Improving User Topic Interest Profiles by Behavior Factorization

        It is true that one of the significant aims of building content recommendation is the construction

of person ahead user profiles. In this case the two problems of feature engineering and the proper

utilization of the signals using data mining and machine learning approaches are faced.


Therefore, the behavioral factorization approach is discussed in here so that the recommendation

of matrix factorization is applied.


Moreover, the contribution of this can be pointed as introducing general nation of separating the

behavioral engagement types, developing a method to perform behavioral factorization and the

evaluation of behavioral factorization on a large scale data set.


Simply the process of this approach can be classified into various division. At the initiative

situation the task is to build the user profiles. In order to construct topic interest profiles of social

media systems personalized recommenders the matrix factorization embedding models are

followed.

The second step in this approach is identified as the personalized user profiles in social media.

For an instance after creating the profiles based on content and item preferences providing the

personal recantations for social media items as bookmarks and social software can be taken.

The next related work is the contextual personalization where the comments on the post on what

topic and when are concerned. Collection matrix factorized context aware generative models are

identified by researchers. Also triadic factorization based approach that is a combination of the

content of two or more different media platforms are also used in practice.


The behavioral factorization captures an important role in here improving the quality of user

profiles. Based on the different type of behaviors various identifies are proposed in here and it

stresses the qualitative significance of this method.


Apart from that Google+ Behavioral analysis also used while if utilizes the behavioral

factorization to model different behavioral types First data description that analyzes all the user

action in all public posts and the representation of each record as tuple can be identified. There

after measuring differences among behaviors from the each aggrieve enactive of posts with the

particular type of behavior is applied and in the final stage the discussion or the analysis of the

result are done. It emphasizes that general non-behavioral specific use profiles may not perform

well in applications that stress different behaviors types.

Therefore, so as to recover from those problems building a multiplies profiles for a user to

represent the behavioral types is presented here. The input behavioral signals that focuses on

clarifying the behaviors on social media posts as inputs and outputs can be numerically

represented as a set of tuples.


Also defining the user profiles as set of vectors in the features space as one of the problem

definition can given


Now the behavioral factorization approach introduced by us is discussed. In the first step

matrices of different behavioral types are build.


Apart from the typical Marty factorization technique the inputters. Item matrix three multiple

matrices are concerned. One of them are the Behavior-Nonspecific user topic matrix BNUM that

discussed of each entry indicating a user’s implicit interest’ on a particular topic.


The next model is the Single Behavior- specific user topic matrix SB-sum. So it generates

separate user topic matrices. The final one is the combined behavior specific user topic matrix

CBSUM where in creating a matrix the single behavior type is represented.


The second step in our approach is the learning latent embedding space. The matrix factorization

technique for building user topic profile as the baseline method is identified. So within the

baseline model mainly the differences between the explicit interest vs implicit interest signals is

discussed.


Also in the behavior factorization model (BF) the users’ different behavior types are separate and

topic preferences for each user are generated. Along with these two aspect the user profiles are

built as the last stage.


In here, with corresponding to the input matrices three types of user profiles can be identified.

Namely they are behavioral non-specific user profile(BNUD), Single behavior specific user

profile (SBSUD) and combine Behavior specific user profile(CBSUP).


The direct profile building (DPB) uses the embedding factors so as to generate the complete user

profiles. Also the weighted profile building stresses how to use linear regression with stochastic

gradient descent to learn these parameters. So it can be used to generate either BNUD, SBSUP or

CBSUP, depending on particular applications.


Through the evaluation of the behavior factorization approach in the aspect of behavior. Non-

specific user topic profiles or behavior specific user topic profiles potential applications as well

as the limitation and future work are identified.


So as the final idea of the behavioral factorization which is introduced as a way to build user

topic interest profiles in social media can be elaborated as a meaning of inspiring other

researches in determine the model user context in topic modeling and recommender systems.