Big Data' and Data Mining' Role in Fight Against Drug Trafficking

Belyaev Ivan Yur'evich ORCID: 0000-0002-0126-0934 Graduate student, Department of Theory and History of State and Law and International Law, Federal State Autonomous Educational Institution of Higher Education "Samara National Research University named after Academician S.P. Korolev" 443086, Russia, Samarskaya oblast', g. Samara, ul. Moskovskoye shosse, 34 ivanbeliaef@yandex.ru Other publications by this author

Belyaev Yurii Vladislavovich lawyer, Law office N 544 443112, Russia, Samarskaya oblast', g. Samara, ul. Sergeya Lazo, 25A vana682@gmail.com

The last decade has been marked by the spread of a new category of narcotic drugs, called new psychoactive substances (NPS) ^[1]. The chemical structure of new psychoactive substances is heterogeneous and includes not only patented components, analogues of traditional drugs or prescription medications, but also new substances ^[2]. Some of these components and their effects are well studied, while the toxicity of others is not known. This is associated with an increase in serious side effects of new psychoactive substances, expressed in mass infections, psychoses, acute poisoning and overdose ^[3].  As a result, many new psychoactive substances are capable of causing serious harm to the health of their consumers ^[4].

Since the legal status of new psychoactive substances varies greatly in different countries, only some of them are included in international anti-drug conventions, which facilitates the turnover of these substances at the international level ^[5].

In response to the rapid spread of new psychoactive substances, the European Monitoring Center for Drugs and Drug Addiction (ECMN) and the United Nations Office on Drugs and Crime (UNODC) have implemented an Early Warning System and an Early Warning Advisory Portal, respectively ^[6].

For example, as soon as information about a new psychoactive substance arrives at the European Monitoring Center for Drugs and Drug Addiction, the procedure for assessing the risk of harm caused by it begins.  However, despite the informative nature of these monitoring systems, the responsiveness of their response to the rapid emergence and spread of new psychoactive substances is limited ^[7]. Firstly, the universal method of identifying new psychoactive substances is lagging behind. Secondly, for the transition of the monitoring procedure from the stage of risk assessment to the stage of control of a new psychoactive substance, strict compliance with three criteria is necessary: 1) proof of its psychoactive effect; 2) the ability to frame addiction; 3) dangerous effects on health and social integration. In this regard, the European Monitoring Center for Drugs and Drug Addiction notes the need to supplement the existing early warning system with more sensitive and flexible technological means.

Based on the large number of new psychoactive substances (unlike traditional drugs), as well as the limited nature of information about their effects, the need to optimize the system for obtaining initial data on their distribution becomes particularly relevant. In this regard, many researchers emphasize the special role of the Internet as a source of early detection of new psychoactive substances ^[8]. Any activity in the online environment leaves digital traces that can be used as an indicator of the appearance of new psychoactive substances. In particular, the history of search queries or communication on forums can be used for these purposes ^[9].

One of the key studies in this area is a project called "The Psychonaut Web Mapping Project", aimed at developing and using a web map of the appearance and distribution of new psychoactive substances in order to identify new trends over a two-year period ^[10]. This project has confirmed that with regard to the availability and use of narcotic drugs, the study of online forums allows you to obtain unique information that cannot be obtained by other means (in particular, sociological surveys). Within the framework of this project, it was possible to identify more than 400 new psychoactive substances even before information about them appeared in the scientific literature and became available to the general public.      

The effectiveness of this approach was also confirmed in another study, where one of the well-known online drug forums was monitored ^[11]. The analysis of the discussions of narcotic drugs conducted there made it possible to identify the direction of evolution of classes of narcotic drugs and individual narcotic components within these classes. Moreover, some of the new psychoactive substances were detected in this way earlier than they were identified by the EU Early Warning System mentioned above.

Another irreplaceable source of information in the field of studying new psychoactive substances is Google Trends. As shown in a number of studies, consumers interested in purchasing narcotic substances previously collect information about them using the Google Search search engine. This tool makes it possible to identify the interest of consumers, which has a high degree of correlation with subsequent actions to purchase the selected narcotic drug. At the same time, the available database for search is extensive and is constantly available almost in real time.

So, in the context of the topic under consideration, Google Trends was used to identify deaths from new psychoactive substances and to study the relationship between search queries and publications in the media. At the same time, the most in-depth studies using Google Trends data were conducted by Al-Imam & Abdul Majeed ^[12]. In each of these studies, the compliance of Google Trends data with the patterns identified on the Dark Web was confirmed. Additionally, the effectiveness of Goggle Trends as a means of early detection of new psychoactive substances was demonstrated.

Google Trends data was used by Gamma et al. to compare Google Search queries related to methamphetamine and related crimes in three countries (Switzerland, Germany, Austria). A reliable correlation was found between search queries for methamphetamine and the corresponding criminal acts, which can be used as an indicator of drug-related crimes.  

In this regard, we conducted our own research aimed at identifying the possibility of using the Google Trends database to predict criminal acts in the sphere of trafficking of new psychoactive substances in order to further improve criminal legislation and law enforcement practice.

As a hypothesis of the study, we put forward the assumption that there is a correlation between search queries in the Google service for the keywords "mephedrone" and "pawnshop". In the case of finding a correlation between these variables, at the second stage of the study, we assumed a linear regression analysis to create an equation that allows us to predict the value of the dependent variable "pawnshop".

As the primary data for statistical analysis, data from the Google Trends service for the Russian Federation over the past five years were used for two search queries: "mephedrone" and "pawnshop". The choice of these variables was determined by the following reasons.

1. Mephedrone is one of the most popular synthetic drugs in the Russian Federation.

2. Pawning things, including those obtained by criminal means, is one of the most common ways to obtain funds for the purchase of mephedrone.

3. There is a high probability of using the Google Search search service by drug–addicted subjects to find ways to purchase mephedrone, on the one hand, and pawnshop addresses to receive money, on the other.

The consumers of mephedrone of the Russian Federation who use a pawnshop as a means of obtaining money for its purchase act as a general aggregate in our study. As a sample, we used Google Search service users in the Russian Federation who have been searching for mephedrone and pawnshops over the past five years.

The representativeness of this sample is determined by the randomness of the selection of its elements and a large number of them, which, according to the central limit theorem, allows us to consider this distribution obeying the law of normal distribution.

The sample for this study was formed in the Google Trends service when comparing two search queries: "Mephedrone" and "Pawnshop" over the past five years. The results of the query in the form of a file were imported into the Statistical Package for the Social Studies (SPSS) program for further statistical analysis.

A fragment of the file with primary data for statistical analysis is presented below.

Table 1 – A fragment of primary data for statistical analysis   

The first column "Date" includes the date of the last day of the week during which the search queries were summed up. Since the sample was formed over a period of five years and the measurement interval is a week, there are 260 rows in this column (the number of weeks in five years).

The second and third columns contain the number of search queries in relative (normalized) percentile indicators. The maximum percentile rank is 100, the minimum is 0. The expression of the number of queries in Google Trends is not in absolute values, but normalized – allows you to use a single measurement scale to compare different queries.

It should be noted that the use of the Big Data methodology and the Google Trends service in our case makes it possible to conduct research that could not be carried out using the traditional method – a sociological survey. In our study, we analyze data collected throughout Russia over the past five years. To conduct such a large-scale sociological survey over such a long period of time would require significant material and time costs. At the same time, even if respondents agreed to answer anonymous questions about the presence of drug dependence on mephedrone and planned to purchase it by pawning things (acquired, including by criminal means), the sincerity of such answers would be doubtful.

The descriptive statistics on the data obtained are given below.

Table 2 – Descriptive statistics

As follows from Table 2, the average number of search queries for the keyword "Pawnshop" for a week is more than five times higher than the average number of search queries for the keyword "Mephedrone": 79.9 and 15.4, respectively. At the same time, the average value of search queries for the search query "pawnshop" is in the range of 15.4+-0.9, and for the search query "mephedrone" in the range of 79.9+-1.

Figure 1 – Frequency distribution of the variable "Pawnshop"

At the next stage of the study, in order to answer the question whether there is a statistically significant correlation between these variables, we conducted a correlation analysis in the SPSS program. The results obtained are shown below.

Table 3 – Results of correlation analysis

The obtained correlation coefficient (r=0.57) between the studied variables can be considered as an indicator of a strong relationship between them. As Cohen Jacob notes in one of his books, which has become a classic in the field of behavioral statistics, in social research, a correlation coefficient of 0.1 or less reflects a weak relationship between variables, a value of 0.3 reflects an average degree of connection, and a value of 0.5 or higher indicates a strong connection ^[13].  

Since at the first stage of the study we established a high degree of connection between the variables, at the second stage we conducted a linear regression analysis. The difference between correlation and regression analysis is that the former measures the strength of the relationship between variables, while the latter measures the nature of this relationship ^[14].

As emphasized in the latest edition of the monograph by a leading author in the field of Data Mining, linear regression analysis is one of the most common statistical methods at present. Linear regression analysis is used to describe the relationships between variables in scientific research ^[15].

In the course of linear regression analysis, the value of a variable is predicted based on the value of another variable. The variable that needs to be predicted is called a dependent variable. A variable that is used to predict the value of another variable is called an independent variable.

The dependent variable in our study is the number of search queries for the keyword "Pawnshop". And the independent variable is the number of search queries for the keyword "Mephedrone".

In the course of this analysis, such coefficients of a linear equation with one or more independent variables are selected so that this equation best predicts the value of the dependent variable.

The result of linear regression can be represented as a straight line on a plane that minimizes the discrepancy between the predicted and actual values.

Below is a graph of the ratio of predicted and actual values of the dependent variable based on the constructed regression line.