Presentation on how to chat with PDF using ChatGPT code interpreter
Recommender system
1. R E C O M M E N D E R
S Y S T E M
F I D A N H A S A N G U L I Y E V A 6 7 2 . 7 E
2. W H AT I S R E C O M M E N D AT I O N
S Y S T E M ?
• A recommendation engine is a system that suggests
products, services, information to users based on
analysis of data. Notwithstanding, the recommendation
can derive from a variety of factors such as the history
of the user and the behaviour of similar users.
• With the use of product recommendation systems, the
customers are able to find the items they are looking for
easily and quickly. A few recommendation systems
have been developed so far to find products the user
has watched, bought or somehow interacted with in the
past.
3. H O W D O E S I T
W O R K ?
• In order to provide customers with service
or product recommendations,
recommendation engines use algorithms.
Lately, these engines have started using
machine learning algorithms making the
predicting process of items more accurate.
Based on the data received from
recommendation systems, the algorithms
change.
• Machine learning algorithms for
recommendation systems are generally
divided into four categories; content-based
filtering, collaborative filtering, and
knowledge-based system.
4. R E C O M M E N D AT I O N E N G I N E
P R O C E S S E S D ATA I N F O U R P H A S E S
• Classic recommender system processes data
through these four steps:
Collecting
Storing
Analyzing
Filtering
5. C O L L E C T I N G T H E D ATA
• Whereas, implicit data may consist of a search log,
order and return history, clicks, page views, and
cart events. This kind of data is collected from any
users who visit the given website.
• Collecting behavioral data is not difficult, since you
can keep user activities logged on your website. As
each user likes or dislikes various items, their
datasets are different. During some time, when the
recommender engine is feed with more data, it
becomes more clever
• And the recommendations become more relevant
too, so the visitors are more inclined to click and
buy.
6. S T O R I N G T H E
D A T A
• To have better recommendations, you
should create more data for the
algorithms you use. It means that you
can turn any recommender project into a
great data project quickly. You can decide
what type of storage you need to use
with the help of the data you use for
creating recommendations. It is up to you
whether to use a NoSQL database or a
standard SQL database or even some
sort of object storage. All of these
variants are practical and conditioned
with whether you capture user behavior
or input. A scalable and managed
database decreases the number of
required tasks to minimal and focuses on
the recommendation itself.
7. A N A LY Z I N G T H E D ATA
In order to find items with similar user engagement data, it is necessary to
filter it with the use of various analyzing methods. Sometimes it is necessary
to provide recommendations immediately when the user is viewing the item,
so the type of analysis is required. Some of the ways to analyze this kind of
data are as follows:
• Real-time system
In case you need to provide fast and split-second recommendations you
should use the real-time system. It is able to process data as soon as it is
created. The real-time system generally includes tools being able to process
and analyze event streams.
• Near-real-time analysis
The best analyzing method of recommendations during the same browsing
session is the near-real-time system. It is capable of gathering quick data
and refreshing the analytics for few minutes or seconds.
• Batch analysis
This method is more convenient for sending an e-mail at a later date since it
processes the date periodically. This kind of approach suggests that you
need to create a considerable amount of data to make the proper analysis
8. F I LT E R I N G T H E D ATA
• The next phase is filtering the data to provide relevant recommendations to the users. For
implementing this method, you should choose an algorithm suitable for the engine you use.
There are a few types of filtering, such as:
Content-
based
filtering
Collaborative
filtering
Knowledge-
based
filtering
9. C O N T E N T -
B A S E D
F I LT E R I N G
• Content-based filtering is based on a
single user’s interactions and
preference. Recommendations are
based on the metadata collected from
a user’s history and interactions. For
example, recommendations will be
based on looking at established
patterns in a user’s choice or
behaviours. Returning information
such as products or services will relate
to your likes or views.
• A particular form of the content-based
recommendation system is a case-
based recommender. These evaluate
items’ similarities and have been
extensively deployed in e-commerce.
10. • To check the similarity between the products or mobile phone in this example, the system computes
distances between them. One plus 7 and One plus 7T both have 8Gb ram and 48MP primary camera.
• If the similarity is to be checked between both the products, Euclidean distance is calculated. Here,
distance is calculated based on ram and camera;
• Euclidean distance between (7T,7) is 0 whereas Euclidean distance between (7pro,7) is 4 which means
one plus 7 and one plus 7T have similarities in them whereas one plus 7Pro and 7 are not similar
products.
11. C O L L A B O R AT I V E F I LT E R I N G
• Collaborative filtering casts a much wider net,
collecting information from the interactions from
many other users to derive suggestions for you.
This approach makes recommendations based
on other users with similar tastes or situations.
For example, by using their opinion and actions
to recommend items to you or to identify how
one product may go well with another. ‘Next
buy’ recommendations is a typical usage.
Collaborative filtering method usually has higher
accuracy than content-based filtering; however,
they can also introduce some increased
variability and sometimes less interpretable
results. They are especially weak in the
absence of previously collected data. Without
meaningful information on others, it becomes
harder, naturally, to participate in any single
person actions.
12. S I N G U L A R VA L U E D E C O M P O S I T I O N
A N D M AT R I X - F A C T O R I Z AT I O N
• Singular value decomposition also known as the SVD algorithm is used as a
collaborative filtering method in recommendation systems. SVD is a matrix factorization
method that is used to reduce the features in the data by reducing the dimensions from N
to K where (K<N).
• For the part of the recommendation, the only part which is taken care of is matrix
factorization that is done the user-item rating matrix. Matrix-factorization is all about
taking 2 matrices whose product is the original matrix. Vectors are used to represent item
‘qi’ and user ‘pu’ such that their dot product is the expected rating.
13. K N O W L E D G E - B A S E D S Y S T E M
where suggestions are based on an
influence about a user’s needs and based on
a degree of domain expertise and
knowledge. Rules are defined that set
context for each recommendation. This, for
example, could be criteria that define when a
specific financial product, like a trust, would
be beneficial to the user. These do not, by
default, have to use interaction history of a
user in the same way as the content-based
approach is, but can include these as well as
customer products and service attributes, as
well as other expert information. Given the
way the system is built up, the
recommendations can be easily explained.
But building up this type of framework can be
expensive. It tends to be better suited to
complex domains where items are
infrequently purchased or hence, data is
lacking.
14. H Y B R I D M O D E L S A N D D E E P
L E A R N I N G
• The most modern recommendation engine algorithms, and the
kind we use here at Crossing Minds, leverage deep learning to
combine collaborative filtering and content-based models.
Hybrid Deep Learning algorithms allow us to learn much finer
interactions between users and items. Because they are non-
linear, they are less prone to over-simplify a user tastes.
• Deep learning models can represent complex tastes over
various range of items, even from cross-domain datasets (for
instance covering both music, movies and TV shows). In
Hybrid Deep Learning algorithms, users and items are
modeled using both embeddings that are learnt using the
collaborative filtering approach, and content-based features.
Once embeddings and features are computed, the
recommendations can also be served in real time.
15. B E N E F I T S O F T H E R E C O M M E N D A T I O N E N G I N E
• A recommendation engine can significantly
boost revenues, Click-Through Rates
(CTRs), conversions, and other essential
metrics. It can have positive effects on the
user experience, thus translating to higher
customer satisfaction and retention.
• Netflix presents you with a much narrower
selection of items that you are likely to
enjoy, instead of having to browse through
thousands of box sets and movie titles. This
capability saves you time and delivers a
better user experience. With this function,
Netflix achieved lower cancellation rates,
saving the company around a billion dollars
a year.
16. W H AT A R E T H E C O M M O N C H A L L E N G E S
A R E C O M M E N D E R S Y S T E M F A C E ?
1.Sparsity of data. Data sets filled with rows and rows of
values that contain blanks or zero values. So finding ways to
use denser parts of the data set and those with information is
critical.
2.Latent association. Labelling is imperfect. Same products
with different labelling can be ignored or incorrectly
consumed, meaning that the information does not get
incorporated correctly.
3.Scalability. The traditional approach has become
overwhelmed by the multiplicity of products and clients. This
becomes a challenge as data sets widen and can lead to
performance reduction.