a valid ranking algorithm for facebook posts

Question

I'm creating a website which will collect posts from facebook's pages to show to the users posts they are interested in (as my graduating project).

I'd like to be able to rank the posts that contain the words which are queried. Hence, I use (solr)for full text indexing which allows me to effciently get the posts containing the words the user searches for.

Now, I am looking for a ranking function for returning most relevant search results first. But there are many variables which could influence the relevance and ranking of the post. I found these factors:

1. date of the post.

2. likes number of the post.

3. comments number of the post(not very correct).

4. is the published page verified ?

5. likes count for the published page.

6. location of the page and the user.

I need a formula to combine these factors, and I think that there is something missing - there must be a weighting for the search keywords themselves in the post. And maybe I am missing something more?

Is there any other way to determine relevance?

Some additional information:

• I think that there is an important factor I am not using. This factor should deal with the other keywords in the post and see if they are related to the major keyword.

Take, for example, these two posts:

1. I love sports.

2. I love basketball , football and all other sports and i think that they are very healthy.

Lets say these posts have the same above factors, but the second has more keywords in the required domain (sports). How to determine that the second post is more related to the "sports" domain?

Answer 1

For creating a formula to map those parameters to a ranking value, first map each parameter to a positive real number. Then a simple approach is to make a weighted sum from these numbers. The Wikipedia article also contains further links to more sophisticated approaches.

Of course, you need to determine "good weights", but hey, it is your graduate project, and you should do some research here on your own.

Moreover, you should definitely inform yourself how the page ranking algorithms of those bigger search engines work and if there are parts of it applicable to your problem.

I cannot tell you why Facebook search changes results after each refresh, and I don't think this is a good question for this site, since we also could only guess here (I would actually recommend to delete that part from the question, some "site cop" guys here are often looking for a reason to close questions because they contain a minor off-topic section).

The missing factor you are looking for could be the number of additional words in that post which do not match any of the search keywords exactly, but are "semantically related" to one or more of them. So how do you measure "semantically related"? Actually, I am not an expert on this, but AFAIK this is whole scientific research area, and tools like CoreNLP or SML might be worth a try.