東方プロジェクト人気投票の解析
Bayesian analysis of Touhou project vote data
China(optional)
- http://thwiki.info/?%BD%D0%B1%E9%A5%EA%A5%B9%A5%C8 title & charactors
- http://thwiki.info/?%A5%AD%A5%E3%A5%E9%CA%CC%A5%B9%A5%DA%A5%EB%A5%AB%A1%BC%A5%C9%2F1#v5aac3bf spell cards &charactors
- https://touhou.arrangement-chronicle.com/statistics music
- https://w.atwiki.jp/toho/pages/842.html#id_713b4b9a title & music & charactors
- pixiv ranking illustration
- https://docs.google.com/spreadsheets/d/1YKpwyCOLRWiC9jucljunxS41i75zUYojI3qx1JnIpVI/edit?usp=sharing
- https://booth.pm/ja/items/2025481
- https://hisayukihonbun.booth.pm/tems/760429
- https://www.pixiv.net/artworks/68538151
- https://docs.google.com/spreadsheets/d/1NbdTXrPqA3df5wS37emN3RMFDH9Re2nApleYTlGKsp8/edit#gid=753814927
run
python analysis_char.py 18 18 sum
Bayesian analysis by using pystan/gensim
python3 analysis.py 18 5 modelsuffix
18 is election number, 5 is duration of time effect
for normalized vote numbers
-
$main_{j(i,t),t}$ i is main character of subtitle just before tth election -
$boss_{j(i,t),t}$ i is boss character of subtitle just before tth election -
$Lv_i$ boss level of character i -
$sub_{j(i,t),t}$ i is character of noninteger subtitle just before tth election
-
$\sigma$ coef of integer title main characters - b coef of integer title bosses
- s coef of noninteger title members
-
$\epsilon_i$ indivisual term
- t index of time(election)
- l index of realtive time
- i index of characters
- j index of mainchar,boss,subtitle table
- V num. of characters
- K num. of titles
- N num. of total votes
- M num. of elections
-
$\alpha, \beta$ hyperparameter
caluculate by using gemsim