User-based Filtering:
Take about 20-50 people who share similar taste with you, afterwards predict how much you might like an item depended on how much the others liked it.
Item-based Filtering:
Pick from your previous list 20-50 items that have similar people with “the target item”, how much you will like the target item depends on how much the others liked those earlier items.
Information needs of user and characteristics of items are represented in keywords, attributes, tags that describe past selections.An item is recommended to the user based on the scores calculated according to these preferences and characteristics.
Then comes the first question: What would happen if we mix them into one filter?
Undoubtedly the filtering may recommend a large range of products based on user's characteristic. In this case, the large range may not be our initial aim because user may lose their target again and ignore the filter.
Therefore we should choose one from the three types of Derived Social Recommendations according to what product we aim to recommend.
Second, I have a idea about Derived Social Recommendations that is background-based filtering.
Background-based Filtering
Background such as education level, hobby, age define a group of customers. The products they have bought create a sort of characteristic according to how much they may like the products.
This may be similar with the user-based filtering. But the key advantage of this filtering is that the defined characteristic keep changing according to the change of the whole background. I believe background-based filtering can match customer's demand best.
