Factors determining consumers’ perception of advertising messages in social networks

Nowadays the content analysis is used for decades to examine advertisements, media content and web sites [Kassarjian, 1977; Roznowski, 2003; Yun et al., 2008]. Moreover, it is applied in anthropology, sociology, human resource management, psychology, literature science, history and philosophy. [Manekin, 1993] Various text types like mass media messages, media reports and statements of political figures, party programs, legal acts, advertising and propaganda materials, historical sources, literary works can be analysed with the help of content analysis.

This method was used in the research of Smith et al. (2012) dedicated to the user-generated content in Facebook and Youtube. The data were analysed by coding and use of frequency tables and Chi-square tests. Nowadays, it is supposed to be the most common method, which, to the great extent, does not require the direct access to the company. Suchkova (2016) in her research devoted to the graphic features' influence on the perception of advertising message also used this method.

There are two types of content analysis: quantitative and qualitative.

Quantitative content analysis is based on the research of words, topics and messages themselves, focusing the attention of the researcher on the content of the message. Therefore, the sense of the elements which were chosen for the analysis should be predicted as well as the every possible result should be determined relatively to the expectations of the researcher. It means that the researcher should create some kind of a glossary where every observation will get a definition and will be classified. The problem is that a researcher must predict not only the mentions which he can get but the elements of their contextual use with a detailed system of evaluating every notice. This goal is usually solved by a pilotage of the analysed totality of messages (it can be done by revealing the messages of the high probability key type mentions on the data of little sample) with a match of arbitration assessments of contexts and ways of using terms. It is preferable to access the notices of several scholars. The harder objective is to assign to key characteristics the particular marks - when we should decide whether the mention is done in a positive or a negative meaning and when we should range the mentions relatively to the strength of their marks (e.g., from the most positive to the least positive). Sometimes researchers use the statements of arbitration group (experts) about the intensity of a term. Often SEO is used in quantitative content analysis [Mannheim & Rich, 1997].

Qualitative content analysis is not connected with concrete sciences; therefore, it has less rules to apply and less risk of confusion in philosophical concepts. During qualitative content analysis the data are coded by the researchers (not automatically); the goal of the analysis is to determine text fragments which match the author's ideas and reflect key ideas through qualitative codes. The core problem is to maintain the rigor and credibility to get valid results. During qualitative content analysis researcher does not count and test statistical significance - he substantiates his conclusions by quotes from text. The researcher can use the same terms as in the quantitative research, whereas in other qualitative methods it is forbidden. [Long & Johnson, 2000]

Bergkvist & Melander (2000) think that the effectiveness of online advertisement should be measured based on the goals of the advertiser:

ѕ To attract visitors to a website(advertisement must be measured by the ability to respond in a desirable way);

ѕ To make consumers become aware of brand (advertisement should be evaluated by ad-related and brand-related responses).

It is also possible to unite these methods if a goal consists of two kinds of responses. In some studies concerning the effectiveness of Internet advertising authors consider that it should be measured by clicks leading to a company web-sites, users' actions on websites and sales [Hofacker & Murphy, 1998; Hoffman & Novak, 1996]. Pelsmacker et al. (2004) also concern the topic of online advertisement effectiveness. They describe the following metrics for measuring the studied phenomena:

ѕ Total (graphical and textual) impression, click-through and advertisement transfer. (for banners and web advertisement);

ѕ Conversion and response rate (for E-couponing, E-sampling and mobile marketing);

ѕ Sent mails, click-through rates, unsubscribe rates, conversion rate, click-stream analyses, E-mail pass along rates and coupon codes redeemed (for E-mails).

Most of the revealed metrics are related to the communicative effectiveness, not commercial, therefore, they can be used as metrics measuring perception. According to the classification by Pelsmacker et al. (2004), the metrics are also used nowadays, thus, they should be revealed. Online advertisement campaigns can be evaluated through various models. They are divided into 3 divisions: impression models (FF, CPM), effectiveness models (CPC, CPA, CPS, CPV, CPL and others) and mixed models - CPM+CPC, CPM+CPS. [Kryсski, 2014; Guziur, 2011]

Impression models

The oldest and simplest model is FF (Flat Fee) where cost of marketing campaign depends on 2 factors: advertisement size and impression time. Advertiser pays for time of advertising showings despite of their amount. This model is usually used during campaigns devoted to image and brand creation as well as for promoting concrete product. Payments are beneficial for publishers because they get fixed amount of money for period independent on generated traffic [Gancarz-Wуjcicka, 2013].

The second impression model is CPM (Cost Per Mille) or CPT (Cost Per Thousand). It is based on CPM index which determines cost of the advertiser with the goal of reaching 1000 recipients. The recipients' actions after impression are not taken into consideration. Like FF model, it is used during campaigns creating brand and image where the accent is made on the brand's memorability. The use of payment in this model is the most beneficial from the point of view of advertising contractors, as they receive payments or a real number of advertising broadcasts, regardless of the impact of advertising on the recipients (for example, sales). This form of calculation often uses the advertising limit for this user, the so-called camping model.

Effectiveness models

The CPC (Cost Per Click) is one of the most popular models, measuring effectiveness of advertisement. It bases on the index of CPC - or Cost Per Click. Money is paid by advertiser only for actual clicks on the advertisement. [Guziur, 2011] The evaluation of CPC effectiveness occurs through CTR index. According to Torop (2006), CTR (Click Through Rate) is a relation of the total number of users' clicks on advertising messages, banners, teasers or textual links to the ad impression. It is measured in percents.This model is usually used by affiliate and contextual advertising networks. From the websites' owners view providing campaigns is not as gainful as CPM or FF because the amount of clicks is influenced by advertising's quality [Kryсski, 2014; Rzemieniak, 2015]. However, the goal of this paper is indirectly aimed at increasing advertising quality.

Other model measuring effectiveness is CPA (Cost Per Action). Advertiser pays for advertisement when user makes certain actions (enter website, fulfil order form or give contact data). The rates are determined according to the users' actions or sales' percentage. The model is usually used during marketing events the main goal of which is selling a product. This model is not popular among those which are used during images' creation. [Kryсski, 2014]

To evaluate effectiveness of CPA model, CR (Conversion Model) can be use to define what percentage of people watched advertisement will do concrete actions. Model of measuring CPA has various forms as CPS, CPL, CPE, CPO and CPV. These models differ according to the action type for which the advertiser pays. For instance, in CPS (Cost Per Sale) he pays for sale made after the advertisement's influence; CPL(Cost Per Lead) includes fees for users' contact data provided after fulfilling a special form attached to the advertisement; CPV(Cost Per View) is the model where advertiser pays for every impression (usually used for video advertisement). [Kryсski, 2014]

Hybrid models

The last group of methods is mixed (hybrid) models which unite several models (CPM + CPC, CPM + CPL, CPM + CPS). They are becoming more popular because they let the publisher and advertiser have benefits both.

Finally, the ROI (Return On Investment) can be useful during measuring effectiveness in Internet advertisement. This measure is relatively unpopular among middle or large advertisers. [Kryсski, 2014]

Marphitsin (2017) determines 7 terms which every SMM- specialist should know. They are the following:

ѕ Reach - the amount of people who could see the post;

ѕ Engagement - the sum of communications between subscribers and posts. Often the index of ER (Engagement Rate) is used which is accounted by relation of users' actions to reach. Among actions are usually likes, comments and reposts;

ѕ Views count or Impressions - the total number of views. If reach counts the number of users who saw the publication, impressions - the amount of all views (user can watch post twice);

ѕ CPA -- price per action;

ѕ CPC -- price per click;

ѕ CTA -- call to action;

ѕ ROI -- the return on investments.

According to the metrics which can characterize perception, Anishenko (2015) have revealed that activism of users in social media is often restricted by the likes of messages, videos and photos. Kirakosova (2014) in her research used term successfulness of message promotion as a number of received responses on a message from users. This concept was characterized by 2 metrics: number of reposts which reveals message spread through other users and number of likes which indicates message's popularity. Moreover, likes are an indicator that a person enjoys or approves something, whereas reposts are not just a way of message spread but public admission of a mostly positive perception.

According to the research papers of Bruner (2009), Godfroi,(1996) , Pelsmacker et al. (2004), Rzemieniak (2015), Anishenko (2015), Kirakosova (2014) and Marphitsin (2017) core metrics of perception have been determined. Moreover, according to the data which the selected company uses, they were complemented. The list is the following: CPM, CTR, eCPC, likes, reposts, comments, total reach, reach among subscribers, group clicks, users joined the group, link clicks, complaints, hidden posts and all posts hidden.

To summarise the literature review, social network as a part of social media is a strong and fastly growing and developing channel of marketing communications. The most popular social network in Russia is Vk.com. Therefore, it has been chosen as a platform for analysis of advertising messages.

Advertising message is a significant element of advertising communication because it does not simply inform customer but affects his perception and even is used for loyalty building. Therefore, the quality of advertising message should be controlled and developed. Several authors have researched advertising messages in social networks. Kwock & Yu., (2013) have revealed that advertising messages' popularity do not depend on their content but only on media type. However, Kirakosova (2014) in her research paper determined that media type does not influence users' reaction. Nowadays to attract people's attention in a mass media mess and to bring important information to the audience users should read a message and then make some conclusion about a community. Ryzhkova (2010) in her research paper claimed that the numbers of symbols do not matter, but if the text and its content are interesting, people will read a message. Therefore, we supposed that the role of content was undervalued.

Two characteristics of advertising message have been chosen for future analysis: media type and content. The first one is revealed in several metrics: text, photo, picture (art, drawing), gif, info graphics,meme, link to a website, video, audio, which were formulated on the research by several authors [Kwock & Yu, 2013; Kirakosova, 2014; Asur et al., 2012; Uhova, 2012; Cook, 1996; Downing, 2000].

The content will be evaluated by characteristics of visual and textual parts of messages and by content type according to the topic and message's structure. The metrics for evaluating textual part were formulated, according to the various researches [Ryzhkova, 2010; Gordon & Gordon, 2012; Suchkova, 2016; Vasilevich & Ledeneva, 2005; Shardakov, 2016; Anokhin, 2017].

They are the following:

ѕ Number of signs

ѕ Number of words

ѕ Nausea

ѕ Water content

ѕ Presence of a title

Several metrics for evaluating visual elements in messages will be applied: colors (colorful or black and white; warm or cold dominated colors), presence of corporate style (for eg. logotype, fonts) and presence of a human and nature. [Suchkova, 2016; Knabe, 2006; Lohse & Wu, 2001]

Content types were divided into 2 groups:

ѕ Metrics describing post's topic and content type: promotional, competitions, entertaining, learning, news, ugc, interactive [Anishenko, 2015; Smith et al., 2012; Specialists WFA, Millward Brown, & Dynamic Logyc ,2011]

ѕ Metrics revealing the message's structure: advertising message, advertising treatment and call to action. [Uhova ,2012]

Perception and the understanding of a text by a recipient is a base for text's effectiveness, because consumer, firstly, read an advertising message and then makes conclusions about a product. Therefore, metrics for evaluating communicative (not commercial) effectiveness can be used for measuring perception. During the analysis of the literature [Bruner, 2009; Godfroi, 1996; Pelsmacker et al., 2004; Rzemieniak, 2015; Anishenko, 2015; Kirakosova, 2014; Marphitsin, 2017], several metrics for evaluating perception have been determined as follows: CPM, CTR, eCPC, likes, reposts, number of comments, character of comments, total reach, reach among subscribers, group clicks, users joined the group, link clicks, complaints, posts hidden and all posts hidden.

The content analysis (both qualitative and quantitative) was chosen as the best method of data collection as it is a common method for evaluating messages of various fields (advertising, in particular) and because it corresponds to our accent made on content of messages. In the next chapter the research design, methodology and data collection by advertising messages' content analysis of chosen online education of entrepreneur and social marketing skills community will be described.

2. Research on Determination of Factors Influencing Consumers' Perception of Advertising Messages

2.1 Research Design

Kotler (2009) underlines that advertisement must be effective to be able to catch attention of the customer. Every day people receive huge amounts of information and try to avoid advertising messages by deleting them without reading. They are in the state when a person's attention can hardly be attracted. That is why it is crucial to create the messages that will be able to catch it.

Our research is devoted to advertising messages and its perception by consumers. Therefore, the aim of the research is to analyse the connection between defined characteristics of advertising messages and consumers' perception.

Our research consists of several steps:

1. To substantiate the choice of research methods and approaches;

2. To state and substantiate the hypotheses;

3. To create methodology on the base of theoretical analysis;

4. To determine a sample;

5. To collect data (contextual advertising messages) through content analysis;

6. To run statistical analysis (descriptive and inferential) in IBM SPSS;

7. To analyse results and discuss them.

According to the results of the previous studies, [Kwock & Yu, 2013; Kirakosova, 2014; Asur et al., 2012] the role of the content was not admitted significant in the message's effectiveness, but it was proved not in all researches. Moreover, two of these papers analysed data in Facebook, thus, it is of upper interest to conduct the research on the base of Russian social network Vk.com.

Ryzhkova (2010) has defined that the numbers of symbols do not matter, because if a text and its content are interesting, a person will read it. Therefore, the method of content analysis was chosen as a common and valid method for analysing advertising texts, which allows us to study topics and messages themselves, focusing attention on a content of a message. We will use mixed (qualitative and quantitative) content analysis which means that we will code data by ourselves as well as use SEO-analysis and will test statistical significance. We will study only the structure and the content of the messages, without studying particular words because this approach to the analysis leads to the following problem - the lexical groups are very difficult and practically impossible to find, they are dependent on the topic of messages. Furthermore, the same words can be used in various meanings and in messages of different topics in popular as well as unpopular communities. Several authors have a common point on this approach and raise the problem of context [Kirakosova, 2014; Cook, 1992; Downing, 2000; Semino, 1997].

The restriction of this method is the following: the sense of the elements which were chosen for the analysis should be predicted as well as the every possible result should be determined and classified relatively to the expectations of the researcher. The problem is that a researcher must predict not only the mentions which he can get but the elements of their contextual use with a detailed system of evaluating every notice. It is preferable to access the notices of several scholars. The harder objective is to assign particular marks to key characteristics - when we should decide whether the mention is done in a positive or a negative meaning and when we should range the mentions relatively to the strength of their marks (e.g., from the most positive to the least positive).Sometimes researchers use the statements of arbitration group (experts) about the intensity of a term.

The data will be statistically processed in IBM SPSS - software which allows us to see descriptive statistics as well as launch correlation analysis which will let us see the correlations between defined metrics and the directions of these correlations.

We have received the demand for the research from a company providing services in the field of online learning (entrepreneur skills and social media marketing). It is noticeable that a company does not have an office and has online business model. Therefore, this research is actual and useful. The name of the company is not mentioned for the purpose of confidentiality.

Vk.com is a vital platform for the company after Telegram was blocked. Moreover, aaccording to the data of enormous Russian services of Internet Statistics, the most popular social network in Russia in the period from 03.2017 till 02.2018 was Vk.com [Rating of social networks, 2018]. Moreover, the research on advertising messages were mostly conducted in other countries, therefore, the social media databases were Facebook, Youtube and Twitter. Thus, it is of upper interest and benefit - to research communities in Vk.com.

After the analysis of theoretical studies several hypotheses have been stated. (Table 2)

Table 2 Research Hypotheses

Hypothesis	Justification
H1:Media type influences the elements of clients' perception	[Kwock & Yu, 2013], [Kirakosova, 2014]
H2:Dominated colors (cold/warm) influence engagement (number of likes, reposts and comments)	[Lohse & Wu, 2001]
H3:The elements of consumers' perception are influenced by a learning material	[Anishenko,2015]
H4:Presence of a woman influences the elements of consumers' perception	[Godfroi, 1996].
H5:Presence of a man influences the elements of consumers' perception	[Godfroi, 1996].

2.2 Research Methodology

The messages were chosen by a census method, which means we study all advertising messages which suit our criteria: total reach of a message should be more than 1000; impression of advertising campaign should be not less than 50000.After this filter we got 106 advertising messages every of which is a unite observation.

We have 2 main concepts: advertising messages and consumers' perception. We have chosen 2 characteristics of advertising messages as media type and its content. Media type is determined as a post's format: text, text with a link, post with a photo and post with a video. In our research we defined the following metrics for media type:

ѕ Text

ѕ Photo

ѕ Picture (art, drawing)

ѕ Gif

ѕ Info graphics

ѕ Meme

ѕ Link to a website

ѕ Video

ѕ Audio

In this paper we closely view the content of two messages' parts: text and picture. The metrics for evaluating textual part were formulated, according to the various researches by Ryzhkova (2010), Gordon & Gordon (2012), Suchkova (2016), Vasilevich & Ledeneva (2005), Shardakov (2016), Anokhin (2017).

They are the following:

ѕ Number of signs

ѕ Number of words

ѕ Nausea (%)

ѕ Water content (%)

ѕ Presence of a title

Several metrics for evaluating visual elements in messages will be applied: colors (black and white or colorful; dominated warm or cold tones), presence of corporate style (for eg. logotype, fonts) and presence of a human (presence of a man, presence of a woman and presence of the company's worker) and presence of a nature. [Suchkova, 2016; Knabe, 2006; Lohse & Wu, 2001] Content classifications of posts was divided into 2 groups: the metrics of the first define the topic of a message and include the following metrics: promotional, competitions, entertaining, learning, news, ugc, interactive [Anishenko, 2015; Smith et al., 2012; Specialists WFA, Millward Brown, & Dynamic Logyc, 2011].

ѕ Promotional posts - the main goal is to promote something (description of products and services, posts about payments of products, discounts, sales, posts about usage of products);

ѕ Competitions* - can be define as a part of promotional content but due to its spread it can be classified as a unite type - the main goal is to widen an audience reach and increase its involvement (contests, quizzes);

ѕ Entertaining posts - the main goal is to remove intellectual tension and cheer up subscribers (humor, anecdotes, quotes, stories);

ѕ Learning material - the main goal is to educate the audience and increase its loyalty (learning articles, the overview of learning resources and services, webinars, description of a company's experience, inscriptions and myths);

ѕ News material - the main goal is to attract users' attentions to a product and to inform them (news about company's field, news about company itself and its products, vacancies);

ѕ UGC (user generated content) - the main goal is to humanize a brand (reviews of products, users' questions published on a company's wall);

ѕ Interactive (engaging content) - the main goal is to involve an audience and increase their activism, create the atmosphere of friendly and live communication (daily posts, surveys, night or daytime chats, request for advice or giving of advice, discussions under a post, games, provocations, complaints).

The second group reflects the key elements (structure) of advertising message. The elements are as follows: advertising message (what the text is about), advertising treatment (to whom it addressed and how the addressee affects the message's style) and call to action (what should be understood by a recipient). [Uhova, 2012]

Perception is defined as the «organization, identification, and interpretation ofsensoryinformation» in order to produce and grasp the given information or the environment.[Schacter et al.,2011] The variables of perception were defined as a result of literature review as follows:

ѕ CPM

ѕ CTR

ѕ eCPC

ѕ Number of likes

ѕ Number of reposts

ѕ Number of comments

ѕ Comments' character (positive, negative, cannot be determined)

ѕ Total reach

ѕ Reach among subscribers

ѕ Group clicks

ѕ Joined the group

ѕ Link clicks

ѕ Complaints

ѕ Posts hidden

ѕ All posts hidden

The operationalization of our concepts can be presented in Table 3. Noticeable that our company uses hybrid models for evaluating advertisement, because the metrics of impression and effectiveness models are used both.

Table 3 The operationalization of main concepts

Textual content	Number of signs	interval
	Number of words	interval
	Nausea	interval
	Water content	interval
	Presence of a title	nominal (binary)	1-yes,0-no
Visual content	Colours	nominal(binary)	1-colorful,2-black and white
	Colours 2	nominal(binary)	1-warm, cold
	Presence of corporate style	nominal(binary)	1-yes,0-no
	Presence of a man	nominal(binary)	1-yes,0-no
	Presence of a woman	nominal(binary)	1-yes,0-no
	Presence of nature	nominal(binary)	1-yes,0-no
	Presence of a worker	nominal(binary)	1-yes,0-no
Media type	Text	nominal(binary)	1-yes,0-no
	Photo	nominal(binary)	1-yes,0-no
	Picture (art, drawing)	nominal(binary)	1-yes,0-no
	Gif	nominal(binary)	1-yes,0-no
	Info graphics	nominal(binary)	1-yes,0-no
	Meme	nominal(binary)	1-yes,0-no
	Link to a website	nominal(binary)	1-yes,0-no
	Video	nominal(binary)	1-yes,0-no
	Audio	nominal(binary)	1-yes,0-no

Table 3 The operationalization of main concepts

Perception	CPM	interval
	CTR	interval
	eCPC	interval
	Likes	interval
	Reposts	interval
	Comments	interval
	Comments2	ordinal	1-positive, 2-negative, 0-hardly can be identified
	Total reach	interval
	Reach among subscribers	interval
	Group clicks	interval
	Joined the group	interval
	Link clicks	interval
	Complaints	interval
	Hidden	interval
	All posts hidden	interval
Content type	Promotional	nominal(binary)	1-yes,0-no
	Competitions	nominal(binary)	1-yes,0-no
	Entertaining	nominal(binary)	1-yes,0-no
	Learning	nominal(binary)	1-yes,0-no
	News	nominal(binary)	1-yes,0-no
	UGC	nominal(binary)	1-yes,0-no
	Interactive	nominal(binary)	1-yes,0-no
Content type 2	Advertising message	nominal(binary)	1-yes,0-no
	Advertising treatment	nominal(binary)	1-yes,0-no
	Call to action	nominal(binary)	1-yes, 0-no

2.2 Data Collection

Company gave us access to the advertising office in Vk.com.

Pict.1 The screenshot of an advertising office of selected company

This is the example (Pict.2) of one of the analysed messages. We see that it is promotional post, we mark this row as 1, we see a man - we also mark a presence of man as 1, but a presence of woman, nature and worker as 0 etc.

Pict.2 Example of analysed advertising message

The data were collected in Excel table and have the following view. (Pict.3) Every line is a single observation - advertising message, columns - are our defined metrics.

Pict.3 Screen of collected data

Textual metrics

Number of signs, words, water percent, nausea - interval metrics, they were calculated in the Internet resource Text.ru as this service was noted by experts what was revealed in literature review. The percent of water (or stop words) has the following interpretation:

ѕ 0-15% - normal;

ѕ 15-30% - higher than normal;

ѕ 30%> - very high percentage of water.

The percent of nausea has the following interpretation:

ѕ 0-30% - the absence or normal presence of key words;

ѕ 30-60% - text is SEO optimised. Mostly search engines count this text relevant according to its key words;

ѕ 60%> - highly optimised or spammed text. [Text.ru,2018]

The presence of a title - nominal (binary) metric which was responsible for any presence of a title: in the text or on the picture/photo.

Visual metrics

We have created 2 metrics for color.

ѕ Colors - nominal (binary) - means the choice between colorful visualization and black and white

ѕ Colors 2 - nominal (binary) - means the choice between visualization's dominated tones: warm or cold

ѕ Presence of corporate style - nominal (binary) - had the meaning of 1 if a post had features of a unite style (fonts, logotype etc), if no - we marked it as 0

ѕ Presence of a man, Presence of a woman, Presence of nature, Presence of a worker - 4 nominal (binary) metrics which were marked by 1 if the mentioned content appeared

If posts had no visual parts, we marked them as 99.

Media type metrics

Text, Photo, Picture (art, drawing), Gif, Info graphics, Meme, Link to a website, Video, Audio - 9 nominal (binary) metrics which were marked as 1 if element existed in message, 0 - if not.

Perception metrics

CPM, CTR, eCPC, Likes, Reposts, Comments, Comments 2, Total reach, Reach among subscribers, Group clicks, Joined the group, Link clicks, Complaints, Hidden, All posts hidden - 15 metrics evaluating consumers' perception. All mentioned variables are interval, except Comments 2 - it is nominal and coded as 1- if positive comments dominated, 2 - if negative, 0 - if the character can hardly be identified. If there were no comments, the observation had a value of 99.

Content type

It had 2 groups of metrics - the first one - Promotional, Competitions, Entertaining, Learning, News, UGC, Interactive - was devoted to the topic of messages and their content. The second - Advertising message, Advertising treatment, Call to action evaluated them from structure part. All 10 metrics are nominal (binary). 1 - if yes, 0 - if no.

In the following chapter the results will be described and analysed.

3. Description of Received Results. Determination of Factors Influencing Consumers' Perception of Advertising Messages

3.1 Research Results

In this chapter the results after the conducted analysis in IBM SPSS will be described and discussed. According to the frequency statistics, we got the following results which are viewed in Table 4 and Appendix 1.

Table 4 Descriptive statistics of interval variables

Name of a metric	Minimum	Maximum	Mean	Median
Number of signs	118	2008	330,87	220
Number of words	16	301	46,25	36
Water content (%)	0	37	0,21	0,185
Nausea (%)	0	48	0,1491	0
CPM	30	230	105,64	80
CTR	0,067	2,962	0,761	0,519
eCPC	4,61	71,7	18,41 42,98	14,505
Likes	0	495	42,98	6
Reposts	0	75	7,7	1
Comments	0	3	0,1	0
Total reach	1087	195805	18578,2	6474
Reach among subscribers	0	7760	155,1	0
Amount of money spent	71,7	35591,22	2457,1	475,29
Group clicks	0	761	51,2	9
Joined the group	0	179	11,5	1
Link clicks	0	2746	200,3	21,5
Complaints	0	144	9,44	4
Hidden	1	730	67,81	26
All posts hidden	1	535	42,54	14,5

We can see that our sample contained metrics with various values. It included short texts as well as long. The percent of stop-words was not higher than 48, thus, the advertising texts of chosen community are not spammed. The percentage of water content varied from 0 to 37. Some texts had a very high percent of water. CTR varied from 0,067% to 2,962%. In Russian Internet it is usually fluctuated from 0,1 % to 2 %. [Torop, 2006] Therefore, we conclude that the coefficient is common and sometimes even is higher which can be a signal of effective media planning and targeting. The higher is this rate the higher is the effectiveness of platform on which the advertising was placed. [Hikari, 2017] CPM varies from 30 to 230 which means that sum for every 1000 impressions differs from 30 to 230 rubles. eCPC as average cost for click varies from 4,61 to 71,7 rubles and is counted as amount of spent money divided to the amount of clicks. The amount of likes varies from 0 to 495, reposts - from 0 to 75. Noticeable, that there were only 5 posts which were commented. Therefore, the character of comments cannot be reflected in the analysis. Total reach was one of the filters - the messages only from 1000 reach were analysed. Total reach describes the number of unique users who have seen a message at least once. 0 - 7760 people from the company's community have seen contextual messages. Amount of money is additional data which were taken from advertising office. It means that sums for advertisement vary from 71, 7 - 35591, 22 rubles. Clicks to companies' groups vary from 0 to 761 and the amount of people who joined the group varies from 0 to 179. Clicks on website links were at the amount of 0 - 2746. From 0 - 144 people made a complaint to a community, from 1-730 people hide particular context advertising messages whereas from 1-535 hide all posts of communities.

As we can see from charts in the Appendix 1, dominated colors in advertising messages were warm - in the percentage of 66, 05. Cold colors appeared in the percentage of 33, 98. We could see a man in 62, 14% of messages, whereas woman was presented only in 15, 53% of advertising messages. Nature appeared in 29, 13% of messages; we could see company's employee only in 5, 83%.

From frequency tables (Appendix 2) we could see the following information: textual characteristic as presence of title exists in all messages. From visual features, all messages are colourful and contain elements of corporate style. According to media type parts, all messages contained text and website link. There were no messages with meme, gif, audio and video. In relation to the content type, all messages were promotional; there were no messages from competitions and UGC. All structural parts occurred: advertising message, advertising treatment and call to action. Finally, the metric «dominated character of comments» characterizes only 2 messages with comments from 5 because 3 of other comments were marked on messages which were deleted. To conclude, mentioned metrics will not be taken into consideration during the future analysis because we cannot compare messages by these parameters.

2 additional variables were created: media type which included sum of photo, picture and info graphics as only these variables differed in observations and engagement which included likes, reposts and comments.

The analysis through cross tabs have been conducted for metric and nominal variables with Pearson's chi square, but all combinations showed a mistake because a lot of cells have expected count less than 5. Therefore, correlation analysis should also be conducted. Firstly, we should check the normality for scale variables as it is one of the criteria for correlation analysis.

Table 5 Crosstabs between presence of a photo and number of reposts

Presence of a photo * Number of reposts

Chi-Square Tests	Value	df	Asymp. Sig. (2-sided)
Pearson Chi-Square	38,375a	25	,043
Likelihood Ratio	38,496	25	,041
Linear-by-Linear Association	1,707	1	,191
N of Valid Cases	106
a. 47 cells (90,4%) have expected count less than 5. The minimum expected count is ,22.

Normality check

P-P plot and histogram

All variables are not normal by P-P plot and histogram.

Mode, median, mean are also different. (If normal - they are equal) (Pict.6)

Kurtosis and skewness both are not less than 0 in all variables. (Pict.6)



Pict.4 P-P Plot of Group clicks

Pict.5 Histogram with normal distribution of Number of words

Pict.6 The frequency statistics of interval variables

Non-parametric test (Kolmogorov-Smirnov) also showed the absence of normality. (Pict.7)

Pict.7 Non-parametric test of Kolmogorov-Smirnov

Therefore, we will use Spearmen's test. Firstly, we checked the correlations between dominated colors, presence of man, woman, nature and a company's worker. The results are the following: dominated colors do not correlate with likes, reposts, comments and engagement, in general. We reject H2 and accept an alternative - Dominated colors (cold/warm) do not influence engagement. However, the presence of woman and presence of nature correlate with likes, reposts, comments and engagement. If woman and nature present on a visual part of advertising message, the number of likes, reposts, comments and engagement decrease. Therefore, we accept H4 - Presence of a woman influences the elements of perception. (Pict.8) Then, correlations between presence of photo, info graphics or picture with likes, reposts, comments and engagement did not occur. Correlations between the content type (entertaining post, learning, news, interactive) were not found with likes, reposts, comments and engagement.

Then, we checked correlations between CPM, eCPC, CTR, total reach and reach among subscribers with dominated colors, presence of a man, woman and company's worker. Dominated colors and presence of worker do not correlate with chosen metrics of perception. Presence of man is positively correlated with reach among subscribers: the more a man is presented on a photo, the more is reach among subscribers. We accept H5: Presence of a man influences the elements of consumers' perception. Presence of woman increases eCPC, but decreases total reach. The presence of nature decreases CTR, total reach and reach among subscribers, but increases eCPC.

Pict.8 Correlations between dominated colors, presence of man, woman, nature, company's worker and likes, reposts comments and engagement

Then, we analysed correlations between CPM, eCPC, CTR, total reach and reach among subscribers and media type (photo, picture, info graphics). Presence of photo is correlated with total reach negatively. The presence of a picture also decreases CPM, but increases total reach. Therefore, we accept H1: Media type influences the elements of clients' perception. Correlations between CPM, eCPC, CTR, total reach and reach among subscribers and content were also examined. Learning material decreases CPM and reach among subscribers. Therefore, we accept H3: The elements of consumers' perception are influenced by a learning material.

Then, we checked the correlations of group clicks, number of users joined the group, link clicks, number of complaints and number of post hidings and all post hidings with presence of woman, man and dominated colors. The presence of a woman decreases link clicks. However, it also decreases number of post hidings and all post hidings. Next, we saw presence of nature decreases group clicks, number of people joined the group, link clicks, but also decreases number of complaints, posts hidden and all posts hidden. Presence of a photo decreases post hidden and all post hidden. Content does not influence such metrics as joined the group, group clicks, link clicks, number of complaints, number of posts hidden and all posts hidden.

Then, the number of signs, words and percent of nausea and water content were checked with perception metrics. The higher number of signs and words the higher is eCPC. The higher is nausea the higher is eCPC. If number of signs increases, the number of likes decreases, but the number of comments increases. If the number of words increases, the number of comments also increases. The higher is nausea the higher is number of comments. The higher is the number of signs the less is engagement. Total reach, reach among subscribers, group clicks and people joined the group are not correlated with number of sighs, words, water and stop-words. The higher is percent of nausea the less are link clicks, the higher elements are used included in media type (picture, photo, info graphics) the less is the number of link clicks. The results are depicted in Table 6. All mentioned correlations were significant on 95 level of significance.

As our results have shown, perception of advertising messages is connected with content of messages and its media type. However, the content plays more important role than media type because the majority of our variables relate to the content metrics. The following result can be achieved because of the lack of variety of media type elements and the absence of possibility to compare the messages with various elements except the presence of picture, photo and info graphics.

Among the factors influencing perception elements are the following: presence of picture, presence of photo, media type, presence of nature, presence of woman, presence of man, number of signs, number of words, % of nausea and learning material. The main general conclusion is that every element of perception is influenced by various factors: group clicks are influenced by presence of nature, whereas the number of posts hidden - by presence of a woman, presence of nature and presence of photo. The same factor has different influence on various perception metrics: presence of woman increases eCPC, but decreases the number of likes.

Table 6 Factors influencing consumer's perception of contextual advertising messages in Vk.com

CPM	Presence of picture	Decreases CPM
	Learning material	Decreases CPM
CTR	Presence of nature	Decreases CTR
eCPC	Presence of woman	Increase eCPC
	Presence of nature	Increases eCPC
	Number of signs	The higher is number of signs the higher is eCPC
	Number of words	The higher is number of words the higher is eCPC
	Nausea	The higher is % of nausea the higher is eCPC
Likes	Presence of woman	Decreases number of likes
	Presence of nature	Decreases number of likes
	Number of signs	The higher is number of signs the less is number of likes
Reposts	Presence of woman	Decreases number of reposts
	Presence of nature	Decreases number of reposts
Comments	Presence of woman	Decreases number of comments
	Presence of nature	Decreases number of comments
	Number of signs	The higher is number of signs the higher is number of comments
	Number of words	The higher is number of words the higher is number of comments
	Nausea	The higher is nausea the higher is number of comments
Engagement	Number of signs	The higher is number of signs the less is engagement
	Presence of woman	Decreases engagement
	Presence of nature	Decreases engagement
Total reach	Presence of woman	Decreases total reach
	Presence of nature	Decreases total reach
	Presence of photo	Decreases total reach
	Presence of picture	Increases total reach
Reach among subscribers	Presence of man	Increases reach among subscribers
	Presence of nature	Decreases reach among subscribers
	Learning material	Decreases reach among subscribers
Group clicks	Presence of nature	Decreases group clicks
Joined the group	Presence of nature	Decreases number of people joined the group
Link clicks	Presence of woman	Decreases link clicks
	Presence of nature	Decreases link clicks
	Nausea	The higher is % of nausea the less are link clicks
	Media type	The higher elements are included in media type the less is number of link clicks
Complaints	Presence of nature	Decreases number of complaints
Hidden posts	Presence of woman	Decreases number of post hidden
	Presence of nature	Decreases number of post hidden
	Presence of photo	Decreases number of post hidden
All posts hidden	Presence of woman	Decreases number of all posts hidden
	Presence of nature	Decreases number of all posts hidden
	Presence of photo	Decreases number of all posts hidden

The results we achieved agree with several authors as well as disagree. Our conclusions are in contradiction with results of Kwock & Yu (2013) because the authors claimed that media type is important for post effectiveness whereas the content is not. We claim that they are important both, but content is correlated with elements of consumer's perception more often. We agree with the results of Anishenko (2015); they showed that 12 % of users are not active at all and that people more like than comment posts and communicate with others. People liked posts more than commented them. Our results are in agreement with conclusions of Kirakosova (2014) about correlation between likes, reposts and media type - they are not statistically significant depend on a media type of post. Our research not agree with the results by Asur et.al (2012) that content do not influence message's popularity because we received that number of likes is correlated with number of signs, presence of woman and presence of nature.

Conclusion and Discussion

The received results have shown that the consumers' perception is connected with the content and media type of advertising messages, but the content plays more significant role. We accepted the following hypotheses: H1: media type influences the elements of clients' perception, H3: The elements of consumers' perception are influenced by a learning material, H4: Presence of a woman influences the elements of consumers' perception and H5: Presence of a man influences the elements of consumers' perception. We rejected H2 and accepted the alternative hypothesis - dominated colors (cold/warm) do not influence engagement.

The value of our theoretical analysis is that we created unique classifications for evaluation of advertising messages and measurement of perception. In relation to the results we received, they can be used by SMM managers of chosen company. This research paper is the first in Russia which studied several metrics of perception in details, media type and content both on the platform Vk.com.

The restriction of this work is that we studied contextual advertising messages of one organization; therefore, the results cannot be applied to other fields, except online business education. Moreover, we could not compare messages of all media types because messages contained memes, gifs, audio and video files were absent in our sample. Moreover, we could not check the correlations with all content types because UGC, competitions were not represented in our sample. However, during the content analysis to decrease its limitations, it is better to predict all possible options, as we did. If we did not evaluate textual and visual messages in details, we would have lost some important factors (number of signs, words, presence of woman etc).

To sum up, the aim of this research was completed - the factors of advertising messages influencing consumers' perception were formulated and described. The future direction of this research is to enlarge the number of observations, to study messages from various research fields and to study various types of advertising messages as well. (e.g., banners, advertising messages of social media communities).

References

1. Anderson E., Mary W. (1993). The antecedents and consequences of customer satisfaction for firms. Marketing science, 12(2), 125-143.

2. Anishenko T. (2015) Company's activism in social networks as instrument of consumer's trust (Master's dissertation).

3. Anokhin R. (2017). 6 best services to check for plagiarism online. Geek-nose.

4. As...