillo
Mental Health Advocacy has exponentially grown in the past decade. In efforts to increase outreach, machine learning models such as Bag of Words have been used to determine text sentiment. From user's comments to the post's users view, the text is analyzed in attempt to target users who may display signs of distress. Bag of Words is a machine learning algorithm often used in Natural Learning Process. By using the algorithm Bag of Words and training the model using Logistic Regression, I will be using "TfidfVectorizer" and "CountVectorizer" in order to determine which method produces optimal accuracy. In this research project, I will be using a web scrapped data from subreddits. Furthermore, this dataset personal annecdotes and commentary in relation to mental health struggles.
The ability to detect mental destress in social media posts is crucial in that any form of outreach can make a difference. Currently, we see many forms of outreach being done in short genres,such as Twitter, but what is currently lacking is the ability for models to detect stress in larger domains. Countless of mental health statistics exist online and many of it revolves around the idea of out reach. Mental health has not always received the attention it requires, but even then, it still remain taboo in certain communities and geographic locations to talk about mental health. For these vulneable communities, or potentially even households, it is pertinent to reach out to them through other means. Although technology is not readily available to all, it encompasses a broad population of teenagers: the current predominant group. Many teenagers reserve to social media as a form of relief. With that said, NLP can be a powerful tool to use in order to target vulnerable users. Although the data being used only encompasses a subsample, it permits us to receive of glimpse at what an optimal NLP model that detects stress may present itself to be.
In this project, given subreddit texts in realtion to mental health struggles, I will be examining and analyzing text sentiment. I will be building a model using the Bag of Words algorithm and training it through Logistic Regression. I will then be using "TfidfVectorizer" and "CountVectorizer" in order to determine which method produces optimal accuracy.
The data consists of subreddits from five different categories(n = 190k )collected via web scrapping. The following is a list of column names the data consists of:
subreddit ptsdassistanceptsdrelationshipssurvivorsofabus... <br>
post_id 8601tu8lbrx99ch1zh7rorpp9p2gbc7tx7et7iphly5m3k... <br>
sentence_range (15, 20)(0, 5)(15, 20)[5, 10][0, 5](30, 35)[25... <br>
text said felt way sugget go rest trigger ahead you... <br>
id 39028174 <br>
label 1488 <br>
confidence 2295.86 <br>
social_timestamp 4308387751949 <br>
social_karma 51828 <br>
syntax_ari 13294 <br>
lex_liwc_WC 244057 <br>
lex_liwc_Analytic 100014 <br>
lex_liwc_Clout 116211 <br>
lex_liwc_Authentic 190272 <br>
lex_liwc_Tone 94869.1 <br>
lex_liwc_WPS 51621.7 <br>
lex_liwc_Sixltr 42167.4 <br>
lex_liwc_Dic 262087 <br>
lex_liwc_function 166304 <br>
lex_liwc_pronoun 56100.7 <br>
lex_liwc_ppron 39696.6 <br>
lex_liwc_i 25730 <br>
lex_liwc_we 2170.16 <br>
lex_liwc_you 2457.06 <br>
lex_liwc_shehe 7685.52 <br>
lex_liwc_they 1653.61 <br>
lex_liwc_ipron 16372.5 <br>
lex_liwc_article 14013 <br>
lex_liwc_prep 37906.2 <br>
lex_liwc_auxverb 29249.3 <br>
lex_liwc_adverb 17158.1 <br>
lex_liwc_conj 21566.1 <br>
lex_liwc_negate 6422.12 <br>
lex_liwc_verb 55146 <br>
lex_liwc_adj 12381.4 <br>
lex_liwc_compare 6491.32 <br>
lex_liwc_interrog 4564.12 <br>
lex_liwc_number 4098.5 <br>
lex_liwc_quant 6375.3 <br>
lex_liwc_affect 17304.4 <br>
lex_liwc_posemo 7656.53 <br>
lex_liwc_negemo 9378.83 <br>
lex_liwc_anx 2594.49 <br>
lex_liwc_anger 2635.18 <br>
lex_liwc_sad 1698.41 <br>
lex_liwc_social 30728.5 <br>
lex_liwc_family 2114.91 <br>
lex_liwc_friend 1575.75 <br>
lex_liwc_female 4474.43 <br>
lex_liwc_male 5610.37 <br>
lex_liwc_cogproc 38582.5 <br>
lex_liwc_insight 8102.8 <br>
lex_liwc_cause 4899.64 <br>
lex_liwc_discrep 5526.11 <br>
lex_liwc_tentat 9435.66 <br>
lex_liwc_certain 4382.64 <br>
lex_liwc_differ 11272.3 <br>
lex_liwc_percept 6376.87 <br>
lex_liwc_see 1626.5 <br>
lex_liwc_hear 1741.48 <br>
lex_liwc_feel 2588.72 <br>
lex_liwc_bio 7409.1 <br>
lex_liwc_body 2010.96 <br>
lex_liwc_health 3538.42 <br>
lex_liwc_sexual 607.64 <br>
lex_liwc_ingest 1150.66 <br>
lex_liwc_drives 22378.2 <br>
lex_liwc_affiliation 7768.64 <br>
lex_liwc_achieve 3896.04 <br>
lex_liwc_power 6851.88 <br>
lex_liwc_reward 3884.81 <br>
lex_liwc_risk 2043.75 <br>
lex_liwc_focuspast 14356.2 <br>
lex_liwc_focuspresent 34975.6 <br>
lex_liwc_focusfuture 3358.13 <br>
lex_liwc_relativ 40453.5 <br>
lex_liwc_motion 5688.48 <br>
lex_liwc_space 17865.6 <br>
lex_liwc_time 17593.7 <br>
lex_liwc_work 5323.22 <br>
lex_liwc_leisure 2228.21 <br>
lex_liwc_home 1802.45 <br>
lex_liwc_money 2231.31 <br>
lex_liwc_relig 328.25 <br>
lex_liwc_death 402.46 <br>
lex_liwc_informal 2343.21 <br>
lex_liwc_swear 699.52 <br>
lex_liwc_netspeak 724.26 <br>
lex_liwc_assent 338.04 <br>
lex_liwc_nonflu 354.09 <br>
lex_liwc_filler 141.49 <br>
lex_liwc_AllPunc 48398.3 <br>
lex_liwc_Period 17216.1 <br>
lex_liwc_Comma 10141.1 <br>
lex_liwc_Colon 620.92 <br>
lex_liwc_SemiC 350.29 <br>
lex_liwc_QMark 1473.11 <br>
lex_liwc_Exclam 546.29 <br>
lex_liwc_Dash 1381.54 <br>
lex_liwc_Quote 1343.83 <br>
lex_liwc_Apostro 9014.79 <br>
lex_liwc_Parenth 2210.69 <br>
lex_liwc_OtherP 4098.82 <br>
lex_dal_max_pleasantness 7937.78 <br>
lex_dal_max_activation 7676.62 <br>
lex_dal_max_imagery 8367.6 <br>
lex_dal_min_pleasantness 3087.75 <br>
lex_dal_min_activation 3178.84 <br>
lex_dal_min_imagery 2838.6 <br>
lex_dal_avg_activation 4889.19 <br>
lex_dal_avg_imagery 4360.3 <br>
lex_dal_avg_pleasantness 5333.7 <br>
social_upvote_ratio 2393.9 <br>
social_num_comments 28234 <br>
syntax_fk_grade 15463.8 <br>
sentiment 115.621 <br>
Turcan, E., & McKeown, K. (2019). Dreaddit: A Reddit dataset for stress analysis in social media. arXiv preprint arXiv:1911.00133.
Rolnick, David, et al. "Tackling climate change with machine learning." ACM Computing Surveys (CSUR) 55.2 (2022): 1-96.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent