Demonstration the Importance of Pre‐processing the Text Fields of Bibliometric Records to Identify Promising Research Tasks. Case Study of Scopus Data on Petroleum Reservoir Engineering

Introduction

Materials and Methods

Results and Discussions

Comparison of Author Keywords Clustering with and Without Term Normalization

The easiest way to compare the clustering of author keywords with and without term normalization is to use the VOSviewer program and construct a co-occurrence network of author keywords for both cases.

The co-occurrence network of author keywords using the original values of the “Author Keywords” field and constructed using the following parameters: the total number of author keywords defined by the program is 16102, of which 966 occur 5 or more times. Using 400 keywords with the highest overall strength of connections and a minimum cluster size of — 50, 5 clusters were obtained, a fragment of which is shown in Figure 1.

The issue of lack of keyword normalization is visible, for example, “enhanced oil recovery” and “eor” are both found in the blue cluster, while “coalbel methane” and “cbm” are found in the green cluster.

For detailed review of this network, the file AuKWsGrt4Top400Min50.json placed in the archive can be used at https://app.vosviewer.com/.

The network of co-occurrence of author keywords during normalization of terms of the “Author Keywords” field is constructed under the following parameters: the total number of normalized author keywords defined by the program is 15215, of which 969 occur 5 or more times. Using 400 keywords with the greatest total link strength and Minimum cluster size = 50, 5 clusters were obtained, a fragment of which is shown in Figure 2.

There are notable similarities between Figs. 1 and 2. For example, the most commonly used terms “enhanced oil recovery”, “permeability” could still serve as the names of their clusters, but the term “heavy oil” got into the red cluster during normalization, while in the first figure it is in the purple cluster.]

For detailed review of this network, the file AuKWsGrt4Top400Min50norm.json placed in the archive can be used at https://app.vosviewer.com/.

Let's compare terms in closely related clusters of author keywords obtained by the above two methods.

Note. In the table headings we will use AuKWs to abbreviate “Author Keywords”, and N for occurrence of a given author keyword.

Table 1. Comparison of two clusters, conventionally labeled as “permeability”.

Table 1. Comparison of two clusters, conventionally labeled as “permeability”.

Normalized AuKWs	N	Original AuKWs	N
permeability	392	permeability	392
numerical simulation	250	numerical simulation	245
hydraulic fracturing	187	hydraulic fracturing	187
coalbed methane	125	shale gas	104
porous media	80	coalbed methane	99
enhanced geothermal system	71	horizontal well	88
shale	69	porous media	80
fracture	66	shale	63
natural gas hydrate	51	stress sensitivity	63
two phase flow	44	fracture	48
natural fracture	42	natural gas hydrate	48
tight gas reservoir	42	adsorption	45
depressurization	41	tight oil reservoirs	45
effective stress	40	depressurization	41
gas hydrate	38	effective stress	40
geothermal energy	35	hydraulic fracture	40
temperature	32	two-phase flow	37
mathematical model	31	geothermal energy	35
stimulated reservoir volume	30	enhanced geothermal system	34
fines migration	29	temperature	32
fracture propagation	29	mathematical model	31
gas production	28	apparent permeability	29
methane hydrate	28	fines migration	29
fracturing fluid	27	tight gas reservoir	29
pore network model	27	fracture propagation	28
coalbed methane reservoir	26	gas hydrate	28
modeling	26	gas production	28
supercritical carbon dioxide	26	stimulated reservoir volume	28
finite element method	25	low permeability reservoir	27
permeability anisotropy	24	methane hydrate	26

The terms “permeability”, “numerical simulation”, “hydraulic fracturing” having the same spelling occur in both columns with the same frequency. But already “coalbed methane” occurs 125 times in the left column and 99 times in the right column. Similarly, “two-phase flow” — 44 and “two-phase flow” — 37 and so on.

Table 2. Comparison of two clusters, conventionally labeled as “reservoir simulation”.

Table 2. Comparison of two clusters, conventionally labeled as “reservoir simulation”.

Normalized AuKWs	N	Original AuKWs	N
reservoir simulation	120	relative permeability	116
relative permeability	116	reservoir simulation	115
carbon dioxide storage	96	machine learning	90
machine learning	91	co2 storage	85
heterogeneity	79	heterogeneity	79
carbon dioxide sequestration	72	co2 sequestration	66
petroleum engineering	65	petroleum engineering	65
carbon dioxide injection	61	oil/gas reservoirs	58
oil/gas reservoir	58	sensitivity analysis	57
sensitivity analysis	57	capillary pressure	55
artificial neural network	55	viscosity	52
capillary pressure	55	waterflooding	52
carbon dioxide enhanced oil recovery	52	co2 injection	50
viscosity	52	optimization	50
waterflooding	52	asphaltene	44
optimization	50	gas injection	43
asphaltene	45	reservoir characterization	43
in situ combustion	43	artificial neural network	41
reservoir characterization	43	carbon dioxide	40
history matching	40	deep learning	38
deep learning	38	history matching	35
recovery factor	29	co2	32
artificial intelligence	28	asphaltene precipitation	31
genetic algorithm	28	unconventional reservoir	31
minimum miscibility pressure	28	tight reservoirs	28
bitumen	27	bitumen	27
carbon capture utilization storage	27	ccus	27
numerical modeling	27	numerical modeling	27
unconventional petroleum	27	oil recovery factor	27
crude oil	26	unconventional petroleum	27

In the original records, the term “reservoir simulation” was referred to in both singular and plural, which caused the observed difference in the data columns; even more frequently, the term “artificial neural network” appears in both singular and plural.

On the other hand, terms with the single spelling “heterogeneity,” “viscosity,” “deep learning,” “numerical modeling,” and “bitumen” occur the same number of times in both columns.

An example of a significant difference in occurrence is the term “gas injection”, which appears on the right side of Table 2 and on the left side of Table 4.

The term “tight reservoir” does not appear in the cluster of normalized keywords, but it appears in the right-hand column in the plural “tight reservoirs”.

The term “waterflooding” appears 52 times in both columns of this table and in the spelling “water flooding” in Table 3. This term was not normalized in the process of preparing the author's keywords.

In the right part of the table there are different spellings of the term: “enhanced oil recovery”, “eor”, and “enhanced oil recovery (eor)”, which can significantly affect the clustering results due to the wide use of this term.

The importance of writing the term with and without hyphen can be seen in this table. “low-permeability reservoir” occurs 114 times in the left column and ‘low-permeability reservoirs’ occurs 29 times in the right column, while ‘low-permeability reservoir’ occurs 27 times but already in Table 1.

The term “heavy oil reservoir” occurs 68 times in the left column, and “heavy oil reservoir” 44 times and “heavy oil reservoirs” 22 times in the right column.

The importance of abbreviation to the singular: “nanoparticle” occurs 54 times on the left side, and '“nanoparticles” 36 times on the right side.

Table 4. Comparison of two clusters, conventionally labeled as “tight oil reservoir”.

Table 4. Comparison of two clusters, conventionally labeled as “tight oil reservoir”.

Normalized AuKWs	N	Original AuKWs	N
tight oil reservoir	116	heavy oil	192
horizontal well	107	tight oil reservoir	69
shale gas	107	co2 flooding	52
fractured reservoir	81	imbibition	49
unconventional reservoir	69	heavy oil reservoir	44
stress sensitivity	66	simulation	41
low permeability	58	low-permeability reservoir	40
shale reservoir	52	unconventional reservoirs	38
hydraulic fracture	48	sagd	34
adsorption	45	air injection	32
gas injection	43	geomechanics	32
improved oil recovery	33	fractured reservoirs	30
geomechanic	32	reservoir heterogeneity	30
fracture network	31	threshold pressure gradient	30
threshold pressure gradient	31	multiphase flow	29
shale gas reservoir	30	oil reservoir	28
apparent permeability	29	recovery	26
multiphase flow	29	steam injection	25
carbon dioxide huff n puff	27	water injection	24
diffusion	25	fractal theory	23
pressure transient analysis	25	analytical model	22
fractal theory	23	pressure transient analysis	22
huff n puff	23	heat transfer	20
flowback	19	steam flooding	20
production performance	19	flowback	19
dual porosity	18	numerical model	19
gas condensate	18	production performance	19
non darcy flow	18	thermal recovery	19
flow regime	16	water cut	19
rate transient analysis	16	water saturation	19

The most important thing to note in this table is that the term “heavy oil” is the most frequent term in the right column, but it appears in the left column in Table 3.

The terms “fractured reservoir” and “unconventional reservoir” occur in the singular in the left column, but in the plural in the right column.

The term “shale gas reservoir” does not appear in the right-hand column.

Table 5. Comparison of two clusters, conventionally labeled as “shale oil” and “porosity”.

Table 5. Comparison of two clusters, conventionally labeled as “shale oil” and “porosity”.

Normalized AuKWs	N	Original AuKWs	N
nuclear magnetic resonance	156	shale oil	137
shale oil	137	porosity	114
porosity	114	ordos basin	113
ordos basin	113	pore structure	101
pore structure	101	tight oil	92
tight oil	94	nmr	74
tight sandstone	77	tight sandstone	70
tight reservoir	70	nuclear magnetic resonance	69
diagenesis	66	diagenesis	66
spontaneous imbibition	62	spontaneous imbibition	61
reservoir quality	58	reservoir quality	58
reservoir	56	fractured reservoir	51
shale oil reservoir	41	reservoir	50
junggar basin	40	tight reservoir	42
tight sandstone reservoir	37	junggar basin	40
sandstone	34	sandstone	33
biomarker	33	pore size distribution	31
pore size distribution	33	sichuan basin	31
sichuan basin	31	tarim basin	28
tarim basin	28	yanchang formation	28
yanchang formation	28	shale reservoir	27
controlling factor	27	carbonate	26
source rock	26	shale oil reservoir	24
oil shale	23	tight sandstone reservoir	24
lithofacy	22	controlling factors	23
reservoir rock	22	oil shale	22
fractal dimension	21	fractal dimension	21
reservoir characteristic	21	lithofacies	21
sandstone reservoir	21	reservoir characteristics	21
songliao basin	20	reservoir rock	20

This table is interesting because the lack of disclosure of abbreviations (nmr→nuclear magnetic resonance) can cause that the most frequently occurring term is different on the left and right side of the table.

This table lists the names of the basins “Ordos Basin”, “Junggar Basin”, “Sichuan Basin”, “Tarim Basin”, “Songliao Basin” - all these basins are located in China. This is consistent with the large number of Chinese publications in the studied sample of bibliometric records.

Note: Keep in mind that the above tables only summarize the 30 most frequent terms from the full keyword occurrence tables, so the term occurrence amounts may not match.

The comparison of the terms in the tables above helped to illustrate the importance of term normalization on the results of term clustering.

The importance of term normalization will become even more evident in the next section.

Using the Scimago Graphica program to identify promising research tasks

Within the framework of this article, the promising research tasks are assumed to be described by the author's keywords that occur more often in new publications that have a high citation rate and connection with other terms. Technically, this can be visualized as a slice of the network of co-occurrence of terms represented in the coordinates “Average publication year of the documents in which a keyword occurs” and “Average normalized number of citations”.

Let's explain the concepts used by quoting from the VOSviewer manual2: “Avg. norm. citations. The average normalized number of citations received by the documents in which a keyword or a term occurs.” “The normalized number of citations of a document equals the number of citations of the document divided by the average number of citations of all documents published in the same year and included in the data that is provided to VOSviewer.”

Note: For a better understanding of the content of the publications selected as examples, their titles and my abbreviated version of the abstract are given in the body of the paper; they are enclosed in quotation marks to emphasize that this is not plagiarism but citation.

Figure 3 shows the graph of the term co-occurrence network constructed for author keywords taking into account their normalization. The following constraints were used to construct the graph in Scimago Graphica software: clusters 1,2,3 out of five obtained for normalized author keywords by VOSviewer software; total_link_strength >= 44; avg._norm._citations >= 1. Degree >= 25. Degree was calculated for the pre-built network, then the data from this network was exported, the co-occurrence network was re-built using the exported data, and then the degree >= 25 filter was implemented to limit the number of terms displayed in the graph. This filter leaves only terms that are well related to other terms in the graph, which is important for identifying a local topic described by a number of co-occurring terms. A analogous result can be obtained by applying the “link strength” and “total link strength” filters to the original graph, but all these parameters are calculated for the network in question, so the resulting values may differ. In this case, the capabilities of Scimago Graphica were demonstrated. Filters are used to display only the terms most relevant to the analysis being performed to improve the readability of the graph in the publication.

Definition: The degree of a node is the number of links it has with other nodes in the network.

According to the VOSviewer manual — “Total link strength attribute indicate the total strength of the links of an item with other items.”

Let us consider examples of identifying publications revealing the topics of two clusters (3 and 1) with terms having a high value of the average normalized number of citations.

Third cluster. The results of a sequential search (search in found) for terms appearing in the text fields of all records: methane hydrate → 49 records AND gas production → 33 records AND depressurization → 20 AND fracture → 4. Of the four papers, the most cited is “Enhancement of gas production from methane hydrate reservoirs by the combination of hydraulic fracturing and depressurization method” [15]. According to ScienceDirect, this publication has been cited 171 times as of February 12, 2025. Short summary of annotation: “A fracturing and depressurisation method is proposed to improve the efficiency of gas production from methane hydrate (MG) reservoirs. A model of a fractured MG reservoir was created and the gas production behaviour under different temperature conditions was studied. The effect of increasing fracture zone size and permeability on gas production rate was more prominent in the early stage of depressurisation for the high-temperature reservoir, while the increase in overall gas production was minimal in the low-temperature reservoirs.”

First cluster. The results of a sequential search (search in found) for terms appearing in the text fields of all records: nanofluid → 120 nanoparticle → 93 carbonate → 33 → enhanced oil recovery → 24 → wettability alteration → 16. The most cited paper “Comparative study of using nanoparticles for enhanced oil recovery: Wettability alteration of carbonate rocks” [16], this publication has been cited 261 times as of February 12, 2025. Short summary of annotation: “Exposure of various nanofluids from zirconium dioxide (ZrO2), calcium carbonate (CaCO3), titanium dioxide (TiO2), silicon dioxide (SiO2), magnesium oxide (MgO), aluminium oxide (Al2O3), cerium oxide (CeO2) and carbon nanotubes (CNT) on the wettability of carbonate rocks were investigated for enhanced oil recovery from oil reservoirs. The results of spontaneous imbibition tests and core flooding experiments confirm the active role of CaCO3 and SiO2 nanoparticles in enhanced oil recovery. It is shown that both irreducible water saturation and inlet capillary pressure increased after treatment with CaCO3 nanofluid.”

Figure 4 shows a graph constructed under the same conditions as the previous one, but for the original author's keywords.

The lack of normalization of term spelling affected the clustering. This is particularly noticeable for the second cluster in this figure. The term methane hydrate is not reflected on it, and it is found in articles with high citations according to the previous figure. In the third cluster, the terms nanoparticles and nanofluids are present, but the term carbonates, as in the previous figure, is not present, and it is found in more recent publications. The first cluster in this figure does not contain the term deep learning, which is found in newer publications with good citations. If we refer to the interactive web page provided in the archive, we can see those two spellings “enhanced oil recovery” and “enhanced oil recovery (eor)” are presented in the graph. This example illustrates well the importance of pre-processing of text fields on the results obtained.

Co-occurrence network analysis of terms in the text of annotations

The VOSviewer program allows to build a co-occurrence network not only for author and index keywords, but also for terms present in annotation texts. VOSviewer generates a list of noun phrases and brings the terms to the singular number, more details can be found in the manual of the program.

This approach to defining key terms works well, but in the text of abstracts of scientific papers it is quite often found abbreviations of the most commonly used terms in the given research area, e.g. EOR → Enhanced Oil Recovery.

Therefore, the previously compiled list of abbreviations (thesaurus_terms.txt file) was used to construct a network of co-occurrence of terms in the annotation texts. It should be noted that VOSviewer performs whole term replacement, i.e. if in file thesaurus_terms.txt we specified that “eor → enhanced oil recovery”, the term “co2 eor process” will not be replaced, but the individual term “eor” will be replaced by “enhanced oil recovery”. Going beyond this example, it is possible to perform an additional iteration of term normalization by analyzing the terms used to build the network and expanding the list of replacements in thesaurus_terms.txt. Another way is to perform term substitution in the original annotation text (or titles).

A private observation of the author of this paper shows that the use of textual fields of titles and annotations is underestimated in many publications, the construction of a network of keywords is significantly more frequent. At the same time, the use of title and annotation fields to build a network of terms co-occurrence, in my opinion, gives more coherent clusters and their number is smaller.

The point is that subject matter experts assess the need for further study of an article not by its keywords, but by its title and abstract. The abstract and title fields in the dataset we studied were filled in completely, while the author's keywords had 1470 blank fields out of 8051 total fields.

Another example, exporting from OnePetro downloads titles and abstracts, but not keywords.

The situation is even worse for The Lens database,

on the query — Filters: Year Published = (2024 - 2024) Subject = (Geochemistry and Petrology) received Scholarly Works (12,482). In 12482 records a little more than half (6851) of the annotations are missing and almost all Keywords fields are empty (11911).

Another example is using the “Publish or Perish” program, which can query a large number of resources, including OpenAlex, and the saved files have Title and Abstract fields, but no keywords.

Also, RSS feeds from topical sites have header and body fields, but not keyword fields.

For example, on the old Springer site (link.springer.com/search), you can use RSS with the fields: title and description (which includes an abstract), but not keywords.

According to the text fields of the given examples, it is possible to build a network of term co-occurrence and evaluate the topicality of the collected materials using VOSviewer.

In accordance with the above, Figure 5 presents the network of terms co-currencies contained in the texts of the abstracts, taking into account the disclosure of abbreviations.

First Cluster (Red)

Most frequently appearing terms (Occurrences score): accuracy (499), stress (461), algorithm (384), horizontal well (326), hydraulic fracture (292), reservoir model (288), coal (280).

Terms most frequently used in new publications (Avg. pub. year score): marine hydrate reservoir (2023.1), extreme gradient (2022.6429), seepage field (2022.6), test set (2022.5455), extreme gradient boosting (2022.5), long short term memory (2022.4375), hbs (2022.4)

Here: hbs → hydrate-bearing sediment, the term HBS in annotations this term for the first time can meet with the decoding, and further used only as an abbreviation, it was not included in the list of exceptions, made from the author's keywords.

Terms most frequently used in highly cited publications (Avg. norm. citations score): rock strength (3.1761), slip length (2.8046), co2 ecbm process (2.5662), gas property (2.5647), ugs (2.4861), hydrate bearing layer (2.4357), water permeability (2.4206)

Here: ugs → underground gas storage (UGS), in annotations this term for the first time can meet with the decoding, and further used only as an abbreviation, it was not included in the list of exceptions, made from the author's keywords.

Here: ecbm → enhanced coalbed methane, this term is present in the dictionary of abbreviations compiled using the author's keywords, but it is part of a compound term, so it has not been modified. This feature should be taken into account when compiling the thesaurus_terms.txt file when performing term clustering based on title and abstract texts.

The abbreviations in the following clusters are similar and are not explained further.

Second cluster (green).

Most frequently appearing terms (Occurrences score): oil recovery (1269), enhanced oil recovery (878), concentration (670), crude oil (630), viscosity (560), wettability (422), interfacial tension (410).

Terms most frequently used in new publications (Avg. pub. year score): dispersion stability (2022.6667), promising alternative (2022.4), microfluidic experiment (2022.25), imbibition efficiency (2022.2308), oilwater interfacial tension (2022.1818), md simulation (2022.0833), experimental finding (2022). Here: md → molecular dynamics (MD).

Terms most frequently used in highly cited publications (Avg. norm. citations score): potential solution (2.9926), disjoining pressure (2.8678), viscoelastic property (2.5119), pam (2.4066), nanotechnology (2.1721), early breakthrough (2.1437), co2 foam (2.1364). Here: pam → polyacrylamide (PAM).

Third cluster (blue).

Most frequently appearing terms (Occurrences score): pore (749), content (487), sandstone (460), basin (450), hydrocarbon (434), dissolution (410), pore structure (348).

Terms most frequently used in new publications (Avg. pub. year score): development efficiency (2022.7857), shale oil resource (2022.7727), qingshankou formation (2022.3333), seepage capacity (2022.3214), basalt (2022.2308), movable oil (2022.1579), baikouquan formation (2022.125).

Terms most frequently used in highly cited publications (Avg. norm. citations score): macro pore (2.324), winland (2.2597), pore geometry (2.198), pore distribution (2.185), burial history (2.1773), ct scanning (2.1137), host rock (2.1036). Here: ct → computed tomography (CT).

Fourth cluster (khaki).

Most frequently appearing terms (Occurrences score): heavy oil reservoir (394), heavy oil (326), steam (246), oil recovery factor (197), oil viscosity (188), co2 flooding (158), solvent (154).

Terms most frequently used in new publications (Avg. pub. year score): underground hydrogen storage (2023.2609), hydrogen production (2023.1333), formation energy (2022.6429), nuclear magnetic resonance technology (2022.6), recovery degree (2022.5882), co2 storage efficiency (2022.4615), cushion gas (2022.4545).

Terms most frequently used in highly cited publications (Avg. norm. citations score): cushion gas (3.6855), underground hydrogen storage (3.5665), hydrogen storage (3.5626), hydrogen production (2.897), good potential (2.7012), co2 geological storage (2.317), residual trapping (2.2258).

Explanation: the term “underground hydrogen storage” is more common in new highly cited publications, but the publications themselves are still few, so this term is not reflected in Figure 6 and Figure 7, where the term occurrence restriction is used.

Conclusions

References

1. Guthrie, S. , Ghiga I., Wooding S. What do we know about grant peer review in the health sciences? // F1000Res. 2018. Vol. 6. P. 1335. [CrossRef]
2. Recio-Saucedo, A. , Crane K., Meadmore K., et al. What works for peer review and decision-making in research funding: a realist synthesis // Res Integr Peer Rev. 2022. Vol. 7, No 1. P. 2. [CrossRef]
3. Moran, R. , Butt J., Heller S., et al. Health research systems in change: the case of ‘Push the Pace’ in the National Institute for Health Research // Health Res Policy Sys. 2019. Vol. 17, No 1. P. 37. [CrossRef]
4. Van Eck, N.J. , Waltman L. Software survey: VOSviewer, a computer program for bibliometric mapping // Scientometrics. 2010. Vol. 84, No 2. P. 523–538. [CrossRef]
5. Ravi Kumar, N. , Kulkarni P., Kalaiarasai V. Unveiling the Viral Thread: A Comprehensive Analysis of Virality Coefficients in Indian Fashion Brand Dynamics // Advances in Marketing, Customer Relationship Management, and E-Services / ed. Tarnanidis T.K., Papachristou E., Karypidis M., et al. IGI Global, 2024. P. 180–193. [CrossRef]
6. Van Veldhoven, Z. , Etikala V., Goossens A., et al. A Scoping Review of the Digital Transformation Literature Using Scientometric Analysis // Bus. Inf. Sys. 2021. P. 281–292. [CrossRef]
7. Chigarev, B. A Proof-of-Concept Methodology for Identifying Topical Scientific Issues in New Publications Whose Citations Have Not Yet Been Established. 2024. [CrossRef]
8. Chigarev, B. Analyzing the Possibilities of Using the Scilit Platform to Identify Current Energy Efficiency and Conservation Issues. 2024. [CrossRef]
9. Handayani, M. , Sunarya M.H., Bahit M. Visualization of Fintech Research Trends (Financial Technology) Using VOSViewer // Proceedings of the 3rd Annual Management, Business and Economics Conference (AMBEC 2021) / ed. Eltivia N., Riwajanti N.I., Susilowati K.D.S. Dordrecht: Atlantis Press International BV, 2023. Vol. 224. P. 3–10. [CrossRef]
10. Chigarev, B. GSDMM Clustering Results Visualization Technique for Short Texts. 2024. [CrossRef]
11. Feldman, R. , Sanger J. The Text Mining Handbook: Advanced Approaches in Analyzing Unstructured Data. 1st ed. Cambridge University Press, 2006. [CrossRef]
12. Cammarano, A. , Varriale V., Michelino F., et al. Discovering technological opportunities of cutting-edge technologies: A methodology based on literature analysis and artificial neural network // Technological Forecasting and Social Change. 2024. Vol. 209. P. 123811. [CrossRef]
13. Gitelman, L. , Magaril E., Kozhevnikov M. Energy Security: New Threats and Solutions // Energies. 2023. Vol. 16, No 6. P. 2869. [CrossRef]
14. Hassan-Montero, Y. , De-Moya-Anegón F., Guerrero-Bote V.P. SCImago Graphica: a new tool for exploring and visually communicating data // EPI. 2022. P. e310502. [CrossRef]
15. Feng, Y. , Chen L., Suzuki A., et al. Enhancement of gas production from methane hydrate reservoirs by the combination of hydraulic fracturing and depressurization method // Energy Conversion and Management. 2019. Vol. 184. P. 194–204. [CrossRef]
16. Nazari Moghaddam, R. , Bahramian A., Fakhroueian Z., et al. Comparative Study of Using Nanoparticles for Enhanced Oil Recovery: Wettability Alteration of Carbonate Rocks // Energy Fuels. 2015. Vol. 29, No 4. P. 2111–2119. [CrossRef]
17. Wang, S. , Javadpour F., Feng Q. Molecular dynamics simulations of oil transport through inorganic nanopores in shale // Fuel. 2016. Vol. 171. P. 74–86. [CrossRef]
18. Waltman, L. , Van Eck N.J. A smart local moving algorithm for large-scale modularity-based community detection // Eur. Phys. J. B. 2013. Vol. 86, No 11. P. 471. [CrossRef]
19. Newman, M.E.J. Modularity and community structure in networks // Proc. Natl. Acad. Sci. U.S.A. 2006. Vol. 103, No 23. P. 8577–8582. [CrossRef]
20. Saxena, N. , Kumar A., Mandal A. Adsorption analysis of natural anionic surfactant for enhanced oil recovery: The role of mineralogy, salinity, alkalinity and nanoparticles // Journal of Petroleum Science and Engineering. 2019. Vol. 173. P. 1264–1283. [CrossRef]
21. Liu, W. , Wu Z., Li Y., et al. Experimental study on the gas phase permeability of methane hydrate-bearing clayey sediments // Journal of Natural Gas Science and Engineering. 2016. Vol. 36. P. 378–384. [CrossRef]
22. Zhong, R. , Johnson R., Chen Z. Generating pseudo density log from drilling and logging-while-drilling data using extreme gradient boosting (XGBoost) // International Journal of Coal Geology. 2020. Vol. 220. P. 103416. [CrossRef]