Content-Based user having few ratings would not

Content-Based Recommendation is a
recommendation system that basically deals with data provided by user profile. The
user will be recommended a video which is similar to the ones that user
preferred in the past,those datas are obtaines from the previously viewed
history from the user profile , similarity between user profile1.

WORKING
PROCESS

We Will Write a Custom Essay Specifically
For You For Only $13.90/page!


order now

User profile  basically consist of ratings and number of
times viewed on a particular video. It is generated based on user’s interest
and favourites, and created profile used to make recommendations. The more
usage of user profile for recommendation, better the accuracy. New users have
to update their interests in their user profile which might take some time to
update their data which is later on used by the recommendation system for
suggesting the videos for the users satisfying based on their interests. Therefore,
a new user having few ratings would not be able to get accurate
recommendations. There are four operations performed in this system:

 User Raw Information
Collection Agent;  User Profile Analysis
Agent;  Content Recommendation Server;
Content Recommendation Client Agent2.

The content recommendation server manages the data  and sends them to the client according to the
client requirements. The content recommendation server deals with  the to-be-pushed contents . By TF-IDF
algorithm, the keywords are extracted from the raw contents and the content
profiles for each category are maintained which will be used as the metadata of
the contents. It is formatted into XML files2.  Thus , this metadata saves time and network
resource.

 The client  receives 
required  data from the server
based on users profile. By receiving the content metadata profile from the
server, the agent  compares it with the
local datas . Later on , it is represented as TF-IDF vectors  and  keyword weights, the relation between them is determined
by  some scoring heuristic, such as the
cosine similarity measure2.  

The workflow of the system involves three steps: Firstly, It
collects the data from the user  profile
history which is done by information collection agent. Secondly, It  analysis  the user preferences from the collected data  with the help of user profile analysis agent. Thirdly,
the content recommendation client agent uses obtained data  to select and filter the information received
from the content recommendation server2. The whole working process is
conducted on the client side with no interference from public servers2.

 

ALGORITHM

Two basic concepts used in
content-based recommendations system are Term Frequency(TF) and Inverse Document
Frequency(IDF). Term Frequency deals with the frequency of interests and favorites
in user profile. Whereas Inverse documents frequency deals with inverse of term
frequency among the whole data provided by user profile. These two concepts are
combined together in order to provide the recommendation for a user based on
the datas provided by user profile. The main purpose of using this concept is
to determine the weightage of the effect of high frequency interests in
determining the importances of an recommended video. It is stored as a vector
of its attributes in n-dimensional space and angle between vectors is
calculated to determine the similarities between the vectors.

 

ADVANTAGES
, LIMITATIONS

 The advantage of this recommendation system is
it analyze all the data provided by the user in their user profile and then it
recommends the video based on their interests (i.e.,user independence) , no
cold start for new item with not enough description or reviews , transparency
 which explains  how the recommender system works, that is
represented explicitly by listing features or descriptions.

The drawbacks of the used
recommendation system is limited content analysis which leads to less accuracy
of the recommendation system , very poor at observing the complex behaviours of
user based on their user profile , serendipity problem (mind cages for a
particular set of users based on their interests) which is also known as
over-specialization , new user who doesn’t
have enough  no of ratings required on
order to before a content-based recommender system can determine  user preferences and provide accurate
recommendations .