Digitization of socio-economic relations leads to the evolution of interaction between users and companies, manifested in communications through innovative information technologies. Marketing strategies of companies are focused on the digital environment since a significant number of users actively use the Internet in everyday life. Interaction with users in the digital environment involves the use of different content related to the interests of different groups of potential customers. The dynamic life of the modern population in the world, especially representatives of the Z and Alpha generations, requires companies to post new relevant content on an ongoing basis. The presence of high-level competition between companies in the digital environment allows users to form demand by their own needs and quickly reorient from one brand to another. The main channel for communications with the target audience in the digital environment is social media, which allows interaction with different groups of consumers through relevant marketing strategies. The population's natural need for social orientation and interaction with other people in conditions of digitalization has led to a significant increase in the popularity of various social media. For each social media, certain distinctive features lead to the use of specialized thematic content and specific communications between subscribers. In the process of interacting with companies and discussing brands, users can use text messages, stylized graphic images (emoticons, emojis, memojis, animojis, etc.), as well as photos and video content. Optimizing the company's marketing strategy involves comprehensive research of the target audience and identification of user reactions to various actions of brands in the digital environment. Artificial intelligence algorithms make it possible to establish, based on comments, the general level of positive or negative perceptions of companies' actions. Modern mathematical algorithms make it possible to transform graphic objects and text messages into information that characterizes the relationship of users to brands, their products, and the implementation of marketing strategies. The obtained results are used to optimize the use of digital marketing tools and ensure a high level of the target audience's loyalty to the company and its products.
The active development of machine learning
algorithms and their integration into artificial
intelligence allows for the development of new
digital products. Advanced technologies are
integrated into digital marketing tools and
bring the effectiveness of interaction with the
target audience to a qualitatively new level [
In today’s world, products with innovative features are constantly being introduced to the market, which quickly become obsolete due to the presence of a lot of competition between companies in the digital environment. Accordingly, the development of innovative information technologies and the directions of their integration into various types of economic activity arouses keen interest among scientists in different countries of the world. Machine learning algorithms make it possible to solve a large number of theoretical and practical problems, which contributed to the emergence of a significant number of research areas of the specified technology, its improvement, and integration into real economic processes. Artificial intelligence refers to an important scientific direction of applied research, which allows for improving modern products and endowing them with qualitatively new characteristics.
The work [
Features of artificial intelligence use in
digital marketing are presented in the work
[
The work [
The image can be used as a valuable source
of information about the behavior of users in
the digital environment, among scientific
works in this direction it is advisable to pay
attention to the following article [
The work [
The development of sentiment analysis occurs due to the evolution of science and information technologies. In the first stages, methods that did not require significant computing power were used. Lexicon-based analysis refers to the simplest of methods, allowing to identification of the sentiments of users, but it allows to estimate of the relationship of the audience only according to a basic set of words. Along with this, the presented approach does not allow taking into account the context of statements and sarcasm. Statistical methods allow more efficient processing of text messages by calculating various characteristics and classifying user utterances. AI-driven sentiment analysis includes such techniques as machine learning and deep learning. Hybrid methods allow to combination of various approaches in the process of sentiment analysis, using their advantages and obtaining the optimal result. Fig. 1 presents the basic techniques of centimeter analysis.
words as positive, negative, or neutral
Statistical characteristics of the text: word frequency, sentence length, use of punctuation The flexibility of using statistical methods according to the features of the text
Learning on large arrays of textual information and graphic objects with emotion labels
Using artificial neural networks to identify factors that shape the tonality of messages
identify hidden context, jokes, sarcasm etc.
several methods to increase efficiency
trade-off between ease of feedback processing and accuracy AI-driven sentiment analysis is the most effective way of identifying audience sentiments in social media and, thanks to the further evolution of machine learning and deep learning algorithms, will ensure the effectiveness of marketing strategies of companies in social media and the digital environment. Fig. 2 presents a comparison of Machine Learning and Deep Learning sentiment analysis techniques, which were proposed by Nhan Cach Dang, María N. Moreno-García, and Fernando De la Prieta. At the current stage of development in sentiment analysis, an approach that involves the use of emoticons and pictures in the implementation of machine learning and deep learning approaches is becoming widespread. The use of text information, emoticons, and pictures as an information base for analyzing the audience in social media allows for a significant increase in the accuracy of identifying the company’s emotional perception and its marketing actions. Machine learning and deep learning approaches are characterized by significant power and productivity, which contributes to increasing the effectiveness of the company’s
marketing strategy in social media and allows the establishment of close communications with the target audience in the long term.
When using standard methods of sentiment analysis, dictionaries of emoticons are formed, which are grouped according to specific emotional reactions. This approach involves grouping and marking with the involvement of employees. In the case of the implementation of machine learning algorithms, the use of a certain set of data is assumed, which involves the analysis of messages with the identification of polarity and the establishment of emoticons present in the text. In the process of identifying the emotional outline of emoticons, they are converted into numerical values, which allows the implementation of appropriate mathematical algorithms. Among the methods of converting emoticons into digital form, it is advisable to pay attention to the following approaches:
1. One-Hot Encoding. According to the presented method, each emoticon is transformed into a binary vector. The index for the corresponding emoticon takes the value 1, and the other indices are equal to 0. For example:
– [
This
approach
involves the calculation of certain statistical
indicators of the use of appropriate emoticons
in text messages. Count-based features allow to
transformation of visualized emotional objects
into useful information used in solving machine
learning tasks: classification, regression, and
pattern recognition [
The presented approach involves the calculation of the following indicators: A. The total number of emoticons in a certain text message, without taking into account the content of the emotion expressed by a specific graphic object. B. The
density of emoticons in a text message characterizes the ratio.
C. The number of emoticons by emotion.
Classification systems provide for the separation of three groups of emoticons based on emotions: positive, neutral, and negative for the needs of centimeter analysis.
shows the basic emoticons that represent each of the three groups of emotions. It should be noted that the list of emoticons may be updated during research, as there are differences in the selection of emoticons for different social media and specific consumer groups. Along with this, the study of reactions in different countries of the world also involves studying the features of communication models in the digital environment and identifying the most popular emoticons among the target audience.
Negative emotions 0 0 0 0 … 0 identifying the sentiments of the target audience in social media.
☹
3. Embeddings. The presented method allows
to transformation of emoticons into numerical
vectors of large dimensions and uses the
obtained data to implement various
machinelearning algorithms. Thanks to the evaluation
of the distances between the vectors
characterizing the corresponding emoticons,
their similarity in terms of emotional reactions
is carried out. Embeddings make it possible to
establish the essence of the emotional
characteristic of an emoticon according to the
context of the text in which this visualized
object is presented [
A. Pre-trained models. Natural language processing models based on embeddings are used in a large number of modern products (Alexa, Google Assistant, Google Translate, Siri, etc.). The presented concept has evolved significantly thanks to the increase in the efficiency of neural media. Among innovative models, it is advisable to pay attention to BERT (Bidirectional Encoder Representations from Transformers) (see Fig. 5) and GPT (Generative Pre-trained Transformer), which allow to representation of words in vector form and take into account the context of messages. Among the machine learning models, it is necessary to pay attention to Word2Vec, GLOVE, FastText, and Baseline, because the natural language processing methods presented allow analyzing emoticons integrated into the text and identifying the emotional direction of messages, as well as graphic elements. B. Custom embeddings. The creation of its objects is expected by the specifics of the company’s marketing strategy implementation in social media. According to the emoticons used, high-dimensional vectors are constructed.
The following custom embeddings for emoticons are highlighted: • Using textual data. If the company has its database, which contains emoticons with a description of the content of the text messages used, in certain cases it is advisable to use the custom embedding approach. Thanks to the application of certain machine learning algorithms, it is possible to train custom embeddings with a high level of probability to accurately identify the context of graphic objects in the text. • Unicode or numeric representation. By the numerical code of a specific emoticon, it is possible to independently create unique representations with a specific vector. It should be noted that to create vectors in this case it is advisable to use both one-hot encoding and learned embeddings. • Independent approach. This approach involves the independent creation of embeddings for emoticons thanks to the application of techniques for vectorization (one-hot encoding, frequency-based representation, Unicode indexing) and techniques for distributed representations (word embeddings, sequence models, attention mechanisms, dimensionality reduction techniques). The flexibility and adaptability of the independent approach allow to adjust the presentation of emoticons by the company by the characteristics of the target audience in the relevant social media in certain periods.
C. Transfer learning. Digital transformation is expected in the case of emoji embeddings due to the use of pre-trained models or embeddings [24]. To achieve optimal results, it is necessary to accumulate a large amount of data that will allow training of the model using the appropriate mathematical algorithm. The process involves the implementation of the following steps: • Pre-trained embeddings. Based on large volumes of text messages with emoticons, embeddings are taught. A variety of texts with integrated emoticons allows us to recognize the context and semantics of graphic objects. • Fine-tuning. The embeddings obtained at the previous stage must be configured by social media, characteristics of the target audience, and other factors of the internal and external environment. The optimization process involves the use of test data sets that can be generated by the probabilistic principles of constructing sample populations. Along with this, test sets can be formed based on targeting a certain group of consumers. • Domain-specific training. Implementation of unique models of communication with certain groups of consumers in social media involves setting the appropriate context for emoticons. It is advisable to use the resulting patterns of content load and context for learning embeddings [25]. • Transfer knowledge. The speed of transformation of the behavior of modern users in the digital environment requires the use of pre-trained embeddings for the identification of a new context and meaningful load in emoticons. Along with this, the trained models allow for analyzing the existing context of emoticons, which may not change for a long time [26].
Evaluate and adjust. Constant testing of emoticons thanks to the use of pre-trained embeddings allows the company to understand its customers and optimize the communication process to ensure an economically justified level of conversion. Performance improvement due to hyperparameter tuning should be done according to the evaluation of the resources spent and the results obtained.
4. Hybrid Approaches. Thanks to the combination of several techniques for identifying the content and context of emoticons, it is possible to increase the accuracy of determining the reactions of the target audience to the company’s actions in the digital environment and optimize marketing strategies for establishing effective communications in the long term [27]. Due to the multiplicative effect and using the advantages of each of the involved approaches, it is possible to obtain rich emoticon representations.
It is advisable to use the following approaches for emoji embeddings:
A. Symbolic and distributional representations.
B. Pre-trained embeddings and custom training.
C. Multi-modal fusion.
D. Attention mechanisms and sequence modeling.
E. Ensemble of models.
Sentiment analysis is one of the effective approaches that companies should use to identify the reactions of the target audience to certain actions within the framework of marketing strategies in social media. However, achieving optimal results is possible only if scientific approaches are followed at all stages of evaluating the attitude of users to the brand, its marketing strategy, posted content, etc. The main stages of analyzing the sentiments of the target audience in social media are:
1. Data collection. Social media act as an important digital marketing tool to interact with the target audience and promote the company’s products. The specificity of social media functioning involves the generation of large volumes of heterogeneous information that can be used by companies to optimize marketing strategies. Thanks to the use of web analytics tools, it is possible to accumulate data on the activity of users in the relevant social media on an ongoing basis. For sentiment analysis, it is also advisable to collect data through the use of APIs and scrapers.
2. Pre-processing of data. At this stage, text messages and emoticons are transformed into a mathematical form thanks to the application of various approaches. 3. Processing of data from social media. Data processing methods in sentiment analysis are divided into rule-based approaches and machine learning algorithms. AI-driven sentiment analysis is more flexible and effective for identifying the judgments of the target audience in social media, as it allows a company to quickly process large amounts of disparate information and identify hidden relationships. Fig. 6 presents machine learning algorithms for sentiment analysis.
Ensemble Methods n o i s s e r g e R c i t s i g o L 4. Analysis of the obtained results. Evaluating the effectiveness of the obtained results and determining the expediency of the model for monitoring the sentiments of the target audience in social media. In the case of determining opportunities for retraining models based on machine learning algorithms and obtaining more accurate results, return to the previous steps.
5. Monitoring in real time. Derivation of the obtained model for AI-driven sentiment analysis. The results are used on an ongoing basis to improve the company’s marketing strategy in social media. Continuous assessment of sentiment is related to the specifics of changing user behavior, first of all, we are talking about Generation Z and Alpha. These generations quickly adapt to the factors of the digital environment and actively change their behavior in social media, which requires companies to constantly monitor the mood of the target audience.
6. Improvement of approaches. The development of information technologies and the introduction of new approaches to the development of artificial intelligence encourage companies to apply advanced scientific achievements to implement sentiment analysis. More productive machine learning algorithms will allow to identification of irony, context, and features of the expression of emotions by consumers in different countries of the world, which leads to the implementation of new approaches in improving the marketing strategies of companies in social media.
The specifics of the construction of text messages by different generations of users and socio-cultural features in different countries of the world do not ensure 100% accuracy of sentiment analysis. The presented study testifies to the effectiveness of using machine learning algorithms in the identification of consumer sentiments and the use of the obtained results to optimize the marketing strategies of companies in the digital environment. The evolution of artificial intelligence will make it possible to increase the effectiveness of the use of sentiment analysis in the field of digital marketing, allowing companies to interact with users on a qualitatively new level of communication. Due to the consideration of the peculiarities of different languages and cultures, multilingual approaches will be used in sentiment analysis in the future. The process of communication in social media involves the use of various types of content, which will allow in the future to expand opportunities for a comprehensive analysis of customer sentiments by building complex models involving audio, video, and visualized materials. The use of more effective deep learning approaches (convolutional neural media, recurrent neural media, transformers, etc.) in combination with cloud technologies will not only improve the quality of identification and interpretation of the results of sentiment analysis but will also allow for realtime data processing.
Increasing competition between companies in the digital environment and the rapid reorientation of consumers to brands that offer more attractive integrated products requires companies to constantly improve their marketing strategies. To ensure long-term communications with the target audience, companies use innovative digital marketing tools and other advanced approaches that allow to achieve a multiplier effect. Achieving the set tasks involves the use of artificial intelligence to process heterogeneous information. When researching sentiments on social media, it is advisable to use AI-driven sentiment analysis. The presented approach has proven to be effective, as it allows for identifying consumer sentiment, context, and irony based on text data and emoticons with a high level of probability. Thanks to the integration of machine learning algorithms into the system of automated collection and processing of data from social media, the company gets the opportunity to identify in real time the change in the attitude of the target audience towards the company and its activity in social media. Monitoring the effectiveness of the implementation of the marketing strategy in social media allows a company to promptly adapt communication models with the target audience to the needs of consumers, which positively affects the achievement of an economically justified level of conversion and allows to ensure a high level of loyalty of the target audience.