Study: Better music, more compelling news? Recommendation engines need more interaction with users

Published 06 Sep 2022

Reading time 3 min

“It’s time to move from algorithm exhaustion to collaboration”

Today, Solita and Alice Labs published an international study that takes a deep dive into the relationship between AI-based recommendation engines and users. The study was carried out in five countries and is part of the two-year Everyday AI project in which two large streaming services were involved, including Finnish Broadcasting Company Yle. The research results call into question the user’s current role as a passive recipient. The report recommends more creative interaction between human and machine by providing users with more varied interactivity.

The report by Alice Labs and Solita, agencies specialising in strategic consumer insight and digital service development, focuses particularly on the recommendation systems used in mobile and online news as well as music and video streaming. The study’s results are based on 60 in-depth interviews of streaming service users in the USA, China, the Netherlands, India, and Finland, conducted during 2020–2021. Alice Labs carried out the study in cooperation with the University of Helsinki’s Centre for Consumer Society Research, and it was funded by the Foundation for Economic Education.

Continued development of recommendation engines

Current AI-based recommendation systems aim, for the most part, to build their recommendations autonomously based on user past behaviour. Researchers, however, are keen to remind us that this may lead to incorrect conclusions, because both the context of use and user expectations change frequently.

This interview-based study presents four typical situations, in which current recommendation systems and their users end up at an impasse:

  1. A user whose wishes change suddenly is frustrated when recommendations don’t react to the change.
  2. A user who follows a particular topic with a keen interest is disappointed when their recommendations don’t develop as their knowledge grows.
  3. A user wants to learn more about an interesting topic but feels the recommendation system’s classification is limiting.
  4. A user wants to find different types of content from the usual and the system does not provide the correct type of assistance.

“Expectations on personal recommendations vary depending on the user’s current desires, awareness, and context of use. When the gap between the recommendations and the user’s own expectations grows too large, a tension forms between the user and the recommendation system. It is often very difficult for a fully autonomous system to solve this kind of tension,” says Alice Labs researcher Kirsi Hantula.

Artificial intelligence and recommendations: From algorithm exhaustion to collaboration

Instead of building fully comprehensive autonomous recommendation engines, researchers recommend providing more opportunities for users to influence the recommendations they get.

“These results call into question the current thinking in which users are passive recipients, and they point towards a new way of thinking about the creative interaction between recommendation and the user. With this new understanding, we will be able to build recommendations that answer human needs much more richly,” concludes Solita’s Insight Lead Antti Rannisto.

The report provides various suggestions on developing future recommendation systems. Download the full report.

Further information:

Alice Labs, Customer Insights Researcher Kirsi Hantula, tel. +358 45 670 5523 [email protected]

Solita, Insight Lead Antti Rannisto, tel. +358 40 504 7196 [email protected]

Embracing user unpredictability