The work is devoted to the development of an IoT ecosystem model for the semantic analysis of Internet posts. The need to create an IoT ecosystem that will allow offline assessment of the emotional state of Internet posts is substantiated. The work provides a graphic representation of the IoT ecosystem and describes its components. The functional model of the IoT system of semantic analysis of Internet posts is described. The principle of learning a neural network, which is the basis of the operation of the IoT system, is presented. The algorithm of the interlocutor bot, which is a means of receiving Internet posts into the system database, is given. Verification of the work of the IoT system of semantic text analysis was carried out. The result of emotional coloring was obtained with a probability of 94%.
(A. Biloshchytskyi);
2023 Copyright for this paper by its authors. CEUR
ceur-ws.org
The task of analyzing the test is relevant at any time, both in peacetime and in military time. The analysis of the text in peacetime is most often aimed at the analysis of the posts of people in social networks in order to identify harm to themselves. In wartime, this task acquires characteristic features of searching for collaborators.
Today, there are several methods of text analysis. In manuscript 3] proposed two innovative unsupervised approaches for the combined modeling and interrelated accomplishment of the two tasks. Both approaches rely on respective Bayesian generative models of topics, contents and clusters in textual corpora. Such models treat topics and clusters as linked latent factors in document wording. In particular, under the generative model of the second approach, textual documents are characterized by topic distributions, that are allowed to vary around the topic distributions of their membership clusters. Within the devised models, algorithms are designed to implement Rao-Blackwellized Gibbs sampling together with parameter estimation. These are derived mathematically for carrying out topic modeling with document clustering in a simultaneous and interrelated manner. A comparative empirical evaluation demonstrates the effectiveness of the presented approaches, over different families of stateof-the-art competitors, in clustering real-world benchmark text collections and, also, uncovering their underlying semantics. Besides, a case study is developed as an insightful qualitative analysis of results on real-world text corpora.
The text often contains images in the form of text that need to be analyzed. It is not only the comments on the image that are important, but also the posts on the image themselves. 4] presents Parallel Deep Fusion Generative Adversarial Networks (PDF-GAN), which use two new components to mitigate inconsistent semantics and bridge the semantic gap between text and images.
Experiments evaluating the generalization of scientific articles and news reports show that the proposed framework achieves comparable or higher correlations with human judgments than popular evaluation models. This study also supports the role of a network of semantic connections in the representation and analysis of texts 5]. The study 6] evaluated the empirical impact of certain syntactic and semantic features on the accuracy of note-taking. An improved feature set that jointly uses syntactic and semantic features is proposed. No one will deny the fact that the use of neural networks to create text analysis software products is relevant. A neural network (artificial neural network) is a mathematical model that mimics the structure and functioning of biological neural networks in order to solve a variety of tasks, such as classification, regression, prediction, and generation 7].
The analysis of short texts and texts with a large set of words has significant differences. The authors of the article 8] propose a new Set-CNN approach. It is aimed at using text convolution neural network based on semantic expansion for short text classification.
The analysis of short texts of comments and reviews under the product in the online store allows to analyze the intentions of customers regarding the purchase of a series of products. Analysis of posts in social networks allows obtaining data on the psychological state and mood of the population. Using this analysis for peaceful purposes brings positive results for the development of society and technology.
Italian online news is analyzed using semantic network analysis and text analysis. For this, they use Semantic Brand Score. In the process of measuring the importance of energy community media, ideas for implementation and public awareness of energy communities are identified 9].
With the increasing complexity of the building process, it is difficult for project stakeholders to retrieve large and multi-disciplinary building information models (BIMs). A natural language interface (NLI) is beneficial for users to query BIM models using natural language. However, parsing natural language queries (NLQs) is challenging due to ambiguous name descriptions and intricate relationships between entities. To address these issues, this study proposes a graph neural network (GNN)-based semantic parsing method that automatically maps NLQs into executable queries. Firstly, ambiguous mentions are collectively linked to referent ontological entities via a GNN-based entity linking model. Secondly, the logical forms of NLQs are interpreted through a GNN-based relation extraction model, which predicts links between mentioned entities in a heterogeneous graph fusing ontology and NLQ texts. The experiment based on 786 queries shows its outstanding performance. Moreover, a real-world case verifies the practicability of the proposed method for BIM model retrieval 10].
In manuscript 11] proposed a new systematic methodology to address these issues and identify technology opportunities using a hierarchical semantic network and dual link prediction. The proposed methodology consists of three modules: 1) constructing the hierarchical semantic network based on SAO structures extracted from patents; 2) identifying technology opportunities in this semantic network through probabilistic-based link prediction; and 3) evaluating these opportunities via similarity-based link prediction. The viability and usefulness of the proposed methodology is proved by empirical analysis of the exploitation technology in the coal seam gas (CSG) industry. The results show that the hierarchical semantic network, including semantic and co-word relationships, can improve prediction accuracy.
So, after considering various directions and methods of applying semantic text analysis, we came to the opinion that this topic is relevant. Since the text we are analyzing is written by a person, the question arose to investigate the components of the ecosystem and build its architecture. Develop an IoT ecosystem model for semantic analysis of Internet posts. To check in practice the operation of the system for semantic analysis of Internet posts.
2. Analysis of the constituent components of the architecture for the task of semantic analysis of Internet posts IoT ecosystem
The Internet of Things combines real things into virtual systems capable of solving completely different tasks. The key idea is to connect all connectable objects together, connect them to a network to collect data and make decisions based on it 12].
The “IoT Ecosystem”: includes all the components that enable businesses, governments and users to connect their IoT devices, including control panels, panels tools, networks, gateways, analytics, data storage and security 12, 13]. The architecture of the IoT ecosystem depends on the problem to be solved. But in general, the main components of the IoT architecture can be listed.
The components of the IoT architecture are the data acquisition area, the data transmission network, and the data analysis unit (Figure 1).
The data collection area includes the user's gadgets and the user himself. The user (upper left corner of Figure 1) is an integral part of the ecosystem as the initiator of Internet posts 14]. The results of the analysis of Internet posts are sent to the user of the gadget from which the request was made, as a result of the analysis of the Internet post.
The data analysis unit (upper right corner of Figure 1) consists of a database that stores both input and output data, a semantic analysis system, and hardware.
The data transmission network consists of a wired and wireless part of data transmission. The user makes an Internet post using his own gadget. The post is sent to the application database with the help of the Internet. Then, if necessary, the text is analyzed by the semantic text analysis system and sent to the requesting user's gadget. The protection of information is ensured by coding of the data transmission network, provided by the provider and the support system of the cloud environments of social networks and the storage environment of the semantic data analysis system.
The hardware of the process of data transmission, reception and storage is not the purpose of this study. The prototype of the system for semantic analysis of Internet posts is placed on a stationary PC and searches for data with the help of a conversational bot 15]. The interlocutor bot is written in Ruby on Rails. The task of the interlocutor bot is to receive data from social networks. The bot was tested on posts, comments and posts by Facebook users. Figure 2 describes the algorithm of the interlocutor bot, which provides data for the system of semantic analysis of Internet posts.
Algorithm of the interlocutor bot: Step 1 Settings before starting work Step 2 Search for posts and information about contributors from open sources Step 3 Submission of the text of Internet posts to the system of semantic analysis of Internet posts Step 4 Execution of Step 1 - Step 3 before executing the system command or changing the settings.
The system of semantic analysis of Internet posts is built on the basis of a neural network. It reflects the emotional coloring of an Internet post. The works 16,17] describe the tasks of text analysis by means of a neural network. Figure 3 shows their classification. In the process of classification, the text is compared with already existing ones. Characteristic words and constructions are checked. For the emotional study of the text, it is necessary to take into account not only the constructions of the words in the sentence, but also the connections between the words.
s i s y l a n a t x e t f o s k s a T
3. IoT model of the system of semantic analysis of Internet posts At the first stage of working with the Internet post, we turn the text into a combination of zeros and ones. The neural network works with the digital description of the text. Let's turn the text into dense vectors. Each word is a separate token. The neural network learns with the teacher. Therefore, we compile a test and training data set. The training data set contains answers for semantic analysis of Internet posts for emotional coloring. Figure 4 shows part of the neural network.
At the entrance, we matter and return the value ℎ .
Feedback allows information to be
transmitted step by step to each node in the network. The neural network used in the system is recurrent.
Connects previously obtained results with future results. One of the characteristics of a recurrent neural
network is to define an error function for each training step. We will use binary cross-entropy (
( , ) = ( ) + ( || ) when ( )- entropy p;
( || )- Kullback-Leibler divergence from q to p.
The results of testing the neural network, which is the basis of the method of semantic analysis of
Internet posts for the emotional coloring of the text, are shown in Figure 5.
(
At the training stage, the neural network learned to classify Internet posts with a probability of more than 97% in 1 epoch (Figure 5). At the test stage, the neural network reached the maximum classification value. Carrying out step-by-step testing, we have a score in the range of numbers from 0 to 1 for each class. Figure 7 shows the functional model of the IoT system for the semantic analysis of Internet posts. Common functions include: system update; information about the system; system retraining; text analysis; introduction of new educational data.
Analysis of the use of IoT system for the semantic analysis of Internet posts Let's download the training sample in .csv format. Table 1 shows an example of a training sample with the following characteristics: ID, text, and emotional coloring of the text.
The emotional coloring of the text presented in Table 1 is represented by 1-presence of emotional coloring, 0- absence of emotional coloring in the text. In the process of learning a neural network under the guidance of a teacher, the following parameters were obtained to obtain an optimal solution: — the number of epochs — 2 and 5; — number of iterations for early termination — 250; — network error calculation — Binary crossentropy; — learning speed coefficient — 0.05 — optimizer — Adam.
5 epochs are selected for system training. As early as the third epoch, we get stable values from the neural network (Figure 8).
After training, we load the text into the system for analysis. The results of the system are shown in Figure 9. So, the text is 94% toxic, 18% moderately toxic, and 84% contains obscene language.
The result of this study is a description of the functional model of the IoT ecosystem for the semantic analysis of Internet posts. The work provides the structure of the IoT ecosystem and describes its components. Data verification was carried out on the example of a previously trained neural network. The network determined with 94% accuracy the emotional color of a short text, which was obtained with the help of a conversational bot from the Facebook social network. 7] Neural network. https://termin.in.ua/neyromerezha/ 8] Yajian Zhou, Jiale Li, Junhui Chi, Wei Tang, Yuqi Zheng, Set-CNN: A text convolutional neural network based on semantic extension for short text classification, Knowledge-Based Systems, Volume 257, 2022, 109948, ISSN 0950-7051, https://doi.org/10.1016/j.knosys.2022.109948. 9] C. Piselli, A. Fronzetti Colladon, L. Segneri, A.L. Pisello, Evaluating and improving social awareness of energy communities through semantic network analysis of online news, Renewable and Sustainable Energy Reviews, Volume 167, 2022, 112792, ISSN 1364-0321, https://doi.org/10.1016/j.rser.2022.112792. 10] Mengtian Yin, Llewellyn Tang, Chris Webster, Jinyang Li, Haotian Li, Zhuoquan Wu, Reynold C.K. Cheng, Two-stage Text-to-BIMQL semantic parsing for building information model extraction using graph neural networks, Automation in Construction, Volume 152, 2023, 104902, ISSN 0926-5805, https://doi.org/10.1016/j.autcon.2023.104902. 11] Zhenfeng Liu, Jian Feng, Lorna Uden, Technology opportunity analysis using hierarchical semantic networks and dual link prediction, Technovation, Volume 128, 2023, 102872, ISSN 0166-4972, https://doi.org/10.1016/j.technovation.2023.102872. 12] Internet of things technologies. Training manual [Electronic resource]: training. manual for study specialty 126 "Information systems and technologies", specialization "Information maintenance of robotic systems" / B. Yu. Zhurakovskyi, I.O. Zeniv; KPI named after Igor Sikorsky – Electronic text data (1 file: 12.5 MB). – Kyiv: KPI named after Igor Sikorskyi, 2021. – 271 p. 13] Tripathy B. Іnternet of Things (IoT): TeChnologies, AppliCations, Challenges and Solutions (англ.) / B. Tripathy, J. Anuradha. – Florida: CRC Press, 2017. – 334 с. 14] Steven T. Smith, Edward K. Kao, Danelle C. Shah, Olga Simek, Donald B. Rubin. Influence Estimation on Social Media Networks Using Causal Inference. IEEE Statistical Signal Processing Workshop (SSP), 2018. Available at: https://ieeexplore.ieee.org/document/8450823 15] Kravchenko O.V., Umanets I.S., Gevko O.O. Using Conversator Bots for Opinion Formation Web Community Participant// Intelligent decision-making systems and problems of computational intelligence: materials of the international of science conf., p. Iron Port, May 21-25, 2019 Kherson: Vyshemirsky V.S. Publishing House, 2019. - PP.93-94 ISBN 978-617-7783-02-1 (electronic edition) 16]Judicial expertise in civil proceedings // Linguistic expertise.
URL:https://stud.com.ua/6784/pravo/lingvistichna_ekspertiza. 17]Azhnyuk L.V. // Linguistic expertiseas a legaltool. URL: http://ekmair.ukma.edu.ua/bitstream/handle/123456789/11606/Azhniuk_Linhvistychna_eksperty za.pdf 18] Bekarystankyzy, A., & Mamyrbayev, O. (2023). ЕND-TO-END SPEECH RECOGNITION SYSTEMS FOR AGGLUTINATIVE LANGUAGES. Scientific Journal of Astana IT University, 13(13), 86–92. https://doi.org/10.37943/13IMII7575