Nowadays the users generate considerable amount of information in the Internet. Therefore, they have a deal with the issue of how to retrieve the required information perhaps much more relevant than he/she supposed. Moreover, the mix of diferent data sources usually muddles users in the case of searching for the instructive information. The authors of this paper will introduce the approach based on text processing and formal concept analysis in order to structure the information from a variety of sources, particularly, social media communities. Additionally, they will clarify the relations between community posts with the same topic, where these relations become a recommendation tool for the user's decision making. In conclusion, the authors will build a diagram that will be a convenient visualization tool in an efort to structure information that was obtained from various sources.
Due to a rapidly increasing amount of data which is generated by users of the
Internet, how to deal with this data becomes the most important issue. For
instance, in 2019, 4.39 billion people were registered in services provided by social
networks, which is 366 million (9%) more than in January 2018 (information
provided by ’We Are Social agency’ and ’Hootsuite’ service [
Copyright c 2020 for this paper by its authors. Use permitted under Creative
Commons License Attribution 4.0 International (CC BY 4.0).
[
Current research that based on the analysis of social network data is
presented in the following works [
This paper includes four-section: an introduction, approach explanation, experiment computation, and conclusion. Firstly, the authors will explain briefly the theory of FCA and basics semantic analysis methods.Secondly, the authors propose a solution based on methods of semantic data analysis and FCA. Finally, they will explain an experiment implementation based on the social network dataset and also they will propose to visualize the results of computation by using concepts lattice as a diagram, in which each circle will be marked in a certain color. 2
2.1
In order to study the text data ,in particular, the meaning of the text obviously
the authors deal with semantic analysis [
= 0, ..., , then the set of words = 0, ..., when the word includes in text ⊆ for ∈ and
∈ . We deal with tokenize, lemmatizer, removing stop words and parts of speech other than nouns in order to compute keywords. So the final set of keywords is
which satisfy between objects and attributes explains ⊆ × for ∈ , ∈ when , if object has an attribute .
Formal concept is pair (,
) : intent of formal concept (, between (2 , ⊆) and (2 , ⊆), ′ = called the concept lattice B (, , ⊆ ,
⊆
) [
satisfy the Galois connection and ′ = . Where
and - extent and ). The ordered set of all formal concept forms is 2.3
In a previous subsection, the authors have defined the main mathematical method
founded on formal concept computation. Therefore, the ordered concepts build
the concept lattice, which is visualised by diagram, Hasse. Every set of formal
concepts has a great common subconcept as supremum. Its extent consists of
those objects that are common to all extents of the set. Every set of formal
concepts has a least common superconcept, the intent of which comprises all
attributes which all objects of that set of concepts have. Additionally, ordered
this way the set should satisfy the axioms defining a lattice, there are
commutative, associative, absorption laws. Thus, a complete lattice is an ordered
poset in which all subsets have both a supremum and an infimum [
Each post includes specific metadata that describe users’ personal interest in
the particular post, they are likes and reposts. Concerning of those measures we
want to compute for every concept the average values of likes and reposts, and
after we are going to clustering the ordered set of concepts by using k-means
(KMeans) [
The authors have investigated the social network dataset concerning the common
topic that takes place in the social network communities. Thus, this dataset
includes information about Id post, the text of the post, value of users’ attitudes,
and value of reposts (see Tab. 1). It was obtained from ’Vk’ social network, It
community, which text of the posts in Russian and English [
In order to compute keywords,the authors used semantic analysis which was explained before. We computed the formal context according to set of key-words will be presented as attributes of a formal context and the set of objects - a number of posts (see Tab. 2).
According to FCA theory the set of formal concepts was computed. Although the set of posts includes no more than 100 items, the number of obtained formal concepts is quite huge,approximately 800 items. Moreover, for each formal concept the average of users’ likes and average of reposts were computed (see Tab. 3).
However, the great number of ordered formal concepts build a huge diagram, for this reason, the authors have explained only the set of concepts and their clustering. According to Tab. 3 in which an object ′ ′ takes place by following the next set of formal concepts. We deal with diagram in order to concentrate on users preferences,for instance, the user who takes an interest in hiking a job or career development and he ∖ she takes an interest on ′ ′ may be recommended the next set of posts: 1297419, 1297266, 1296779 and etc. This argument makes sense due to lattice lines properties that explain the relation between formal concepts.
By using the measures (likes and reposts) we computed three concept clusters, where centroids according to likes are ( : ) : ( : 108.333, : 55.3985, : 24.227), and reposts are ( : 9.1345, : 3.254). Additionally, each circle of the diagram has its own color that explains a cluster number. This opportunity allows users to visualize their searching and to rank posts according to measures. For instance, the user concentrates on ′ ′ therefore high priority has 1296779 and after 1296779,1297419 as stated by post likes. However, measure repost supports a few users so its values explain only two clusters but another hand this measure makes more sense than likes measure in the opinion of an issue of ranking.
This paper has discussed the approach which manipulates the social network dataset by using FCA, particularly, it assists the user of social media to deal with communities’ posts. The authors take into account the specific framework of a dataset that is satisfied with a variety of social services. Moreover, the authors concentrated on such criteria as accessible, high quality, immediate, and relevant information that can be provided to the user by his/her request. Although this approach partially satisfies this number of criteria, so it will be tried to improve in the future by using FCA advantages.