One of Hive’s most amazing features is the recommendations that users will get. Hive recommends users other users who are likely to be similar based on their interests. Also, a user will get interest recommendations based on their past behavior or their friend’s behavior. In this post we briefly explain what recommendation systems are, discuss the different approaches and then we will discuss how Hive’s recommendation system could work.
What recommendation Systems are?
Recommendation systems or recommender systems are a type of information filtering that is trying to present products (movies, books, music) or social elements (users, groups) that are likely of interest to a user. In other words, recommendation systems are trying to filter the information that is going to be s
hown to the user based on his recent behavior like purchasing a book, watching a movie or listen to music. A common recommendation from amazon is “Customers who bought items in your basket also bought…”.
To achieve this kind of information filtering, an effective algorithm is required. Today, many web sites are using recommendation systems such as amazon, IMDB, Google, Last.fm etc. For instance, a user registered to a movies web site declares that he likes war movies, either by searching war movies regularly or indicating that he has watched many of them. Then the recommendation system is able to understand this trend and will recommend him other war movies that he hadn’t watched yet. This can bring many benefits for a company and can boost its sales and that’s because many users may love things which they didn’t know existed. That’s why recommendation systems are trying to find similar things that a user likes and brings them closer by recommending them.
Recommendation systems can be used in Hive to recommend to users other similar users that share many interests who they didn’t know before; Hive is giving them the opportunity to meet. This can be used for interests too, so a user may find many new things that they would be interested in, using the recommendations of Hive.
Different Approaches
There are some different approaches to recommendation systems and each one has its advantages and disadvantages.
Collaborative Filtering
Collaborative filtering or user-based is an approach to 
recommendation systems and it’s widely used by many web sites. Its aim is to find similar users and then recommend to one what the other similar user likes. How similar a user is with another user is defined by their distance. Distance can be measured by using n-dimensional Pythagoras or some other techniques.
Advantages:
- Very good recommendation results
Disadvantages:
- It’s not efficient to map every user to similar users within sites with many users
- A new user cannot get good recommendations
- Algorithms have more complexity
Item based
This approach finds similarities between items rather than users. This technique recommends items that are similar to those that a user liked in the past. specifically, various candidate items are compared with items previously rated by the user and the best-matching items are recommended
. So for example if a user tends to buy sport shoes, then the system will recommend him more sport shoes that the system has classified as similar to the ones he already bought.
Advantages:
- Good recommendations for a new user
- Better efficiency if there are more items than users.
Disadvantages:
- Limited in scope
These 2 techniques are the common techniques for recommendation systems. An example of each one is given below.
User-Based: A site that uses a user-based approach is last.fm. Last.fm creates a list of recommended songs for a user that users with similar tastes are listening too.
Item-based: A site that uses an item-based approach is Pandora. Pandora tries to classify songs into similar songs and recommends to users some similar songs to what they are currently listening to.
Hybrid Approach
A combination of the above techniques can result in a hybrid approach which can sometimes bring better recommendation results. Many web sites now use this technique because they can avoid the disadvantages of one technique while using the advantages of the other. Netflix, Amazon and Google have some of the most popular hybrid recommendation systems.
Hive – A Paradise for Recommendation Systems
As mentioned above, recommendation systems are trying to find similarities between users and items so they can recommend new items to the users based on these similarities. Because of Hive’s approach to connect similar people, the recommendation system can connect people who are similar by analysing the common interests of 2 users. Also, if two users are friends then they probably share many interests. For example if 2 users are friends and they both have pop music as the common interest, and user A likes a different artist or a song then user B will probably like it too, so Hive will recommend that item to user B as well.
Also most of the items in Hive will belong to a category. For example the movie Transformers 3 will belong to the category action, science-fiction, etc. So, since most items are categorized and we know which of them are similar, again the recommendation system can easily recommend the right items to a user.
So because of Hive’s nature to keep similar people and similar items together, a combination of user-based and item-based recommendation algorithms can be used to have a hybrid recommendation system that will use both user similarities and item similarities to recommend new friends and new items to the users.