- Product specialization;

This approach is characterized by separating the market into different segments and detecting the narrowest niche where a company may concentrate its efforts. The choice of the niche is explained by the need to allocate the product into the market with less competition where the product range is more elaborated. (Lusch et al., 2007).

In any chosen by the company strategy there are some success factors that ensure the proper performance of the company in terms of the chosen strategy. Vasconcello (1991) in the research explores three major marketing success factors that define the whole performance. These are social networks, personal selling and the exchange of tacit knowledge. According to the article, these factors are directly connected to the strategy that is most useful in game development. After thorough analysis and case study, the author concludes that the factors play a crucial role in the promotion of the game. The findings have been later bolstered by other independent research.

Three main marketing success factors:

- Social networks;

Robins (2010) explores the role of the social media in the promotion of the game within the marketing strategy. Author pointed that social media has become an integrated part of games` production for the game industry. Via this channel, game developers receive an information from customers that is needed to improve the game, strengthen their loyalty and potentially appeal to new customers according to the author. Besides, customer support is also experiencing a shift from the originally local game platforms where users used to receive any help to the social networks where they can directly contact the game developers either in personal messages or through posts. The author also concludes that although social media remains a powerful tool, it, nonetheless, is not enough to achieve the desired effect without appliance of other promotion tools.

- Personal selling;

From the standpoint of marketing perspective, another author emphasizes the necessity of another vital promotion tool. Jaworski (2018) in his recent work that mainly serves as a commentary to another work examines the lessening success factors. One of those refers to personal selling. The author in his article addresses the problem in marketing that is dominant in the current period of the field development. An analysis of the current situation on the market of marketing service gives a reader an incentive towards the beforementioned problem. As a conclusion, the author proposes a solution regarding the reversion of the problem and the means. Robins (2010) bolsters the view by applying the marketing theory on the game development industry. Even though game industry is not the one which uses the tailored approach to each customer, it relies on the opinion of customers, conducts research and generates the unique product that is able to satisfy large groups of customers. In any case, game industry is considered to be a customer-oriented industry that strives to maintain long-term relationships with its existing customers and cultivate the new ones (Jaworski, 2018).

- Exchange of tacit knowledge;

To ensure the best and successful performance of the company, there is a need to establish communication between the functional areas in the company. Arnett & Wittmann (2014) examine the role of the tacit knowledge exchange in the promotion and how it may improve the success of marketing. Through pointing out main hypotheses, the authors examine the influence of the tacit knowledge and prove the correlation between the sales and marketing. As a result, the article is concluded with the findings that tacit knowledge exchange may enhance the success of the marketing.

Another author explores tacit knowledge with the emphasis on the game development industry. Anderson (2011) states that this factor is even more crucial for the IT company that is involved in development of the games. While the communication between the customers and the company is on the content production department, is it vital to ensure that this department delivers users` comments in a proper and timely manner. Through the field research of the current game marketing campaigns, the author concludes that the communication is important as game developers should be able to think about the possible changes and communicate those changes to the main office where these changes are further implemented (Anderson, 2011).

1.5 5C of digital marketing

Founding further analysis on the previous marketing strategy theory exploration it is concluded that the most efficient framework to research video games industry is 7C framework of digital marketing as it allows to understand how developers shape the marketing strategy. Similar to other digital industries, video game development is dependent on the customers. Evaluating the ratings and reviews, companies strive to get customer insights on the games they developed (Dobele, 2015). At the same time, customer engagement heavily depends on the content (Lee & Benbasat, 2004) which focuses on telling a certain story with maximum player involvement from the description, key words, pictures and trailers in order to attract their attention and create a connection between a customer and a game.

Conversion is yet another important point to include into the analysis as it revolves around the rate the occurs when people visit the website and turn into paying customers (Dobele, 2005). Thus, it is crucial to further concentrate on the research of the games which customers have to pay for to play as they are more likely to fill in an honest review for the payed game rather that free, because they would want their money to pay for something that is reasonable and brings satisfaction.

Context implies providing relevant information for customers via appropriate channels. Eliciting information from the relevant gaming platforms will increase the probability to better understand the message that game developers send to potential players. Another influential factor in the gaming industry is the community especially if it is an online game or with co-op features. Community can play a significant role when players decide to stay loyal to the game and the developer (Vulkan, 2003).

Another reason for game developers to focus on the digitalization is convenience which signifies the opportunity to find the fastest and simplest way to deliver games to players. Cohesion, on the other hand, requires a joint conversation between customers and developers. In a way it allows to increase customer retention (Lee & Benbasat, 2004) as players want to have a unified channel where all the necessary information about the game is stored.

Marketing strategy investigation provides an overall picture of the techniques exploited in the industry; however, it does not give an insight on the specific explicit features that drive the approaches. To get a better understanding of the connection between marketing and the end product (a game), it is crucial to examine the marketing mix model of the video games industry.

1.6 4S digital marketing mix model

Since video game development industry remains a highly competitive one (there are more than 100 prominent developers; the top 10 developers can be observed in Appendix 1, Figure 2), the marketing strategy cannot create a solid basis for success within the rivals (Wang, Wang & Yao, 2005). Developers of video games try to pursue other ways that can help them to stand out and appeal to more customers. This can be analyzed with the use of marketing mix. Providing that video game development industry, which is chosen as an example of the current paper, pertain to the digital area, traditional 4P analysis is not suitable for the purposes of the research in evaluation of the market situation in e-commerce. Intangibility of the product that is being offered by the digital companies opens the way to use the 4S web marketing mix. This mix helps to understand the situation on the market better that is applicable to the digital product and the quality that is created (Wang, Wang & Yao, 2005).

Constantinides (2002) explains digital marketing mix as Scope, Site, Synergy, and System.

- Scope, which is the objective and strategy, refers to the market segments that include several variables such as geographic or behavioral, potential customers that are assessed by their needs, profiles or motivation, internal analysis conducted in terms of the value process and resources, and strategic role of web activities.

- Site stands for either promotion activity or online experience of the certain digital product produced by the organization that operates online. These activities may include branding or attracting traffic.

- Synergy, as a third pillar of the mix, is an integration. It refers to the e-commerce implementation into the company`s basic operations in order to maintain an equilibrium and in attempt to improve company`s processes of activity.

- System is related to the technological issues and the maintenance of the product in a proper condition in order to make it operational and practical for the customer needs.

Taking this model as a basis, some authors (Tomaselli, 2008; Kuo, Wu & Deng, 2009) who study the industry of video game development have acknowledged the key trend that becomes crushing with time. By implementing a differentiation strategy that aims at keeping an equilibrium between extensive advertising and the supreme product quality, majority of the present developers move the focus largely to the promotion and advertising (Tomaselli, 2008). As a result, the hole between the perceived quality of a video game and the expected one is evolving. To estimate the influence of the promotion practices on players' picks it is crucial to investigate prevailing promotion success factors.

1.7 Video games' promotion success factors

The key to success in game development industry lies in creating a product of one of a kind. The developer strives to enhance its quality and appeal to more possible consumers. In case of the failure to meet the expectations of the customers, it may lead to their switch to other games produced by other developers (Oliver, Ash, Woodley, Shade & Kim, 2014). Nonetheless, before players can estimate the real quality of the game, they are influenced be the marketing conducted by the producer company.

Many studies have investigated the most crucial characteristics of the promotion activity so that to establish the prerequisites of the marketing campaign success and causes of their efficiency drop (Ramaseshan, Ishak & Kingshott, 2013; Oliver et al., 2014; Wang & Nordmark, 2015). The following list demonstrates the factors that remain crucial in online marketing campaigns.

Acceptance. Although promotion may become an overwhelming activity which is able to reach its customer by any means, it does not necessarily emphasize the fact of its acceptance by the target audience. To stipulate the campaign`s efficiency, the promotion activity has to be viewed by the target audience and has to be also accepted as a source of observing the information displayed in the promotion tool body (Yousif, 2012).

Complementarity. The effort put into the promotion cannot be concentrated completely on one media instrument, but a wide range of tools to enhance its effectiveness. Utilization of all available means, on the contrary, leads to no more efficiency of the campaign. The solution lies in the proper sense of balance of the best means that are central to a certain industry (Barnes, 2002).

Content. This factor relates to the way of the message communication. This includes the language, the length and the idea lying beneath the entire message and its delivery to the target audience. Thus, Haghirian et al. (2005) stresses that the message that is delivered to the target audience should be more like instructive, entertaining, and relevant (Dickinger, & Haghirian, 2010).

Cost. Apart from the mentioned aspects, the financial side remains no less important. Justified costs should correspond proper marketing campaign. If the resources, on the contrary, prevail the received result of the campaign, it is doomed as financially inefficient. Therefore, to reap the most benefit, the cost of the campaign cannot exceed the accepted terms (Barnes, & Scornavacca, 2004).

Credibility. Credibility has more long-term effect that is created for the product in terms of the campaign effectiveness. Communicated to the customer message should exclude vague and initially wrong information that can possibly confuse the customer. The factor also affects the attitude of the customer generated on the basis of the message truthfulness (Haghirian et al, 2005).

Permission is relevant in terms of the desire of the users to control the amounts of advertisements targeted on them and are constantly trying to find the means to get the information they are interested in. Acquiring undesired advertisements rises dissatisfaction if the users towards online-platforms (Fransen, Verlegh, Kirmani & Smit, 2015).

Frequency grants an opportunity to calculate the number of advertising messages received per one day. The updates on the newly developed games or discounts are usually sent to the players by video games distribution platforms via e-mails. Because of the facts that users can read the messages at any convenient for them time, this kind of advertising is considered unobtrusive. Nevertheless, some platforms fail to identify the appropriate frequency of the e-mails which leads to intrusive advertising unwanted by the customers.

Personalization. For gaming platforms, it its common to allow players to customize their profile so that to enhance user experience. This also lets gamers get more personalized recommendations in the future. Because of personalization the information that customers get becomes more suited for the customer and, hence, more valuable, which as the result, augment the impact on the consumers' choice and boosts customer retention (Pfeiffer & Zinnbauer, 2010). Reach shows the number of customers that can be influenced by the chosen marketing campaign. It is quite important as the main sources of income for companies in game development industry are online users (Fransen, Verlegh, Kirmani & Smit, 2015).

Incentive. With the direct marketing exploitation users can get various benefits (such as price discounts) which are beneficial for the customers (Aslam & Karjaluoto, 2017). In the recent research it was discovered that the boost of the sales can be reached by offering price discounts to the customers, which is largely implemented by different digital distribution platforms with their season passes or special packs of games (Arnosti, Beck & Milgrom, 2016).

To conclude, the investigations conducted showed that many authors emphasize different success factors of digital product marketing campaigns. Nevertheless, none of the research strives to identify most influential factors. In the game development industry where product quality can be easily concealed by the means of advertisement, it becomes vital to distinguish those means. As entertainment has been defined as a source of pleasurable experiences that holds emotional value for the customers (Oliver et. al., 2014), the quality of the entertaining product stays a crucial factor. Thus, video games that possess mentioned emotional value must be taken into consideration in line with the product quality while determining the promotion success factors.

1.8 Video game design and principles of usability

Video games success depends not only on the development process and promotion techniques exploited by the companies, but also on the perception of the game by the player and how comfortable the player is while playing. This can be estimated by the usability phenomenon, which is also important for current research as it will allow to give definitions to the promotion factors identified when researching the reviews of the games.

The most common method to evaluate the potential problems that arise from using a productivity software is with usability characteristics (Gifford, 2013). They imply several categories of issues that the software possesses along with the special ratings to assess the extent of each problem (Livingston, Mandryk & Stanley, 2010). Several studies focused on identifying characteristics applicable for video games in order to observe their usability (such as Desurvire, Caplan & Toth, 2004; Federoff, 2002). Federoff's (2002) investigation unraveled that the central characteristic of a video game should be a set of components that allow to evaluate it fun, as the goal of a video game as a product is to provide entertainment and bring satisfaction for the player. However, in the research conducted by Bogost (2008) the author states that fun cannot be the ultimate estimator of the game's usability as it denominates the subjectivity and does not provide the impartial attribution instrument.

Another research on the video game features that best estimate usability was conducted by Pinelle, Wong and Stach (2008) which simplified the concepts discovered by previous research. As games involve almost constant interaction and input form the player, the authors identify principles of usability that concentrate on user experience and on how the design of the game can facilitate or obstruct it. By investigating the reviews of games of different genres, the authors identified 285 problems with usability of games and classified them into 10 usability characteristics with examples, all of which can be observed in the table below.

Table 1 Video game design usability principles*

*Source: Pinelle, Wong and Stach. Heuristic evaluations for games: Usability principles for video game design. (2008, p. 1458)

1.9 Remarks on the ethical marketing issue

The usage of deceptive techniques is not a new phenomenon in the marketing and promotion field. Companies have practiced different methods of concealing the real quality of the product or service they offer. From unfair pricing and deceptive packaging, it all came to the misleading advertising which rose a question of ethics.

Some scholars define ethics as individual values placed behind someone`s decisions in marketing (Ferrell, 2005). Others, on the other hand, refer ethics to the organization`s values and codes that drive their ethical decisions into the field of “societal marketing” (Kotler, 2008). “Societal marketing” means the correspondence of the interests of consumers and the society that is created by ethical considerations of the organization. In other words, it is not just profit that a business should be concerned with, but the morality of their actions that should be transparent and trustworthy.

Today misleading advertising has become almost an inseparable part of the marketing campaign of many companies. What is more, with the problem being acknowledged, customers are now even more confused as they cannot detect which campaign has implications of deceptions and which does not. Video game development industry is not an exception. With rising unsatisfaction of the gamers, it becomes evident that the problem has not gone but intensified. The presence of dozens of companies in the market and only eight of them occupying the market makes the competition unfair providing that the giants employ unaccepted from the standpoint of ethics means. The situation on the market prompts questions on whether companies are allowed to use deceptive marketing techniques and whether it can harm the industry overall with the subsequent unsatisfaction of players' needs (Garda & Grabarczyk, 2016).

Within the ethical marketing issue there is also the problem of fair competition in terms of the rivalry between businesses (Hill & Alexander, 2017). In the video game market, the most disturbing concerns rise when looking at the conditions in which two types of business operate - the independent developers and famous companies. The independent developers are often left in the dark as they do not possess the necessary budget for extensive promotion (Wright, 2018). At the same time, these developers are ready to offer revolutionary techniques in terms of the game elaboration. Thus, the industry is operating in the state of unfair competition which eliminates the opportunity for the independent developers to enlarge their customer base (Garda & Grabarczyk, 2016). Sometimes players can even ignore the unsatisfactory perceived quality of the products released earlier by the famous companies and still purchase the newest games as they are promised the most pleasurable experiences.

All in all, the use of the deceptive marketing techniques can cause harm to the industry overall as the players would leave in search of the other entertainment sources. None of the research prior to this one has concentrated on identifying the techniques with which companies achieve the effect of customer enticement. That is why it is crucial to attract more attention to this issue in order to establish fair conditions on the market and prevent customer dissatisfaction elevation. By focusing on distinguishing the factors exploited in the promotion part of marketing campaigns that invoke the desire to buy the product without the regard of the real quality of the product, it will be possible to achieve the mentioned goal. The next sections will be dedicated to the analysis of feasibility of using reviews for the research and methodology that will grant an opportunity to elicit said factors relevant for current paper.

2. Statement of the research question

In an attempt to compete with other rivals on the market, companies turn their efforts to promotion activity more than to the proper quality elaboration of their products or services (Oliver, Ash, Woodley, Shade & Kim, 2014). This becomes even more relevant for the digital industry with the intangible product being offered as the customers can weight their decision to buy a product only by the description of properties that the company offers. Such descriptions are distributed in an appealing form as every business strives to attract more customers and, hence, earn more money. To create this appealing outlook, companies use marketing techniques though various promotion channels. In the literature that studies marketing there is not much said in regard to the most influential means that appeal to customers in a digital industry, although the best practices to attract customers attention are already discovered for traditional markets. This becomes even more relevant for the video game development industry in particular. A wild race over customers' attention has brought nothing but disappointment among the players who still remain fans of the major companies due to the reputation and credit that they gained once entered the market in its infancy (Kasurinen, Maglyas & Smolander, 2014). On the one hand, the whole situation enhances a dissatisfaction of the customers while deepening the gap between the expected quality and the perceived one. The expected quality is the anticipation of the customers created by the marketing campaign and promotion. The perceived quality is, on the contrary, the real quality experienced while utilization of the product or service (Eckert & Hughes, 2010). The expected-perceived quality match is crucial in creating the customer value as it fulfils the company`s goals and mission. This fulfillment is important for company`s operations, its clear vision and transparent activity in the future that should lead to proper decisions and long-term prosperity.

On the other hand, the situation is not that propitious for small companies which are entering the market nowadays with fresh ideas and high-quality games but limited financial support except for their own funds or crowdfunding activity to promote their products (Aleem et. al., 2016). The intense marketing campaigns that hide the real game quality conducted by the giants on the markets leave those small independent companies without deserved recognition (Kasurinen et al., 2014). Besides, this kind of situation in the market rises the question of unfair competition and the importance of ethics. Because of the large market share occupied by the famous developers, the independent developers are mainly overshadowed by their presence and almost do not have an opportunity to compete with them (Aktas, & Orcun, 2016).

The stated problem has raised concerns in past years. The latest trend has proved the gradual decrease in the game quality of the mass-marketed games during the past two years. At that time, statistics shows that the whole market is predominantly occupied by only eight famous companies that produce most of the games today. Since the numbers of famous companies is far greater than eight, this means that the rest of the market is occupied together with the other mass-marketed game developers and independent indie game developers leaving the latter the tiny market share alone. Hence it is important to check whether the difference between the quality of the games produced be different types of developers is statistically significant.

The usage of deceptive techniques by the famous companies creates unfair competition on the market that lead to the problem of ethics. It is not only the companies or the individuals who do not have an opportunity to compete equally, but also customers who are deceived in their decision to purchase a certain game. Any market for a customer should be open and transparent so that they know what they are purchasing in the first place (Wright, 2018). The recent evidence shows unsatisfactory trends that marketing campaigns take significant place at the operations of major companies (Kadic-Maglajlic, Boso & Micevski, 2018). Therefore, there is a need to understand what makes the campaign effective, what are the factors that lead this campaign to success and how independent companies can benefit from this knowledge and the utilization of the results of this research.

To do that, the researchers of this paper have set the following research question: what are the key promotion success factors that are exploited to conceal the quality of video games? And which have the highest impact of customers' purchasing decision?

On this basis, there are four objectives set for the researchers. First, there is a need to identify the type of promotion strategy which is commonly used in the digital industry and video game development in particular. To achieve this objective, the first task is established and completed in the theoretical foundation of the research. Second, the preferences of each of the developer in choice of the method of promotion are to be detected for the applied part of the research. Third, as a result of the reaching the second objective, the next one would be to detect the promotion success factors and track the most influential ones. Fourth, there is a necessity to track a relationship, if prevalent, between the response coming from the market and promotional methods used as well as between the real game quality and the level of customer satisfaction. In order to answer the established research question and reach the objectives set before there is a set of tasks needed to be done:

a) To build profound theoretical foundation. The foremost thing to do in the beginning of the research is to understand how the customers make a decision to purchase a digital product. Then it is important to unravel how video game development industry functions and what marketing methods, strategy and techniques are applicable for the companies operating in that industry. In addition to that, there is a necessity to pay attention to the promotion component of the whole marketing campaign, and investigate what components influence the success most and what factors are to use in the promotion activity. The analysis should also concern reviewing as the source of relevant information about customers' decisions.

b) To decide on the methodology suitable for the purposes of the research. Since the researchers have decided that the best way to find the factors would be the topic modelling of the game description and the reviews, there is a need to find the most appropriate method out of the existing ones. Therefore, the first thing to do is to study the existing methods, find the most appropriate one and explain the choice.

c) To collect data. Another crucial step in achieving the goal of the research is to create a list of games released in the chosen period (2016-2018), collect the description of each game, reviews, and other crucial information for data analysis necessary to conduct statistical tests.

d) To detect the promotion success factors. Data analysis and tests, which the researchers conduct, are necessary to evaluate the differences in the approaches of the famous companies and the independent ones. These are needed to evaluate the discrepancies in the approaches and track the factors that both types of developers employ.

e) To allocate the most influential factors and create a list of recommendations. The strength of influence can be detected with factor analysis. By distinguishing the strongest factors, the researchers will be able to give recommendations to the independent indie developers on the matter of improvement the marketing campaign and promotion activity. The said list can be also exploited for other digital products that encounter similar problem on their markets.

In order to find the factors that spur the success of the marketing campaign of the game, there is a necessity to analyze the way the game is delivered to the customer (game description) and the response from the customers (reviews). The differences between the implications of the game description and the reviews are needed in order to identify the gap in the expected and perceived quality. By proving that indie games get better market response in comparison to mass-marketed games, the ground for further research is established.

Currently, in academia it is common to use reviews to draw on the customer opinion about the product. However, it is important to delve deeply into this matter to understand whether reviews really are a feasible source of information that allows to determine the reasoning behind customers' purchasing decision, and of digital product in particular.

The analysis of reviews and game descriptions of both parties - mass-marketed game developers and independent game developers - is conducted in order to track the preferences of each developer in the choice of the approach and to identify the factors that lead the marketing campaign to the desired success.

Video games have been a focus in many studies dedicated to a variety of topics with the research method implying content analysis as well. Nonetheless, the analysis of game reviews remains a relatively new method in content analysis. For instance, Ivory (2006) applied reviews analysis of video games gathered from the GameSpot website in order to measure the representation of the gender in video games. Although the researcher confirmed set hypotheses, he did not pay a deliberate attention to the review analysis in detail. Likewise, Soukup (2007) in his study aimed to learn about the community mentality of video gamers on the basis of the reviews rather than detailed analysis of the reviews.

The advent of the Internet has brought a variety of websites that rate and evaluate games in the numeric values and provide consumer reviews. Among many there are Metacritic (Metacritic.com) and Game Rankings (gamerankings.com) that accumulate a collection of comprehensive reviews and scores for each game (Pearce, 2004). The scores that each game receives is largely used as a method to evaluate the game. These scores also serve as a basis for creation of the expected quality of the game for the customers, therefore, they are given an important part in the game evaluation. Content analysis of reviews, on the contrary, is left aside and not commonly used in the game evaluation and measurement of its quality (Schmierbach, 2009). Nonetheless, the observations prove that video game reviews possess an influence and power that cannot be underestimated. Even though Boatwright, Basuroy and Kamakura (2007) state that the consumers are not able to estimate the product quality before its consumption, they, nevertheless, agree on the point that the reviews help consumers create a picture of the product and anticipate the quality they are only aware of.

The crucial point to consider is that the reviews come from two different parties - the customers who purchased the game and opinion leaders. While the latter carries more credit and is better accepted as unbiased evaluation, the former is hardly given an appreciation. The greatest problem of the reviews left by the customers is the prevalent bias formed in a shape of personal feelings and emotions. Because of that the evaluation of the reviews (as positive or negative) cannot be perceived as impartial. In order to eliminate the bias, the analysis should be concentrated on eliciting the topics that a single review is referring to. By processing large quantities of data and eschewing the words that possess the sentimental meaning, the bias can be eliminated.

Providing that this research focuses on detecting the success factors in the game description and the reviews of the customers, the content analysis is primarily applied.

A variety of content analysis methods allows the researchers to choose the most suitable for the study purposes one. For automated methods there are computer-aided text analysis methods which split into supervised and unsupervised that differ in their dependence on the researcher. Those methods possess certain advantages and drawbacks while processing the large quantities of text data which are determined in the theoretical overview of the current paper. For the purposes of the current research there is a need to concentrate on unsupervised method as it allows to distinguish the meaning of the allocated results. The best suitable method, therefore, is LDA.

Latent Dirichlet Allocation is an unsupervised method useful in splitting the text into topics and allocation of the certain words to those topics based on the established by the model pattern (Neuendorf, 2002). The choice of the approach was mainly stipulated by the fact that it allows the researcher to define the topics manually in contrast to the other well-known approach, STM, with the automated corpus summary. In more detail, the method separately processes the reviews and the game descriptions and finds groups of words that are suitable for a certain topic. Found topics are defined by the researchers in accordance with the factors and, then, topics in reviews and topics in game description are compared. Since each topic is defined after the factor, there, then, the trends are established on the matter of the dominance of each topic. The analysis is applied to both the mass-marketed game developers and the independent indie games developers. The received results are compared and the most influential factors that make the mass-marketed games successful are tracked.

As for the results of the research, the most influential factors that stipulate the success of the campaign are identified by using multinomial logistic regression analysis and on that basis the list most influential factors is allocated and a set of recommendations for independent indie game developers is created. This set is designed in order to help independent developers not only to improve their marketing campaigns in the short term but also to gain better appeal on the market and increase the awareness of them as of the individuals who are in the industry to produce high-quality content for their target audiences in the long term. Lack of financial support of the developers cannot be a hurdle in giving the customer the best product they deserve. What is more, by establishing equal terms and conditions for all competitors, the competition can become fair and transparent.

3. Methodology

3.1 Content analysis methods

Content analysis is not a new phenomenon in the research world. With the appearance of paper documentation which is needed to be stored in an accessible way, it becomes vital to interpret the large amounts of text data in the most convenient and understandable way (Lowe & Benoit, 2013). Thanks to the digitalization that is gradually taking over the entire world, all of the archives, records and any public documents as well as current Internet activity in a form of surveys, reviews and feedback are turned into the digital product that can be analyzed and interpreted by a variety of means. Those means usually split into manual and automated ones. For the purposes of this research, automated methods will be given the most attention (Tetlock, Saartsechansky & Macskassy, 2008).

The core of the automated methods of analysis lies in the treating the text data as a single vector the length of which equals the number of words in the text that are unique since those methods count words regardless its order. As a result, those algorithms produce large quantities of unique words that cannot be practically comprehensive (Krippendoroff, 2004). However, there might be a possibility to significantly reduce the amount of words by eliminating infrequently occurring, which may also have a detrimental effect for a text interpretation. Current tools of processing the data such as stop word removal, compounding and decompounding and many others allow to amend these and many other flaws. These tools are vital to review for better understanding of the textual data processing and its preparation for further automated analysis (Pollach, 2012).

Removal of stop words. To enhance the quality of analysis, the first thing what analysts do is stop words removal that are unrelated but still are very common for the textual data. The words like “but”, “and”, “the” or “also” are eliminated before the analysis to prepare the text. However, at this moment, the researcher has to identify what words present to be frequently occurring “stop words” and what words remain an important part of the language that can affect the whole analysis in the end (Vagnani, 2015). In the current research paper, however, the authors are exploiting already compiled dictionaries (packages) of stop words that are provided in the open sources by RStudio developers and Python programmers.

Stemming and lemmatization. Stemming implies removal of the word ending such as in plural nouns and conjugated verbs. Lemmatization, in turn, identifies the root of the word and, instead of removing any “unnecessary” part, it groups all related words and changes their forms into infinitive if needed for better comprehension (noun “thought” will stay as it is; verb “thought” will be turned into the infinitive “think”) (Taddy, 2013).

Compounding-decompounding. For some languages it is common to represent a certain meaning in two words which separately convey two different concepts. In case those words are separated and treated individually, there is a chance of getting a distorted picture. To lower the probability of distortion, it is important for the researcher to decide whether it is crucial to compound two words into a single entity or decompound into two separate concepts (Krippendoroff, 2004).

After completing the preprocessing part, the next step for a researcher is to remain building a document-term matrix or DTM. In simple words, this matrix is a table with rows, which represent a text or a document, and columns consisting of a single word in each. In the intersection of single row and a column there is a number that stands for the number of appearances of the certain word in the certain text regardless its position in the text. Once this preprocessing stage is completed, the researcher applies an automated approach to the prepared text (Short et al., 2010).

3.2 Computer-aided text analysis

Two approaches of analysis of automated text currently exist: supervised and unsupervised methods. The main difference between those two approaches lies in the human effort needed to analyze the text. Supervised methods require a manual categorization of the first documents which are later used by a model as a basis for the categorization of the remaining documents (Pollach, 2012). Unsupervised methods are not supposed to build their results on the example provided by the researcher. Instead, a model allocates the text into the categories by searching for a low-dimensional summary that explains the way words are systemized. In the end, the researcher is needed for the result interpretation rather than giving the instruction beforehand (Short et al., 2010).

Grimmer and Stewart (2013) have described two vital supervised methods - document scaling and classification and prediction.

Supervised methods

Document scaling. The most primitive method implies the necessity to count all interested words manually. By reaching the quantity of the word needed to be fully represented, the researcher achieves the goal. An example may be an occurrence of a certain word in newspapers throughout the set period. An increase in quantity signifies a certain trend such as the usage of the word “moderate” instead of the “downturn” in economic periodicals may signify the stabilization of the economic situation in the region. In addition to simple word counting, researches also assign weight to a word, most commonly for words possessing emotional implication (so that “love” get higher score rather than “like”) (Salton & Buckley, 1988).

The method is beneficial in identifying certain words and being able to tell a story by the rate of occurrence in a certain document. The advantage here goes to the simplicity and the absence of the necessity to read the whole document.

Classification and prediction. This method remains one of the most common in categorizing the sets in the document. By creating a training set and teaching a model parameters that are important in directing words into certain categories, the model is able to perform the similar activity on the actual dataset (Hillard, Purpura & Wilkerson, 2008). The simple illustration of this method is commonly used in categorizing the mail. When the model is trained, it is able to direct emails to the folders they belong to - presence of the words “price”, “limited' and “sell” will likely signify its belongness to the folder “Offers” while “university”, “tuition” and “scholarship” to the folder “Education”.

The major advantage of the method remains the ease in evaluating the results manually which ensures more accuracy. Besides, it is also easy to assess the model`s predictions calibration and modify it by making more precise (Short et al., 2010).

Despite the ease of handling the results and the process of the beforementioned methods, it is still difficult and time-consuming for a researcher to apply these methods as the prediction the models yield need to be calibrated, or, in other words, checked for the validity of results. That is why for the purposes of this research the authors have chosen to use on of the unsupervised methods (Vagnani, 2015).

Unsupervised methods

In comparison with supervised methods, which basic sense lies in replicating small-scaled work that can be conducted by people onto a larger amount of textual data, unsupervised methods imply structuring texts and interpreting the gathered results (Taddy, 2013).

The basic algorithm that runs under unsupervised learning can be described in a generative model, where 100 points are allocated to two dimensions or, hence, divided into two groups. The model assigns the means for each of the groups and locates the points to a certain group based on the probability with which the point pertains to which dimension (Figure 1).

Using the statistics, it is possible then to interpret each group in the context of the corpus (Krippendorff, 2004).

Keywords and scaling. When implementing keywords method, it is most common for an analyst to apply manual approach on selecting key words from each document. When the population is validly represented in a sample, the model counts the chosen words appearance in a document. After that each word is given a weighted frequency which represents the scaling of the word order and relationships of the words between each other and groups where they occur. One of the significant drawbacks of this method is the fact that a word is considered out of context in each document which can distort further interpretation (Pollach, 2012).

Figure 1. 100 points allocated to clusters in the two-dimensional space*

* Based on the source: Rabinovich, M., & Blei, D. (2014, January). The inverse regression topic model. In International Conference on Machine Learning, 199-207.

Document clustering. With the document clustering method application, the researcher is able to assign each document to a specific group, which subsequently explains the use of the word. The documents can form clusters and hierarchies or be individual representing a single meaning. This approach can be restrictive when it comes to analyzing compound documents as distributing a document to a single topic can convolute the meaning (Hillard, Purpura & Wilkerson, 2008).

Latent Dirichlet Allocation. Unlike document clustering, Latent Dirichlet Allocation (LDA) implies mixed-membership approach which means that a word can pertain to one topic, and a document can be allocated to a mix of topics. The main advantage of this method is that the topics are assigned not based on the researcher's analysis but automatically on the basis of word co-occurrence patterns and the probability to meet each word in each topic (Short et al., 2010).

Structural Topic Modelling. When the research involves comparing the thematic nature of the text presented for analysis, it is best to utilize Structural Topic Modelling (or STM). The method allows to determine the distributions of words inside of the documents and to create a summary of the corpus content divided in topics. One of the advantages of STM is the opportunity to avoid pre-specification of the topics that the algorithm must look for. On the contrary, it looks for the topics itself based on the content of the textual data and the pattern of co-occurring words. Moreover, STM builds relationships between the topics found in the corpus and the metadata connected to the textual information (Lowe & Benoit, 2013). Metadata may include any relevant to the document information, for example, for the review it will be the time of the review, the name of the user, some characteristics of the user and so on. The opportunity to connect the textual data and user-characteristics brings new light for the analysis of the textual data in terms of is interpretation which is of extreme value for understanding the relationships within the dataset (Tetlock, Saartsechansky & Macskassy, 2008).