Categories
"Various"

The meaning of Research Interest in ResearchGate

Abstract. There are over 20 million users in ReseachGate (RG). ReseachGate uses internal metrics, RI and Total Research Interest (TRI), to measure how the authors’ peers assess their work. Formulas used by the RG team for those metrics are not published. This article proposes a simple model of calculating Research Interest (RI) and getting results close to the actual RIs on the RG site.    

How to cite this article – Victor Torvich, The meaning of Research Interest (RI) in ResearchGate, https://vtorvich.com/meaning-of-RI-in-RG.

Introduction

The meaning and formula for RI are of interest to members of RG. It is also the main topic for some articles [1,2].

     There are multiple parameters in RG, which are probably, taken into account by the RG team while calculating RI. For example, there are six types of Reads, five types of Recommendations, and so on.

I would not re-engineer the original formulas used by the RG team. Even if we could, we would be lost in those details. My goal is to find out as simple and transparent formula for RI as possible. That RI formula should also use as few input parameters as possible. Preferred input parameters are the ones that could be seen by any RG member and not only by a specific author. I would not rely on published by RG team [1] weights of different parameters.

     That should provide the meaning of RI based on the big picture.

Terminology

Total Research Interest (TRI) is the sum of the RI for each research item an author has added to their profile. The RI and TRI score is how the RG team estimates scientists’ interest in RG member’s research.

     All further terms are also related to whatever happens within the RG community.

     Publication (P) is not any publication by an author, but just a research item, a.k.a. “publication,” which is included in the author’s profile on RG. The RG member could have up to ten types of research items in their profile on the RG site.

Citation (Cit) of publication is a citation of mentioned P. Publication Read (PR) is a read of mentioned P. Publication’s Recommendation (RecP) is a Recommendation by a member of RG made for mentioned P.

Parameters in ReseachGate

There are three main parameters in RG, which measure the impact of an author’s research on the RG community: Cit, PR and RecP.

Per the RG team, “when researchers read, recommend or cite a research item, its Research Interest goes up.” [1]

Let us take a look at how that PR, RecP, and Cit parameters correlate with RI.

I’m using data from my personal RG account and data from accounts of other members of RG. All those data, i. e. Cit, PR, RecP, and RI, are publicly available and retrievable by anybody on RG site. I will not mention which data belong to which member of RG.

Case A: RI without Citations

RI metrics are all about the author’s publications and reactions to those publications.

My formula is, probably, different from the unpublished formula used by the RG team.

    First, let us consider a simplified case with the positive reaction to publication – when Cit=0, but RelP/PR >=0.5 (No Citations, but, at least, one Recommendation per two Reads of the publication). Raw data with 18 data points are in Table 1. The sums of publication Reads (PR) and publication Recommendations (RecP) are also there. An important ratio of Recommendations per Read (RecP/PR) is in this table too.

Table 1. RI as a function of a sum of publications’ Reads and Recommendations (PR+RecP). Cit = 0. RecP/PR >= 0.5. Data retrieved on September 2 – 5, 2021

Presented data cover a wide range of parameters that occurred in actual accounts on the RG site. RI range is 0.3 – 249.9; PR range is 1 – 789; RecP range is 1 – 842; RecP/PR range is 0.5 – 4.09; (PR+RecP) range is 3 – 1631.

     I plotted the graph for RI as a function of (PR+RecP). RI = f(PR+RecP).

Figure 1. RI as a function of a sum of publications’ Reads and Recommendations (PR+RecP). Cit = 0. RecP/PR >= 0.5. Data retrieved on September 2 – 5, 2021

In the absence of Citations and with a high value of RecP/PR (greater or equal to 0.5), the RI is a linear function of (PR+RecP). The formula for expected RI is this: Expected RI = 0.1581 * (PR+RecP) + 5.2377).

    R-squared, a statistical measure of how close the data are to the fitted regression line, is very high – 0.9791.

   For example, according to my formula, the expected, i. e. fitted to the plotted line, RI = 50 would be at (PR+RecP) = 283. The expected RI = 100 would be at (PR+RecP) = 599.

Case B: RI with Citations but low Recommendations

Let us consider how to calculate RI when Cit >0 and low RecP (RecP = 0 or 1) and low PR < 200.

    Raw data with 19 data points are in Table 2.

Table 2. RI as a function of Citations (Cit). RecP = 0 or 1. Data retrieved on September 2 – 14, 2021

     The range of presented data, in this case, is as follows. Cit range is 1 – 133. RI range is 0.6 – 67.4; PR range is 10 – 198; RecP range is 0 – 1; RecP/PR range is 0 -0.03.

   The graph with RI = f(Cit) is in Figure 2.

Figure 2. RI as a function of a sum of Citations (Cit). Cit >=1. RecP= 0 or 1. Data retrieved on September 2 – 14, 2021

The value of (PR + RecP) in Table 2 is relatively low, i.e., less than 200, and, in most cases, less than 100.

With a low impact of RecP and PR, we have the RI as a linear function of (Cit). The formula for expected RI is this: Expected, i. e. fitted to the plotted line, RI = 0.5009 * Cit + 0.8898. R-squared is very high – 0.9946.

According to the above formula, in Case B with Citations, the expected RI = 50 would be at Cit = 98.

Now we could compare the expected RI when we have only (PR+RecP), in Table 1, and when we have Cit > 0, and low RecP, i.e., RecP = 0 – 1, in Table 2. In Case A, with no Citations and with wide range of (PR+RecP), we expect RI = 50 at (PR+RecP) = 283. A simple calculation gives us 283 / 98 = 2.9.  

We conclude that the weight of three (Reads + Recommendations) is the same as the weight of one Citation. 3*(PR+RecP) = 1 Cit.

Case C: General case with RecP/PR > 0.01

Based on Cases A and B, I would assume that the equivalence of one Citation to three (Reads + Recommendations) is used all the time.

     That allows us to calculate RI, taking into account Citations, Reads, and Recommendations of publications. I combined Tables 1, Table 2 and more raw data.  

     The number of used data points, in this case, is 54. Presented data covers the following range. RI range is 0.3 – 1930.8; Cit range is 0 – 120. PR range is 1 – 6893; RecP range is 1 – 5934; RecP/PR range is 0.014 – 5.91; (PR+RecP) range is 3 – 12827.

      Now RI could be presented in more generic form as a function of (Cit + (RecP+PR)/3).

    The graph with RI = f(Cit + (RecP+PR)/3) with RecP/PR > 0.01 is on Figure 3.

Figure 3. RI as a function of (Cit+(RecP+PR)/3)). 54 data points.
RecP/PR > 0.01. Cit >= 0. Data retrieved on September 2 – 14, 2021

The meaning of RI is that, when RecP/PR >= 0.01, RI is a linear function of (Cit+(RecP+PR)/3). The formula for expected RI is this: Expected, i. e. fitted to the plotted line, RI = 0.4375 * (Cit+(RecP+PR)/3) – 4.4878.

R-squared is very high – 0.9909.

     According to this formula, the expected RI = 50 would be at (Cit+(RecP+PR)/3) = 104. The expected RI = 100 would be at (Cit+(RecP+PR)/3) = 218.

RI case D with low ratio RecP/PR

Cases when Cit = 0 and the ratio RecP/PR is low, or even 0, are special cases and should be considered separately.

     Let’s take a look at RI with Cit = 0 and RecP/PR < 0.01.

Table 3. RI as a function of (RecP+PR). RecP/PR< 0.01. Cit = 0.
Data retrieved on September 2 – 5, 2021

Table 3 is based on 11 data points. The range of presented data is described here. RI range 0.6 – 7.4; PR range 22 – 5946; RecP range 0 – 4.

This case is the case of low RecP/PR – less than one Recommendation per 200 Reads. Sometimes it is even just one Recommendation per thousand Reads. It is no wonder that the RG team decided to assign a much lower RI in Case D compared to Case A.

Figure 4. RI as a function of (RecP+PR). RecP/PR < 0.01. Cit = 0.
Data retrieved on September 2 – 5, 2021

The curve is logarithmic. The formula for expected TRI is this: Expected, i. e. fitted to the plotted line, RI = 1.1817*ln(RecP+PR) – 3.9515.

R-squared is high, 0.6601, but not very high. More raw data are needed to improve the R-squared value.

In Case A, with RecP/PR > 0.01, you would have RI = 6 with RecP+PR = 4.8. In Case C, you need to have RecP+PR = 4500 to get RI = 6. This dramatic dumping of RI value is a consequence of the low assessment of publications by researchers in Case C.

Relationship between RI and TRI

TRI is calculated by RG algorithm by summing up RIs for all research items in author’s profile on RG site. Or it could be calculated by summing up weekly RI additions across all research items in author’s profile.

     Complications arise during this summation process. The reason is that we have two RI formulas: logarithmic for cases with the low assessment of publications by researchers, and linear for cases with not very low assessment.

     It is very common for authors to have some publications with high appreciation by members of RG community and some with a low appreciation. Therefore, TRI would be a sum of some RIs with linear formula and some RIs with logarithmic formula.

     The more you have research items with Cit = 0 and RecP/PR< 0.01, the more TRI would skew towards lower value.

Summary for RI formulas

In this work, I presented the model, which allows making sense of RI. I aimed to get my own formulas, which could provide an estimated RI close to an actual RI on the RG site in most situations.

     RI could be estimated using only three parameters related to the author’s publications in their profile on the RG site. Those three parameters are Reads (PR), Citations (Cit), and Recommendations (RecP) of the mentioned publications. My formulas are based on input provided by Citations and integrated parameter, Reads plus Recommendations (PR+RecP).

RI score is a weighted index. Publication citation weight three times more than publication read and recommendation combined. RI score is a composite index with two different formulas for two domains of RI applicability.

RI is a linear function of (Cit+(RecP+PR)/3) when RG community valuation of author’s research is not too low, i. e. RecP/PR >= 0.01.

     The formulas presented in this work are based on raw data from the RG site with an extremely wide range of values. My formulas are approximate ones, making sense of the RI score and calculating an estimated RI if needed. The formulas themselves and domains boundaries could be fine-tuned with more raw data.

References

  1. RG Hep Center, What is Research Interest?, https://explore.researchgate.net/display/support/Research+Interest
  2. Sergio Copiello, Research Interest: another undisclosed (and redundant) algorithm by ResearchGate, Scientometrics, Volume 120, Issue 1, July 2019, pp 351–360, https://doi.org/10.1007/s11192-019-03124-w

Go to Comments on this blog post.

Go to the Directory of Blog Posts.

*** Switch to Sign-Up page! ***

By victortorvich

I'm the author of the book “Subsurface History of Humanity: Direction of History”. It is available on Amazon marketplaces and on audible.com.

Leave a Reply