I agree with Johannes, but I also need to add to the equation the need for a product/content catalogue, as without it in place you can't get the details about the pages viewed. The catalogue needs to contain your unique identifier of each page, and metadata with e.g. page category. Once the catalogue is in place, you need to add the unique ID to each trackPageView, like this where you replace INSERT_ITEM with the page ID:
_etmc.push(["trackPageView", { "item" : "INSERT_ITEM" }]);
Before a catalogue is in place, you will also not be able to enable the integration to the data extensions.
Once you have set up the collect code, you must join the IGO_PRODUCTATTRIBS + IGO_VIEWS + IGO_PROFILES or PI_CONTENTATTRIBS + PI_CONTENTVIEWS + IGO_PROFILES (depending on whether you use product or content catalogue) to derive the interests of the identified users. For this you will be using Automation Studio and SQL Query activities.
PI_CONTENTATTRIBS/IGO_PRODUCTATTRIBS will contain metadata passed to the catalogue as name-value pairs and are crucial to this, as you will need to take the categories from here.
Keep in mind, that tracking categories does not feed into any data extensions, and can't be used for segmentation:
_etmc.push(["trackPageView", { "category" : "INSERT_CATEGORY" }]);