. The system that implements the process of automated determination of indicator characteristics of social network profiles, as well as determining the response of social network participants to certain events has been developed and tested on the real text examples and data. The developed system showed that proposed approach has substantially improved the accuracy and the amount of the characteristics of user profiles inside the Twitter social network by using the approaches that combine various natural language processing models and technologies.
Social networks are very actual research field in computer sciences today because a
vital information for most people on the planet is already in the digital form. Consider
an approach to determining the indicator characteristics of profiles of participants in
social networks (SN) based on the modeling of some generalized image. Typically,
collecting data to solve such and similar problems is sought by 1) determining the set
of significant (indicator) characteristics of related objects (profiles of SN participants),
2) obtaining values of such characteristics for one object, 3) accumulating statistics for
values for of all objects under study. The next, fourth stage should ensure that you gain
useful knowledge of statistics in one way or another - for example, statistical analysis,
mashing learning, deep learning. The result is a breakdown of the primary set of objects
into specific classes or categories [
We take advantage of the fact that a large part of the indicator characteristics of
the profiles of SN participants is verbally assigned, that is, belongs to a plurality of
words of some natural language. The idea behind the proposed approach is to use
linguistic methods of analyzing the plurality of words that characterize a particular profile
of a SN participant in order to obtain a concise natural-language description of that
participant, understandable to humans. We formalize the problem by using the model
of associative imaginative thinking of the person, first proposed in [
The subject of formalization will be certain mnemonic processes inherent in man.
In [
In terms of the model of figurative thinking, a basic synthesis operation provides,
according to Vygotsky [
We show the fundamental possibility of constructing a resolution image within the
model of the mechanism of memory [
Focus Weight Max(Weighti ), iN (1) where Focus–Bi is binary image code in focus of attention; Focus–Weight is the weight of the associative relationships of the image in the focus of attention with the imagescomponents of the emotion vector Vector–Set; Weighti is the weight of associative relations of the i-th image of the image ensemble (IA) with the images Vector–Set.
Formally, we will look for the synthesis or construction of a solution image as an
InsZX statement similarly to work [
2. Operation New–OM transfer of images-components of emotion vector Vector– Set in IA RAM: 3. Operation New–Vector insertion into the vector of emotions Vector – Set values from the Choice – Set stack:
4. Operation New–Choice uploading to the Choice – Set stack of RAM memory:
Vectori is binary code of the i-th image of the emotion vector Vector–Set;Bi–OM is binary code of IA RAM.
We add to the algebraic BAS system [
1. Search operation of Seek i-1 unit (s) in Bi-OM binary code and End-Bi
predicate, which is true when not all bits of code have been viewed
(2)
(3)
(5)
(6)
n
Bi Vectori NewOM Bi OM .
i1
(4)
In graph diagrams for algebraic constructs, we include the following notations of
structural programming operators [
Do – loop by parameter or condition; * – composition.
Consider the solution of problem (1) on the basis of such constituent operators.
1. Operator Pyramid–Images of binary code decomposition Bi–OM to the codes of the image-components. As a result of the action of the operator presented in Figure 1, the well-known Bi – OM code defines the codes of the images that are part of the IA. 0 i
Do
Bi
Endi+1 i
Seek»1»i k
For this purpose, a cycle along the length of the Bi – OM code is organized, if there is a unit, then another element with a unit at this position and zeros on all others is added to the list of images (the cycle is limited by the number of units n in the Bi – OM binary code):
Pyramid Im ages :: 0 i *{[End Bi] (i 1 i *
k 1 n Seek"1"i k * Conc"0" "1" Conc"0" Bi Ii )}, j1 j k 1 (7) where Bi–Ii is the binary code of the i-th image in RAM; Conc“0” is concatenation of "0" characters into binary code digits.
2. Operator OM–Change–Vector (figure 2) simulates the launch of the mechanism of constructing the image of the untying, because New–Choice, New–OM and New–Vector operations are performed sequentially:
New-Choice
New-OM 3. The InsZX unblocking constructor (Figure 3) runs the corresponding OM – Change – Vector mechanism and submits its results to the Orient – Reflect Oriented Reflex model of p.4.5.3.:
(9) InsZX
OM-Change
Vector Fig. 3. A Graph of the operator InsZX.
Orient-Reflect
The algorithm for constructing the image of solving a problem situation is implemented
within the model of the orienting reflex [
Let us illustrate the possibilities of the proposed approach by solving problem (1)
for experimental data of a cross-sectional test example “Semantic WEB Standards”
from [
Formally, in accordance with the statement of problem (1), we define Check– Set=011111000000, Vector–Set={000000010000, 000000000010, 000000000001}. As a result of the action of the OM-Change-Vector operator, the Vector – Set linguistic images are transferred to the IA and replaced by the LI from the Check – Set. The Orient – Reflect oriented reflex operator consistently determines the weights for all linguistic images (Bi – OM = 000000010110), which ultimately results in the largest of them being Focus – Weight = 8 for the LI language (56). However, even with the limited natural-language material of the cross-cutting example, the “network (17)” is gaining weight 4, which makes it a competitor to the “language (56)” LI in generalizing meaningful concepts of the Semantic WEB Standards topic.
Thus, based on a formal look at the phenomena of the ensemble of images,
emotions vector, focus of attention and orientation reflex, the possibility of constructing an
image of solving a problem situation by means of an algebraic system is shown. Unlike
works [
New application based on the model for determine and detail the image-solution of the problem situation is proposing. The program fulfills the tasks of analyzing the reactions of users of the social network Twitter to current news, as well as determining indicator characteristics of the profiles of participants in this social network.
This development is planned to be used to collect statistical information about the
response of users to an event that occurs in real time, as well as the analysis of indicator
characteristics of network user profiles. The goal is to analyze statistical information
and display it in the form of histograms and tables. Data collection, analysis and
processing takes place on the device itself, which increases the security of the data the user
is working with [
Like the vast majority of programs for Android, the program consists of a graphical interface and other application resources (graphical, text and sound resources), as well as the business logic of the application.
For the program to work, you need to configure the access settings for the Twitter social network API. These parameters are key: value pairs: • consumer key; • consumer secret; • access token; • access secret.
These values are used when sending requests to the servers of the social network and
necessary for the process of collecting information. To get these parameters, you must
have a user account with developer privileges on Twitter. These parameters are unique
for each user and allow identifying his program when accessing data on the server Data
collection, analysis and processing takes place on the device itself, which increases the
security of the data the user is working with [
User interaction with the program starts from the main screen. In the center of the screen, there is an input field where the user can enter a user nickname whose profile he wants to analyze or a hashtag to which some discussion on the network is attached. Nicknames should be entered with the use of the symbol "@" in front of the nickname itself without spaces between them. In case it is necessary to analyze the reaction to the news, the user enters the corresponding hashtag starting with the input symbol "#".
After that, the user can click on the “Start analysis” button, which will start the process of analyzing user reactions to events on the network or analyzing the profile of the network member depending on the input.
In addition, on the screen there are two additional buttons “Use test data (tags)” and “Use test data (user)”. Each of them starts the process of analyzing relevant content using pre-stored data from the Twitter network. This is necessary if the user wants to see the result of the work of the program, but does not have a Twitter developer account and therefore cannot access its API. In this case, the program will not access the Twitter API, but will use pre-prepared response files from Twitter servers.
The main screen of the application is shown in Figure 4.
When entering a username or hashtag, the system first calls the Twitter API. The corresponding module of the program sends the locks to the server using the previously specified access parameters, as well as the user's request. The server returns data in JSON format (Figure 5).
At the stage of parsing each reaction of the user, the text of the reaction itself is extracted and divided into tokens, after which these tokens are added to the list of tokens. The list of tokens is necessary for further text processing, since many other text processing methods are needed in preliminary tokenization.
Next, the program determines the names of people, locations and dates that are in the text of tweets using the Named Entities Recognition API library Apache OpenNLP.
Then, using the Chunker API, the program selects keywords and phrases (sets of words combined in meaning and grammatically) in tweet texts. For this function to work, the program needs a list of tokens, as well as the result of the POS Tagger API, whose task is to determine the parts of speech for each word in the incoming text. This will allow the Chunker API to identify words and phrases related in meaning in the text, which in the end result will help determine the reaction of a particular user to an event on the network, as well as to the general reaction of users.
In the case of analyzing the profile of a Twitter participant, the Language Detector API is additionally applied, which, based on the model embedded in it, recognizes the language of the text and presents a list of languages that it was possible to recognize together with the probability coefficient that this language is the language of the text.
To study and analyze the reactions of users to events, the responses of users on the Twitter network regarding events in the network united by the tag “#ukraine” were selected as input data for analysis, which means that if a given tag or word is present on a tweet, it will fall into search results.
To conduct the study, 1,500 user responses in English were collected, since the recognition model works with English speech. This figure is small for real analysis, since it covers only a small percentage of network users' reactions, but it is dictated by the limitations of the Twitter API. This limitation can be circumvented by sending requests at fixed intervals, after which the request counter is reset to zero and new requests can be sent. However, this approach will not work in the context of a mobile application, since even if it will collect data on demand in the background for a certain period of time, there is a risk that the system will stop the background process. For the test needs, this number of tweets is enough, because the goal is not to collect the most accurate statistics, but to demonstrate an approach to solving the problem.
Figure 6 shows the results of an analysis of user reactions to events on the network.
The user can interact with the graphs, zooming in and scrolling through them to see all the data.
In the course of processing the results of the analysis of user reactions, the program identified the five most common names in the text of tweets, they turned out to be "Donald Trump", "Joe Biden", "Rudy Giuliani", "Vladimir Putin", "Volodymyr Zelensky" (Figure 7). With small amounts of data (1-5 thousand tweets), there is no need to display more results, since often the names of other people are rarely found in tweets, more often they are in the @nickname format.
Named Entity Recognition API was able to highlight the names of locations in the text, in this case, the names of the most mentioned countries. It is worth noting that, despite the high accuracy of the results obtained, the model may not determine the names of little-known cities, which may be a problem when analyzing more “local” events in the network.
Figure 9 shows a diagram with the most common keywords and phrases in the texts of reactions to news on the network.
It is worth noting that the approach using Tokenization, POS Tagging and Chunking API gives much greater accuracy and greater semantic load embedded in the final results compared to using Tokenization and Tuples (for generating pairs of keywords in the text). The data approach works well in English, since it has a strict word order in a sentence, therefore there is a high probability of finding matches among the received words and phrases.
To study and analyze the profile of a member of a social network, the profile of the current US president was chosen, as well as his tweets in the amount of 2500 pieces. This figure is small for real analysis, since it covers only a small number of recent user tweets and is dictated by the limitations of the Twitter API.
Figure 10 shows the results of the analysis of the profile of a member of the Twitter network. The analysis results show that the system recognized four languages that were supposedly used when writing tweets by a user. The result with highest balls was the accuracy of the English language, which is true. The remaining languages were recognized, since the text could contain names, names or locations that are written similar to these languages, as well as due to errors in the recognition model. Figure 11 shows the pie diagram with the languages detected in the user tweets.
We also analyzed the most mentioned places and personalities on the same principle as when analyzing user reactions to events on the network. These results were put on one chart in order to determine the dependence between the name and location.
With this approach, between related pairs of names and locations, the difference in the frequency of references will be insignificant. However, in the case of this user and data set, this dependence was not detected.
Summing up, it can be argued that the created program successfully fulfills the tasks of analyzing the profiles of members of a social network, as well as determining the response of users to events in it.
In the course of the work, a new coordinated approach to solving the above problems was proposed, implemented using the Apache OpenNLP library. It made possible to increase the accuracy of the received results, which was confirmed by the testing results of the developed program. 4
The system for implement the process of automated determination of indicator characteristics of social network profiles, as well as determining the response of social network participants to certain events has been proposed in the article. Based on a formal look at the phenomena of the ensemble of images, emotions vector, focus of attention and orientation reflex, the possibility of constructing an image of solving a problem situation by means of an algebraic system is shown. The architecture and main modules of the proposed system were developed and tested on the real text examples.
To develop the software for solving the goal of research, the Kotlin programming language was chosen because of it and its ease of use when creating Android applications, as well as Apache OpenNLP library for parsing data about news and network users.
So, the developed system showed that proposed approach has substantially improved the accuracy and the amount of the characteristics of user profiles inside the Twitter social network by using the approaches that combine various natural language processing models and technologies.
According to the authors, the further development of the research subject is the study of different natural language processing models, which can be used as base of the proposed system for definition of indicator characteristics of social networks participants’ profiles.