Personal management of personalized recommendations

Every enterprise search software vendor promotes the value of personalization, based on some mix of signals from previous clicked results and context information from what it assumes is your role and expertise. This is all neatly wrapped up in a Black Box labelled AI that seems to bypass all the good intentions of explainable AI (xAI). As a member of the Editorial Board of the International Journal of Information Management I receive a regular alert from Science Direct relating to the broad areas of information retrieval and information management. This has been a very useful service. Then a few months ago I used Science Direct to track down some references in Elsevier journals to the neurochemistry of odour detection in support of a monograph I am writing on the life and work of chemist and information scientist Malcolm Dyson. I now have all the information I need. The project is closed.

However, Science Direct seems not to have recognised that I have not clicked on an odour research paper for several months, and in any case the number of downloads was negligible in relation to those on IR and IM. The result is that I continue to get regular updates on odour research.

There are three issues worthy of attention.

  • What was the extent of the trigger to move to odour research? It would have been such a weak trigger among all the IR and IM literature I look at.
  • Given I have not looked at any papers on odour research for three months why does the system either delete the topic or at least ask if it is still relevant?
  • Why is there no option to report the issue or adjust the profile from my desktop. It seems that I am doomed to receive odour research for the rest of my career.

The wider concern is that in an enterprise employees have multiple roles, and especially in research might explore an issue on a very exploratory basis. They might be searching on behalf of someone else and not have a direct interest themselves. At the heart of the matter is the extent to which you know the basis on which personalized recommendations are pushed at you and can control the scope and frequency of the recommendations.

If this is not the case then why not?

Martin White