Recommender Systems

What is Recommendation System :

The recommendation systems are defined as a software tools to give suggestions for items to the users in which they might be interested. The suggestions might be related to decision-making processes which can be within these – what movie / TV series to watch, what playlist to listen, what items to buy, what news to read, or what videos to watch and many more in this list. The main objective of the recommendation systems is to provide recommendations to users on their online activities for better choices from many options.

.

Each Recommender System Is Composed Of :

Users:

Each user may have set of user attributes. For example: age, gender and etc.

Items:

Each item may have a set of item attributes or properties. For example, an actor in a movie, author of a news article or color of an appliance.

Preferences:

These represent users’ likes and dislikes. User meets item in the preferences space, for example, a user can rate a movie 5 on a scale of 5 stars.

.

Type Of Feedback:

Explicit:

user preferences can be identified by their own. For instance, in Netflix users can rate the movies. Hence, recommender system can offer the new movie based on their rating.

Implicit:

user preferences can be identified by their activity and these activities need to be analyse. For instance, recommender system offer new product based on their past activity on the website. 

.

Popular Methods Of Recommender Systems:

  1. Non-Personalised System
  2. Content-Based Filtering
  3. Collaborative Filtering
  4. Hybrid System

.

1. Non-personalised system

These involve summary statistics and in some cases product associations by using external data from the community like, a product that is the best seller or most popular or trending hot. It may also provide summary of community ratings, for example, how much a population likes a restaurant or summary of community ratings which turns into a list like, which is the best hotel in town.

Also Read:  Reinforcement Learning in Marketing Campaign

.

2. Content-Based Filtering

Here, users rate items and from that a model of user preferences against the item attributes is built. An example could be in the domain of movies. Suppose someone likes science fiction, fantasy and action movies, and doesn’t like romantic movies. Overtime the algorithm can accumulate this and Figure out that the user has positive scores on genres like science fiction, fantasy and action, and lower scores for romance. The algorithm might also find out that there were some actors that user likes or dislikes. For example, the user can be a fan of movies with the actor Bruce Willis, and not a fan of Ben Stiller.

.

3. Collaborative Filtering

collaborative filtering is a method based on information about the actions of other users whose preferences are similar to the existing user. This method studies the pattern of other user’s behavior rather than extracting features from the product. The key advantage of this approach is that prior analysis and understanding of the existing data is not required to make recommendations to the user. However, these systems need a large amount of data from various users in order to make recommendations.

.

4. Hybrid System

In these methods, a combination of two or more recommendation algorithms are used to take or maximize advantage of some techniques and avoid or minimize the drawbacks of another.

.

1241cookie-checkRecommender Systems