- input data → web pages of questions of users
- output → users to follow, ranked
- two process
- personality based recommendation
- topic interest based recommendation
- intermediate data stored as json
- in topic hierarchy, for each node, find base topics at some level
- use networkx library to travel topic hierarchy
- find sort path(s) to root
- and find nodes at level in the ans
- other nodes which have the same nodes in this level
- and are closest means they posses similar interest in topics
- networkx is python library for complex networks
- personality recognizer module
- 5 vector personality test
- 0 to 7 score
- Linguistic Inquiry and Word Count → LIWC
- personality recognizer
- based on words used by user, find his intent or personality
- to do combination → 1st get results based on personality scores
- rank them based on topic hierarchy
- topic hierarchy
- topic tags