A corroborating argument for dropping the stars: A lot of people who've tried multi-value rating systems have found that they get a heavily bimodal distribution anyways: people vote when they think something is awesome, or when it sucks, but rarely when they think it's just meh.
So you really won't be losing much information with a two value (eg, upvote/downvote) system.
Should I keep the 5 star rating control for rating articles, and change the voting on items to thumbs up/down? Or should I change both the voting on items and voting on articles to thumbs up/down?
Both articles and items should have the up/down vote options, rather than the 5 star, IMO.
There may be value in the determination of consumer-side article rank by a (weighted) average that encompasses both the net rank of items posted to the article + the rank of the article itself.
I'm sure you've already thought about the item/article interaction in terms of effect on overall article ranking. UI/X-side, the effect (of ranking an item within an article) is unclear now, and I am fine with that as an early user.
Moving forward, developing a ranking formula that weighs article rank against net rank of all of the article's items will be a usage-informed improvement that provides great value to users.
That ranking formula, as well as making the result of a "ranking" interaction obvious to the user, will help the understanding of how such interactions work.
With a social answering/ranking/voting application, creation of, and/or tweaks to, the formula that determines the effect of interactions will have a huge influence in shaping the culture of the individuals that interact with the application.
Judging from the concept of "decision fatigue" outlined in that NY Times article the other day * , you're better off with thumbs up/down since it's less work to decide that than to have to choose 2 or 3 or 4 stars for an OK article.
Think about it from a game theory perspective. If the rating of something is higher than you would like it to be, why would you vote more than a 1? If the rating is lower, why would you vote less than a 5?
Some people may express their honest preference, but a major motivation for people to interact with something like this is to change the average, and voting at the extremes maximizes their power.
So you really won't be losing much information with a two value (eg, upvote/downvote) system.