Recommender System

 

Test day, Recommender systems and input-output

Recommender Systems

A recommender system, or a recommendation system, is a subclass of information filtering systems that seeks to predict the “rating” or “preference” a user would give to an item. They are primarily used in commercial applications

Examples of Recommender Systems:

Netflix, YouTube, Tinder, and Amazon. 

The systems entice users with relevant suggestions based on the choices they make. Recommender systems can also enhance experiences for News Websites.

Types of Recommender Systems

Recommender systems are typically classified into the following categories:

• Content-based filtering
• Collaborative filtering
• Hybrid systems

Depending on whether a model is learned from the underlying data, the recommender system can be divided into:

• Memory-based
• Model-base

The test was on collaborative filtering and content-based filtering

collaborative filtering

collaborative filtering uses similarities between users and items simultaneously to provide recommendations. A collaborative filtering model can recommend items to User A based on similar User B interests. Additionally, embeddings can be automatically learned without resorting to the manual development of features.
An example of a recommended movie
Consider a movie recommendation system whose training data consists of a feedback matrix.

A Movie Recommendation Example

Consider a movie recommendation system in which the training data consists of a feedback matrix which:

• Each row represents a user.
• Each column represents an item (a movie).

The feedback about movies falls into one of two categories:

Explicit— users specify how much they liked a particular movie by providing a numerical rating. Implicit - When a user watches a movie, the system infers that the user is interested.
For simplicity, we assume the feedback matrix is ​​binary. So a value of 1 indicates an interest in movies.

When a user visits the home page, the system should recommend movies based on both:


Similar to movies the user has liked in the past
Movies liked by similar users
advantage

No domain knowledge required

Embeddings are learned automatically, so no domain knowledge is required.
lucky coincidence

This model helps users discover new interests. Alone, the ML system may not be aware that a user is interested in a particular item, but similar users are interested in that item, so the model recommends it. There are cases. great starting point

To some extent, the system only needs the feedback matrix to train the matrix factorization model. In particular, the system does not require context properties. In practice, this can be used as one of the multiple candidate generators.

Cons
We do not carry fresh produce

The model's prediction for a given pair (user, item) is the dot product of the corresponding embeddings. So if an element is not visible during training, the system cannot create embeddings for it and query the model with it. Hard to include query/item site features

A minor feature is a feature that goes beyond a query or item ID. For movie recommendations, side features can include country and age. Including available auxiliary features improves the quality of the model. Including side functions in WALS may not be trivial, but generalizing WALS makes it possible.


Content-based filtering

Content-based filtering uses the Articles feature to recommend other articles that are similar to the user's tastes, based on the user's previous actions or explicit feedback. It also represents a user within the same functional space. Some user-related features may be provided explicitly by the user. For example, a user selects "Entertainment Apps" in their profile. Other features may be implicit based on previously installed apps. For example, suppose a user installed another app published by Science R Us.

The model should recommend relevant articles for that user. To do this, you first need to choose a similarity metric (such as dot product). Next, we need to set up the system to score each candidate item according to this similarity metric. Note that the model does not use information about other users, so the recommendations are specific to this user.

Advantages and disadvantages of content-based filtering 

advantage

The model doesn't need data about other users because the recommendations are specific to that user. This makes it easier to scale to large numbers of users.
This model can capture a user's specific interests and recommend niche items that most other users don't.
Cons

This technique requires a lot of domain knowledge, as the feature representation of the elements is created somewhat manually. Therefore, the model can only be as good as the hand-made properties. Models can only make recommendations based on the user's existing interests. In other words, this model has limited ability to extend the user's existing interests.  

SFIA level 

for the recommender system, I am in level 5 which is Ensure, advise.

 I have been able to fully responsible for meeting allocated technical and/or group objectives.  recommender system has to help me to analyze, design, plan, execute, and evaluate work to time, cost, and quality targets.

I have  Makes decisions that impact the success of the assigned test, i.e. results,   it Has significant influence over the allocation and management of resources appropriate to given assignments.  recommended system have Ensures users’ needs are met consistently through each work stage.

I can Perform an extensive range and variety of complex technical and/or professional work activities. Undertakes work that requires the application of fundamental principles in a wide and often unpredictable range of contexts.  I can understand the relationships between my search and the recommender someone else will receive.

Assesses and evaluates risk.

Takes all requirements into account when making proposals.

Shares own knowledge and experience and encourage learning and growth.

Creatively applies innovative thinking and design practices in identifying solutions.

  I can Apply knowledge to help to define the standards that others will apply.

References

https://www.google.com/search?q=recommendation+system&tbm=isch&ved=2ahUKEwjNn_iblYT-AhUOmicCHTl0A14Q2-cCegQIABAA&oq=RE&gs_lcp=CgNpbWc

https://medium.com/@toprak.mhmt/collaborative-filtering-3ceb89080ade