|
I am building question answering system witch recommend questions to user. I have match algorithms which match user tags with question tags, it is function from 0 to +Inf. The user can response on incoming question by pass, answer or mark spam. Then answer can be helpful or not. So, I want to create user rank weight based on those data, and have association formula like this match(q,u)=tag_match(q,a)*rank(u), where rank(u) = f(pass_count, answer_count, asked_count, mark_spam_count, helpful_answer, not_helpful_answer). Any idea how should this formula look like? And how to train it? |
|
I suggest you read Anatomy of a Large-Scale Social Search Engine, which describes aardvark (And the related blog post). You may get some ideas there. |
