collaborative filtering - Is this an approach to user-item recommendations that could work -

i designing application incorporates recommendation system base on user interactions (collaborative filtering). user on homepage presented set of 6 items interact with. there between 50 , 300 items. following actions possible:

1. click on item (strong interest)
2. refresh item (some interest)
3. open read-more dialog (some interest)
4. don't move on (no interest)

this data collected , stored. system should recommend items of interest user. i'am thinking turning data rating system.

option a) if user clicks on item, translated implicit lifetime rating of 5. refreshing item 4 , on. user->item matrix this:

       item 1 | item 2 | item 3 john   5                 4 jane   4 

in example john has clicked on item 1 , refreshed item 3. rating can go really, i.e. if user has refreshed item write 4 , update 5 if item clicked later.

option b) each time user 1 of above actions, i'll increment scalar value item, means can grow unbounded.

       item 1 | item 2 | item 3 john   55       1        30 jane   41       9 

maybe problem, since numbers harder translate rating scale 1 10

option c) count every interaction separately

       item 1 click | item 1 refresh | item 1 read john   3              1                        jane   1                               1 

here problem "reading about" item done once.

independent of whatever option choose, idea first find similar users using cosine similarity or pearson correlation. pick top 10 30 users list , compile toplist of favorite items. list, recommend items current user has had little interaction in past.

is work? worried finding similar users eliminate chance of finding interesting (new) items current user.

what suggest sounds reasonable. concern not finding new items reflection of collaborative filtering method metadata-based. find new items have undoubtedly content analysis separate stage. example, if items news articles might try identify important keywords each user.