Search for expert - general idea

Owned by Henryk Rybinski
Jan 03, 2024
2 min read

The expert search system works according to the following algorithm:

A.      If there are certain filters applied to scientists:

  1. The system searches for scientists

  2. For the selected group of scientists, the system searches for achievements according to the

    1. specified domain criterion (keywords)

    2. user-specified types of achievements (publications, patents, projects, etc.).

    3. user-specified period in which the best experts are sought; if the period is not specified, the system can automatically narrow the range for scientific achievements from below, so that the total number of achievements does not exceed the N threshold set for the installation;

    4. the characteristics of the output: if for the criteria selected above the output includes more items than N, the system can automatically raise the threshold of "quality" of the publication output, increasing the threshold of the number of points of the Ministry of Science and Higher Education/MEiN;

  3. For each scientist and for the selected output, the system recalculates the points according to the adopted criterion;

  4. Having the score calculated, the system arranges the scientists according to the points obtained and displays K scientists in tiled form

B.       If there are no filters imposed on scientists, the system implements points (2-4) as above, except that the output is searched across the entire knowledge base (not limited to a subset of scientists).

In the above algorithm, the score for a single item o and domain query q is calculated according to the formula

S(q,o) = sim(q,o) × s(o)

where

  • sim(q,o) is calculated according to the formula sim(q,o) = 1 - e-rel(q,o) whereby rel(q,o) is a measure of the similarity of the object o to the query q, returned by SOLR

  • s(o) is the value of object o according to the selected expert ranking criterion.

In the case of multi-criteria selection, the unified measure is the modified MEiN scoring system, with a scoring normalization function for publications prior to 2019, so that the old scoring system is comparable to the new system.