The importance of weak social ties in professional communities is well studied and widely accepted. In our paper we analyze the structure of strong ties based on the co-authorship relation and use the formal concept analysis framework to gure out weak ties. The research is motivated by fast growing need in cross-disciplinary research, which requires experts from di erent areas to understand the bigger picture and identify potential fellows for collaborative research projects in nearest future.
1.1
To keep up with state of the art developments in the Oil & Gag industry the engineers have to regularly examine specialized conferences organized under the umbrella of the Society of Petroleum Engineers (SPE)5. The built in mechanism for expert selection of materials for these conferences is designed to provide an appropriate level of knowledge, and thus eliminates the need to waste time on publications, which are not of top quality.
All the conferences in the eld are divided into regions to represent the regional development in the industry. Besides, the SPE has a rule explicitly stated in every call for papers, according to which the article may 5 SPE is a not-for-pro t professional organization for oil and natural gas exploration and production (E&P) professionals. It was founded in 1957, and today brings together more than 165,000 engineers, scientists, managers, and educators be submitted to only one conference. Therefore the authors of crossdisciplinary articles have to choose which of the specialized conferences to apply to.
Nowadays the easily accessible hydrocarbon resources have run out, so the industry is focused on hard-to-mine resources (brown elds), which requires an integrated approach and cross-functionality. Therefore, the number of cross-disciplinary articles grows from year to year.
As a result, it may happen for a cross-disciplinary work that the article falls out of focus. On the other hand there is a common theme for all conferences, such as those associated with machine learning and big data. For those interested in topics such specialists should either keep track of all conferences at once, or use automated search engines. 1.2
Our goal is to develop a methodology and tools for automated analysis
of a collection of research papers available at the SPE digital library. On
the basis of these analyzes one should be able to:
{ gure out the most important and relevant research topics,
{ assess the in uence of di erent researchers and scienti c schools,
{ identify strong and weak ties in the professional community,
and use all of these in daily research management process. This paper is
focused on the third item in the list. It continues our study of professional
communities started in [
The analysis of social networks of co-authorship has a long history [
The identi cation of weak ties within a professional community has a great practical importance. Firstly, identi cation of people who are working on the same topic and substantial research idea is very important for information gathering and knowledge di usion. Secondly, knowing the social environment, e.g. weak ties within the community can be important in collaboration and cooperation establishment. In this paper we aim to identify the strong and weak ties within the professionals of Oil & Gag industry based on their collaborations which can be inferred from their coauthorships. In this paper we assume that two researches have weak ties if they both work with the same objects or concepts and their research topics are very close to each other. 1.4
Formal concept analysis (FCA) gives a way to analyze collections of
objects and their properties. Recall some basic de nitions from [?]. A
formal context is a triple K = (G; M; I), where G is a set of objects, M
is a set of attributes, and I G M is a binary relation that expresses
which objects have which attributes. Implication A ! B for subsets A,
B of the set of attributes M (A; B M ) holds if A0 B0, i.e. every
object possessing each attribute from A also has each attribute from B.
An association rule is an expression of the form X ! Y , where X; Y M
and X0 Y 0 may not hold. The strength of an association rule can
be measured in terms of its support (denoted by supp) and con dence
(denoted by conf ), where
supp(X ! Y ) = j(X [ Y )0j ; conf (X ! Y ) = j(X [ Y )0j
jGj jX0j
Support determines how often a rule is applicable to a given data set,
while con dence determines how frequently items in Y appear in
transactions that contain X. See [
Then the association rules are interpreted as indicators of connectivity between di erent research elds, and also used to recognize weak ties between authors of di erent papers.
The idea to apply FCA in the context of social network analysis is not
new. In [
Our study is based on materials of annual SPE Russian Oil and Gas Conference and Exhibition 2016. The main features of this event are as follows: { Multi-disciplinary. The conference presentations, selected on the basic directions of development Oil & Gas industry. These areas are listed below. { Periodic. This is an annual conference. { Regional. The majority of the participants represented mainly the
Russian companies. { High selection criteria. The conference acceptance rate is approximately 15%. The selection process is conducted by Subject Matter Experts. { The conference program consists of four parallel sections. { At least one co-author must attend the event and present the work. The data we work on is retrieved from open portal of Society of Petroleum Engineers (SPE) at http://www.onepetro.org.
Clean up and preparation of meta information was produced using Python on hybrid cluster at Gazpromneft NTC LLC. Text analysis was done using Python NLTK library. Statistical analysis was performed using SciPy library. 2.1
The collection comprises 404 articles written by 839 co-authors. It includes papers in the following areas: 1. Well construction drilling and completion. 2. Static and dynamic modeling. 3. Hard-to-recover reserves. 4. Well and formation testing. 5. Field development monitoring and control. 6. Well intervention. 7. Shelf development experience and prospects. 8. Field geophysical survey/well logging. 9. Gas condensate and oil gas condensate eld development. 10. Brown elds. 11. Geomechanics. 12. Oil and gas production - equipment and technologies. 13. Cores recovery, examination and analysis 2.2
In the retrieved data each publication record includes the following information: { title and abstract of the article; { the list of authors and their a liations; { year of publication.
The most time-consuming step was to prepare the data and make the data set clean and useful. Unfortunately, the portal does not have a directory for authors. As a result sometimes we had up to 6 di erent spellings of the same name in di erent articles. 2.3
Almost every paper in the collection is written jointly by a few authors. It usually takes at least several months to write a good paper, so in the context of professional community each publication could be considered as a proof of strong ties between the co-authors.
The descriptive statistics for the co-authorship network is given below. { Number of nodes: 839 { Number of strong ties: 2315 { Number of connected components: 127 { Size of the largest connected component: 198 { Size of the second largest connected component: 20 2.4
The visualization of co-authorship networks is presented in Fig. 1. This
and the other graphs in this paper are produced with yEd Graph Editor
[
An inspection of the largest connected component shows that it mostly consists of participants of the well established collaborative program between Gazprom subsidiaries and Schlumberger. Otherwise the picture is very typical for a large industrial research conference, where the audience consists of big number of small cliques, which hardly communicate with each other. It shows that authors prefer working within their small community and it is di cult for them to establish new links with other colleagues.
As it was already mentioned above the goal of our work is to help the members of a professional community identify participants with similar interests and then convert weak ties into strong ones by establishing mutually bene cial collaborative research projects. 3 3.1
The importance of weak ties is well studied in the literature, see [
The heuristics is implemented in the following way. First, we start from extracting keywords for each paper in order to create a formal context, i.a. object-attribute relation in which objects are words, attributes are authors, and the relation is \a keyword w is used by an author a". Second,
the association rules with high characteristics of support and con
dentiality are computed using Concept Explorer tool, see [
As it was mentioned above our data set stores titles and abstracts of papers. As these texts are rather small we initially consider all words as equally important.
After the clean up the object-property table has 729 objects (keywords) and 839 attributes (authors). 3.3
we conclude that members with IDs 333, 754, 42, and 133 work on the close subjects as they use 14 common keywords, so each two of them are considered weakly tied.
In general, each rule of a form
s j a1; : : : ; an = c% ) s0 j b1; : : : ; bm produces Cn2+m pairwise weak ties within the union set fa1; : : : ; an; b1; : : : ; bmg. 4
For the data set of SPE papers the suggested procedure yielded the following. First, we have got 216 association rules with con dence greater than 80% and support at least 5 objects (keywords). Some of them are listed on Table 1. That resulted in 436 weak links out of which 149 were unique. Finally it turned out that the bigger part of them duplicates some of the existing strong ties and only 46 out of 149 suggest new connections. The network graph with the added weak ties is presented in Fig. 2 and in Fig. 3.
Brie y, most of the isolated islands are not a ected and remain isolated.Three cliques got connected to the largest component, see Fig. 2. Another two joined the second largest component, see Fig. 3. The fact that out of 149 identi ed weak ties 103 are duplicates of the already established strong ties shows that the suggested heuristic is rather conservative, two thirds of the found connections are certainly relevant. For the remaining new links we rely on expert opinion. To this end visualization in Fig. 4 was used together with the respective table of suggested candidate pairs for collaboration. In this paper we have used Formal Concept Analysis for the identi cation of weak ties in a social network of co-authorship. This task has a lot of applications, for example identifying colleagues with similar academic and professional interests and aims. The identi cation of people with similar interests can also signi cantly improve the mechanism of academic and professional recruiting.
We also believe that the current methodological approach can be reframed for the case of dynamic social networks and identi cation of weak ties formation and dissolution in a professional community. 5
The article was prepared within the framework of the Basic Research Program at the National Research University Higher School of Economics (HSE) and supported within the framework of a subsidy by the Russian Academic Excellence Project '5-100'. We are also thankful to the anonymous referees for their comments and suggestions, which signi cantly improved the text of this paper.