使用户兴趣与内容匹配（基于标签）

Question

I have a lot of content items stored in the database and I know which tags a user is interested to. Alice, for example, shows interest in tags like "healthcare", "sports" and "social". Each content items contains one or more tags. How would I match these in order to recommend new content to Alice?

Consider these basic database tables:

CREATE TABLE `content_tag` (
   `id` INT(11) NOT NULL AUTO_INCREMENT,
   `item_id` INT(11) NOT NULL,
   `tag_id` INT(11) NOT NULL,
   PRIMARY KEY (`id`)
);

CREATE TABLE `tag` (
   `id` INT(11) NOT NULL AUTO_INCREMENT,
   `name` VARCHAR(50) NOT NULL,
   PRIMARY KEY (`id`)
);

And I have Alice's interests alongside with a (relevance) score, which act like weights:

array:3 [
    'healthcare' => 2.20
    'sports' => 1.30
    'socal' => 0.5
]

How would you approach this?

Is there a way to use an algorithm for this, like the cosine simularity, or is this only ment for sentences?

Answer 1

You can find similarities between Alice and items and then sort them. So, most similar n items will be recommended.

One of the similarity metrics is cosine similarity (as you suggest) and works like follows;

For each item you can create a vector by using tags. As far as I understand your items don't have scores so values of item vectors will be 0 or 1. Each value will represent one tag for the particular item.

Item representation;

[0,0,1,1,0,0] -> Lets say First value represents 'healthcare', second one is for sports and last tag is for tag5. This item does not have tag5 so its value is 0

And users also have vectors which is similar to items. For instance Alice's vector is; [2.20,1.30,0.5,0,0,0]

After creating vectors you can compute similarity (eg by using cosine similarity.)

Note that size of each user and item vectors equals to number of all tags in the system. In this example there are 6 distinct tags in the system.