Machine Learning & Data Science Recommender Systems Content Based Recommender Systems

Content-based recommender systems provide personalised recommendations by leveraging features of the items themselves, as well as the interactions or preferences of the users with those items.

Each user has a "profile", i.e., a set of parameters $\theta^j$ and each item has a feature vector $x^i$. Some users have rated some items. $r(i,j) = 1$ if user $j$ has rated item $i$. A prediction can then be made on item $i$ by user $j$ : $$p = (\theta^{j})^{T}(x^{i})$$The intuition here is that $\theta^{j}$ has been learnt so that the features of the item which the user prefers are prioritised.

Optimisation Objective

To learn the parameters for each user, we want to minimise the cost function $$J(\theta^{j}) = \frac{1}{2}\sum\limits_{i:r(i,j)=1}((\theta^{j})^{T}(x^{i}) - y^{i,j})^2$$ I.e., go through all the items that the user has rated and sum the difference between our prediction and the actual rating. We would need to do this for each user. Machine-Learning-&-Data-Science-Supervised-Learning-Gradient-Descent with regularisation can be used to fit the parameters for a particular user.

Advantages

• No Cold Start: Content-based systems do not require other users to have interacted with an item to recommend it.
• Personalised: Recommendations are tailored to individual users based on their own behaviour.

Disadvantages

• Limited Scope: Can only recommend items similar to those the user has already interacted with, potentially resulting in a lack of diversity.
• Feature Engineering: Requires meaningful feature representation of items, which can be challenging.