Personalized Recommendations: Users’ Expectations and Assumptions

Twenty years ago, Jakob Nielsen claimed that personalization was overrated because technology hadn’t advanced enough to predict what users would be interested in. However, today, as he predicted, personalization is a growing trend on the web. In contrast to role-based personalization, individual customization refers to adjusting content and features specifically for a user based on their preferences and behaviors.

Personalized user interfaces can include various elements, from command shortcuts to color schemes. Among them, the recommendation feature is particularly useful on websites and mobile apps. Recommended products or content items can be commonly seen on e-commerce sites, social media, news sites, streaming videos, music services, and more. For instance, services like YouTube offer millions of videos, but a specific user may only see a few recommended videos and channels based on their browsing patterns and recorded data after visiting the site’s homepage or watching a video.

Individualized recommendations can be based on machine learning or other artificial intelligence technologies, explicit user-defined instructions, or a combination of both.

To understand users’ perspectives and meta-models of personalized recommendations, we conducted remote usability research with eight participants located within the United States. In each session, participants performed tasks assigned by the facilitator on 2-3 websites where they had accounts and answered recommendation-related questions during the interviews.

Our research participants were sensitive to the fact that sites generally present personalized recommendations based on their browsing patterns, purchase history, and other data sources. Overall, these recommendations were well-received and perceived as helpful in narrowing down various options provided within the site. To achieve this, users were willing to give up some personal information, and they expected that much of their behavior would be tracked and analyzed.

Selection through Recommendations

Participants overwhelmingly responded positively to recommendations. They found that the suggested items matching their interests helped them avoid choice overload and quickly find items of interest.

“Recommendations don’t overwhelm you with a lot of products. Instead, they show you things that you might be interested in, and you can choose among them. I think that’s pretty neat.”

“I’m always on Sephora. I use the app, mobile website, and computer website. I think Google is probably creating personalized ads for me based on everything I’ve seen. If Sephora did that, it would be really cool. … It would look like it’s personalized for me, and that would be really cool. It would save a lot of scrolling.”

“I really appreciate a way that helps me discover things that I wouldn’t have discovered myself. … I don’t have the patience to listen to everything to find various things I might not enjoy. So if Spotify could do that for me, that would be great. So I tried Stitch Fix for that reason. I like shopping a little bit, but I don’t like, ‘Oh, I need black pants, and I need to look at all these black pants and find the one that fits me.’ I don’t like doing that. Instead, answering some questions and having Stitch Fix say, ‘This is a good pair of pants. You would enjoy this, and it would look good on you.’ That would substitute some of the finding work for me and provide me with ‘Here are things you might like. Try them.’”

Perceived Data Collection and Privacy Expectations

When asked about the information used to generate recommendations, participants had fairly clear thoughts. The following are data types mentioned as possible sources, with responses from participants with the most confidence listed first:

  • History: Past purchase records (on e-commerce sites) or consumed content (e.g., watched videos, songs played on a music service)
  • User-input profile data: Demographic information like age, gender, location, as well as domain-specific interests or other details
  • Ratings
  • Saved or “liked” items
  • Browsing behaviors
  • Search history

“I’m not really sure how this is being chosen. Because it all looks pretty appropriate, I would assume it’s based on my shopping habits: what I’ve clicked on, what I like, what I’ve bought.”

Some users acknowledged that personalized recommendations could benefit them even without making purchases, but most preferred direct activities like past purchases, saved items, or profile information to have more weight than simple browsing.

“Sometimes when I click on something, I think it’s going to be good, but then I’ll read a review or … I don’t think I would be interested in something like a glitter mask. But I would like to see things I bookmarked and more of the brands I’ve purchased. I’d like to be recommended that kind of brand and type of product.”

Users assumed that some recommendations are based not only on their direct or indirect activities but also on algorithms that identify similar users or items of interest.

“The algorithm probably has something like, ‘If you like this black checkered button-up shirt, you would also like other black checkered or other button-up shirts.’”

“They’re probably using algorithms based on a lot of different data. I assume it works like how Netflix works and is based on what I’ve watched. I also assume it’s based on non-personal factors like my age, gender, and so on.”

In our study, no participant expressed excessive concerns about privacy (perhaps because of the harmless nature of the tested sites). They perceived tracking site usage as a reality and accepted it as part of the status quo while using the internet and living in the modern world. The benefits of personalized content were seen as more significant than privacy concerns.

“I don’t care because it’s cosmetics.”

“I’m a little worried about it, but I accept myself as being used to it since I don’t expect to quit using the internet anytime soon. It’s just how it is.”

“I think it’s kind of cool. Some people might be scared about something trying to capture their next step. But they’ve always been researching that and showing how it works with everything else. I think it’s really cool and shows that the company is trying to offer the best service.”

Judging General vs. Personalized Content

As personalized content becomes more prevalent, how do

people determine if content is tailored to them? Naturally, our research participants primarily relied on explicit titles (e.g., “Recommendations for You,” “Because You Watched”) as indicators. Apart from these explicit cues, indirect clues such as overall popularity or promotional potential of the recommended items were also used to judge whether the content was personalized.

The structure of the service played a significant role in this judgment. Users of Hello Fresh meal service didn’t believe that the company provided them with personalized recipes because they thought it would require too much effort on the company’s part.

Users browsing Netflix questioned whether shows promoted on the top banner area were tailored to them. After some consideration, they concluded that Netflix likely presented those shows to everyone for clear business reasons.

“When I go to my home screen, ‘Orange is the New Black’ is the first thing that comes up. I’m not sure if that shows up for everyone or not. … Maybe it’s just because a new season just came out and I’m one of the first ones to see it. So I think what shows up first is probably the same for everyone.”

“I guess it looks like how much I’ve watched, checking how many people have watched reviews on popular shows. It doesn’t seem personalized to me.”

In conclusion, personalized recommendations can meet users’ expectations and assumptions. Users appreciate the recommendation feature, and personalized content helps them avoid choice overload and find items of interest more easily. Personalized recommendations can be based on various data sources such as past purchase records, user-input profile data, saved items, and browsing behaviors. Users expressed some privacy concerns, but the benefits of personalized content were perceived as more significant. Judging whether recommendations are personalized or not relies on explicit cues and indirect clues, with the service’s structure and promotional strategy also influencing the judgment.

Source: https://www.nngroup.com/articles/recommendation-expectations/