Today the dynamics and amount of modern information flows created by the Internet media content make a significant load of work for an expert while preparing an analytical conclusion on certain subject, industry and general trends. In the paper, the estimation of the limit values for detecting the PReffect of electronic mass media integrated index is experimentally substantiated by conducting comparative and statistical analysis of numerical sequences of linguistic statistics for sets of messages which are intended a priori for purely "unbiased" reporting and documents with vocabulary characteristical for information influence (direct action commercial texts).
The dynamics and amount of modern information flows created by the Internet media
content (news feeds, blogs, micro blogs, audio and video streams etc.), make a
significant load of work for an expert while preparing an analytical conclusion on certain
subject, industry and general trends. In addition, as proved by the works of scientists
and specialists devoted to the information space [
Thus, the use of automated detection of "PR-effect" by applying an automated algorithm for the initial analysis of a given text and/or a load of texts, which would allow to pre-evaluate and decompose the flow of messages into those representing collections of documents with the vocabulary characteristical for “PR-effect” and the others, which, in our opinion, is an essential tool for analytical work and information support of decision-making.
Such a technological method was proposed by us in [
PRI (MTIPt ) M PRI (MTIPt )i , where TIF "PR-effect" index for the i-th day makes 1/3 of the number of messages with the features of "non-eventivity", "emotionality", "manipulativity" to the total number of messages of the thematic vector MTIP: (1) (2) (3)
MTIPti Pr (m ) 1 where r={EV, EM, MA} is the index for the characteristic with linguistic features, respectively {"eventivity", "emotionality", "manipulativity"}; |MTIP∆ti| is the power of a set of TIF messages for the i-th day.
Based on the construction, the value of PRI(MTIP∆t) is always positive and falls within the range between 0 and 1.
The threshold value α is set as the criterion for making a decision on the presence of "PR-effect" in the TIF. Thus if
PRI (MTIPt ) , (4) then such a TIF represented by the thematic vector MTIP∆t possesses features of "PR-effect".
Consequently, the given value α sets the basis for the automated decision-making on detecting the PR-effect. Therefore, the experimental substantiation of its estimate value is significantly relevant. 2
Nowadays, research and evaluation of information flows (IFs), including those
created by electronic media, is mostly carried out within convention-analytical research.
The respective criteria for assessing the quality of quantitative content analyses based
on the requirements of national and foreign methodologists is suggested by
Ivanov O.V. [
Among the approaches to solving the problem of assessing IF texts created by
electronic media, the methodology for assessing the quality of messages’ resonance
should be noted, which is the original development of one of the leading national
consulting companies NOKS FISHES [
K(tl) Logical presence coefficient – reflects the proportion (%) of the company’s infodrive logical presence in the text array. Includes the tone value and subjective saturation of reference for both the given company’s and the other companies of the given trend = (logical presence coeff. + emotional presence coeff) / 2 =[K(l)+K(t)] / 2 K(l) – subjective representation coeff. = D of the given object / Σ (D of all the objects in the publication) K(t) – emotional representation coeff. = |To| of the given object / Σ (|To| of all the objects in the publication) Pg – potential influence of mass media rate Step 3. Integrated characteristics of the reference quality W – media’s loyalty to IF and the demand for media’s IF Includes the qualitative characteristics of media’s perception of the company's infodrive or shows the editorial policy vector for the company's infodrive = image * abstract * distribution type * author’s tone value * (text array size + publication type) * reference type * saturation * logical presence coeff.
Y – Event saturation of the client’s media area Includes qualitative characteristics of the text array’s event filling by the company’s infodrive = image * abstract * eventivity * event tone value * (text array size + publication type) * reference type * saturation * logical presence coeff.
Z – Probable influence – informational drive (ID) towards the media’s audience Determines the probability of reporting the company's information to the general audience = marginality * eventivity * potential influence of mass media coeff. * (7 – publication size) * reference type * subjectivity * logical presence coeff.
Step 4. Rating R – integrated resonance quality index If W < 0, Y > 0, |W|>|Y| , β = - √ W2 - Y2 + Z2 If W < 0, Y > 0, |W|<|Y| , β = √ Y2 - W2 + Z2 If W < 0, Y < 0, β = - √ W2 + Y2 + Z2 If Y < 0, W > 0, |W|<|Y| , β = - √ Y2 - W2 + Z2 If Y < 0, W > 0, |W|>|Y| , β = √ W2 - Y2 + Z2 If Y > 0, W > 0, β = √ W2 + Y2 + Z2 However, these tools, as well as the previously mentioned areas of research, are based on methods of expert evaluation, which in the general case does not exclude the systemic error problem mentioned above.
In previous works, by analyzing the trends of the TIFs created while categorizing the
general IF and the synthesis of linguistic statics (both standalone messages and
calculated for a specific TIF load that had the characteristic features of PR-effect) [
In this paper, we will focus on the detailed substantiation of the "PR-effect" index and the estimation of the criterion α for the decision-making on "PR-effect" in the TIF (4). Thus, the purpose of this paper is to formally present the technology for detecting the "PR-effect" of electronic media by: conducting a comparative analysis for sets of messages that are a priori intended for "unbiased" reporting, and documents with the vocabulary characteristic for the "PR-effect"; conducting statistical analysis of the numeric sequences of estimates of PRI(MTIP∆t)i and experimental sets of data (documents) of both types; setting (substantiation) of the limit conditions for estimation of the value α of the method "PR-effect" detection, which is the task of the present work. The subject of the study is two groups of information messages. One of them is to be focused on "unbiased" reporting, the other one contains vocabulary characteristic for the "PR-effect".
The object of the study is the boundary criterion α of the "PR-effect" index. 3
To conduct a study to determine the limit value of the criterion α we performed the
following:
1. In the system of information space analysis (SISA) [
Taking into account the above, the algorithm for the experimental determination of the limit values of the value α for the "PR-effect" detection can be represented by a block diagram (Figure 2).
Fig. 2. Algorithm for determining the limit value of α for "PR-effect" detection
Thus, for the purpose of visual comparison and estimation of the type and nature of
the probability distribution FC1 of the random values obtained for C1 and FC2 for C2
frequency distribution histograms of estimates PRI(C1,2)i for each document in the
corresponding group are constructed. Given that the construction of the set of values
PRI(C1,2)i, at i 1, n where n is approximated by a continuous function in the
range between 0 and 1, and the number of class intervals calculated according to the
Sturges rule [
As it can be seen from the figures (qualitative analysis of probability distribution for
FC1 and FC2):
1. Probability distributions both of FC1 for C1 (three modes clearly expressed) and
FC2 for C2 are not Gaussian (that is, they have a non-parametric form), and therefore
the use of standard statistical estimates for the calculation of numerical characteristics
(e.g. mean value or deviation) of the distribution function, as well as the use of
Student's criterion to confirm the hypothesis regarding homogeneity of the samples, will
not be correct;
2. 83,03% of C2 (set of texts by copywriters - Fig. 5) received an estimation of
PReffect index at PRI(C2)>0,975, while the same level of evaluation corresponded to
only 3,01% of the documents in C1 (texts from GED - Fig. 4);
3. Certain number of documents (18.33%) from C1 (GED texts - Fig. 4) have an
estimate of PRI(C1)>0,75, which is due to the literary nature of the information
presentation in some of the encyclopedic articles. These texts include:
- biographies of outstanding artists, historians, and politicians, for example:
- description of historical events and countries that have directly participated in them
or had geopolitical influence [
In addition, the median value is at the level of 0.44 for C1 and 1 for C2, based on which we will subsequently obtain the desired limit value of α for detecting "PReffect".
The hypothesis of the homogeneity of independent samples C1 and C2 is then checked. Theoretical distribution functions FC1 and FC2 are unknown, that is, they belong to the same general set.
Then, the verified zeroth hypothesis Н(0) is presented as: FC1 = FC2 unlike the opposite hypothesis Н(1): FC1 ≠ FC2, while FC1 and FC2 are considered continuous. To test Н(0) the statistics of the non-parametric Kolmagorov-Smirnov criterion is applied: '
n1n1n2n2 max FC1 FC2 , where FC1 and FC2 are empirical distribution functions with two samples of volumes n1 and n2.
Hypothesis Н(0) is not confirmed, if the given statistic value λʹ is more than the critical λʹcr, ie λʹ > λʹcr, otherwise it is considered to be confirmed.
The hypothesis Н(0) testing with the Kolmagorov-Smirnov criterion is conducted using the Statistica 6.0 software. The Kolmagorov-Smirnov criterion has shown that the distribution functions FC1 and FC2 do not belong to the same general set (that is, the hypothesis Н(0) deviates) with an error rate of 5% (Figure 9).
In other words, there is the 95% confidence in a significant deviation between the
experimental data C1 (messages that correspond to the "unbiased" reporting) and C2
(documents that have characteristics of PR-effect vocabulary) PR-effect index values
PRI(C1,2)i, as shown by the Box-and-whisker diagram (Figure 10) [
At this, the value α = 0,696035 is taken as the limit value of the decision-making criterion for the presence of "PR-effect" (4), with an error rate of 5%.
To summarize, as a result of the study, we gave a formal statement for detecting "PReffect" by electronic media (the algorithm is depicted as a flowchart in Figure 1). An experimental substantiation and qualitative analysis of the PR-effect index were performed, which showed the difference in the distribution of the probabilities of its frequencies for both experimental samples (C1 for the articles of the Great Encyclopedic Dictionary, C2 for articles by copywriters) from the Gaussian one (having a non-parametric form). By conducting statistical analysis of the numerical sequences of experimental data (C1 and C2), the hypothesis of the difference in the messages of "unbiased" reporting (C1) and those that contain characteristic vocabulary of PReffect (C2) in the characteristic space of linguistic features was confirmed. The limit value of the decision-making criterion for the presence of "PR-effect" (4) α = 0.696035, with an error rate of 5%, was obtained. At the same time, both qualitative and statistical analysis of the experimental material showed a certain difference in the value of the integral PR-effect index for the characteristic groups of documents. Therefore, as the tasks for further research, we will aim at developing an algorithm for automatic grouping of an arbitrary set of messages in the characteristic space of linguistic features based on the PR-effect index.