PageRank-network of heterogeneous objects

This note is a development of the previous text, " Problems of generalization of PageRank ". The bottom line is to more or less fully rate people using the PageRank algorithm . Why choose PageRank? Well, of course, you can make up something like a sum of questionnaires from different terms and calculate it for each user. For example, education means so many points, higher education so much, office plankton position so much, top manager so many, academic degree there / no, work experience for so many years (we calculate the function of the number of years), rating on Habre is such and such, the number of friends on Facebook is so many, etc., etc. Not only will the list be long and it is not clear if you will take into account all the most significant factors. But it will also be necessary in some way (rather “by eye”) to determine the coefficients of significance for each term, and this is also a task. The PageRank method gives in my opinion a curious way to solve this last problem.

In the above text, I proposed to interpret the concept of voice in the PR method extensively. For example, having a degree will “vote” for its holder. How strong will this voice be? - It depends on how many authoritative people refer (in some way refer to) the degree as a significant factor. In other words, we must include an abstract “degree” object in our PR network along with human objects, and not only outgoing links, but also incoming links will come from it.

Or let's say work experience. Relatively speaking, work in Yandex and Rambler has a different significance in the eyes of the IT community, i.e. these two entities will have different values ​​of PR in this network (although traditional indicators of PR and TIC can also be defined as abstract rated entities and take into account the voices to them and from them). We associate with the concrete user the abstract essence of “experience”, to which links from Yandex or Rambler will go, if he worked there, and give greater or lesser importance to the experience of this user.

Thus, setting the task of rating people, we will at the same time actually rating objects of different types - organizations, content, titles and positions, etc.
')
In conclusion, I would like to think a little bit about how to solve another mentioned problem - taking into account all the most significant factors when determining the PeopleRank of a person. There are some general provisions about which I wrote earlier, for example, if you are cool / commented / read by cool people, then most likely you are also cool. Such information can in principle be extracted from social networks. We can suppose we also define the essence of the “projects” in which the user participated. Such things and especially specifics (such as entering specific projects, organizations) are probably better to entrust users with the web-dual way — if someone believes that in order to correctly define his PR, you need to add a new entity to the network, he adds it. Just as the sites are added to the network, which Yandex and Google identify.