![\"\"\/](\"https:\/\/research.atspotify.com\/wp-content\/themes\/spotify\/images\/icon.png\"){kind=icon}
\n <\/div>\nMarch 23, 2023<\/span> Published by Yu Liang, Aditya Ponnada, Paul Lamere, Nediyana Daskalova<\/p>\n<\/p><\/div>\n Recommender techniques usually look to customers\u2019 previous consumption to foretell what they might need subsequent. In apply, this strategy tends to work finest when what the person desires is just like what they’ve consumed not too long ago, and when it’s comparatively straightforward for that particular person to judge new objects. On the opposite hand, this strategy works much less nicely when what the person desires is one thing completely different and doubtlessly one thing that isn’t captured by their previous consumption [1]. This situation is made tougher nonetheless when the beneficial objects take a while to judge, as is the case for issues like films, books, and \u2014 the main target of this research \u2014 podcasts.\u00a0<\/p>\n So how would possibly recommender techniques adapt to deal with these shortcomings? In a brand new research, we discover an strategy that places customers in management by giving them the power to specify their listening targets and obtain suggestions designed to assist meet these targets. To accomplish that, we created an interactive analysis prototype, known as \u201cGoalPods,\u201d and evaluated it with avid podcast listeners, in search of solutions to 2 particular analysis questions:<\/p>\n RQ1: <\/strong>What (if any) targets do customers have for multimedia content material consumption and what do these targets entail?<\/p>\n RQ2:<\/strong> How does goal-focused multimedia consumption change the best way customers work together with customized suggestions?<\/p>\n We carried out our research in two phases. First, we carried out a large-scale survey to study in regards to the varied targets that listeners wish to accomplish by way of podcast listening (to reply RQ1). Second, we used the knowledge from the survey to construct an interactive prototype and assess how customers would go about setting targets that they want to obtain by listening to podcasts (to reply RQ2).<\/p>\n To tackle the primary query \u2014 RQ1: What (if any) targets do customers have for multimedia content material consumption and what do they entail?<\/strong> \u2014 our pattern of customers acquired an in-situ<\/em> survey on the Spotify platform. The survey was triggered when customers visited a podcast episode web page, to make sure that solely latest podcast listeners on the platform obtain it. The survey consisted of two questions asking about (1) customers\u2019 instant intent for listening to the podcast episode that triggered the survey and (2) any long-term targets that they want to obtain by listening to podcasts.<\/p>\n We collected responses from 68K premium account subscribers within the US. From the responses, we noticed that even when customers had a long-term aim (e.g., to study one thing new), they tended to as an alternative be listening to podcasts that met their instant wants (e.g. to entertain themselves). This highlights a spot between what customers aspire to realize and what they usually devour, pointing to a transparent alternative for recommender techniques to raised help with goal-focused consumption.<\/p>\n To reply our second analysis query \u2014 (RQ2)<\/strong>: How does goal-focused multimedia consumption change the best way customers work together with customized suggestions? <\/strong>\u2014 <\/strong>we included the learnings from the survey to assemble our prototype, GoalPods.\u00a0<\/p>\n The interactive prototype allowed customers to specify listening targets after which obtain customized podcast suggestions catered to these targets. The interface contained 4 key elements, visually separated into columns, corresponding to every of the next person actions:\u00a0<\/p>\n In the second a part of our research, we explored how customers interacted with the interactive prototype to devour goal-focused suggestions over the course of 1 week. After one week, customers have been interviewed to find out about their expertise of listening to the podcast episode from their aim playlists.<\/p>\n Our qualitative evaluation revealed 4 important themes.<\/p>\n Theme 1: Two kinds of person targets: low-involvement (e.g. \u201crelieve boredom\u201d) and high-involvement (e.g. \u201clearn something new\u201d) targets.\u00a0<\/em><\/strong><\/p>\n We recognized two sorts of targets that customers had for listening to podcasts. First, their high-involvement targets <\/strong>have been life targets that customers felt podcasts might assist them with, corresponding to self-improvement, information, and studying. They have been typically geared in the direction of private improvement and utilizing podcasts as a method to achieve information that may be utilized in the true world or shared with others. Low-involvement targets,<\/strong> then again, have been centered round instant necessities of leisure or catching up on pursuits. These targets typically concerned maintaining with the instances and protecting one-self entertained whereas doing different issues.\u00a0<\/p>\n Theme 2: Users want extra construction and help to set high-involvement targets<\/em><\/strong>.<\/strong><\/p>\n We discovered that customers discovered it straightforward to determine related suggestions for his or her low-involvement targets. They typically targeted on acquainted content material to lower the cognitive load required to decide on goal-relevant content material to take heed to. However, when it got here to high-involvement targets, customers wanted extra scaffolding and help. Since these targets require greater cognitive effort, recommender techniques might assist customers obtain them by decreasing the barrier to discovering brief cognitively-heavy podcasts.<\/p>\n Theme 3: Setting targets helped customers uncover content material exterior their filter bubbles<\/em><\/strong><\/p>\n By anchoring on customers\u2019 private targets to discover suggestions, the interactive prototype (and goal-focused podcast consumption) led to insightful content material discovery exterior the customers\u2019 filter bubbles. Overall, personalization was most helpful for the low-involvement targets because it helped determine acquainted and related content material to the person\u2019s listening historical past. However, within the context of the high-involvement targets, customers aimed to search out drastically completely different content material, and the present recommender system was not optimized for that use case.\u00a0<\/p>\n We discover that our device for goal-setting helped customers to find content material in two important methods: (1) it helped customers get into the mindset of being open to new content material and elicit their needs for branching out of their typical habits, and (2) the structured setup of the device helped floor novel content material in a handy manner, supporting the necessity for recommender techniques that drive discovery. As one participant summarized, \u201cit\u2019s kind of hard to know where to start. And so I think this feature gives you a good place to be like, okay, this is what I want to do for this week. And this is how I can do it. And here are some options for what I can listen to instead of just scrolling through a tab for like 10 minutes, trying to find one thing you might find interesting. So I think that was very helpful.\u201d<\/em><\/p>\n Theme 4: Contextual cues and descriptors have been vital to assist customers decide goal-relevant content material.<\/em><\/strong><\/p>\n Regardless of the cognitive effort required to observe by means of on a aim, we discovered some alerts that helped customers determine what content material they wish to take heed to. These alerts are vital cues for future recommender techniques for podcasts and different audio content material to deal with.<\/p>\n We developed GoalPods as a analysis prototype to probe customers on their podcast-related aim setting, suggestion analysis, and goal-focused episode consumption. While there aren’t any plans to combine this prototype into the Spotify expertise, some vital classes have been pulled to tell future product iterations. Our research highlights just a few vital design implications for recommender techniques, particularly the place the extent of funding on a part of the person is greater, corresponding to films and audiobooks, along with podcasts. We noticed that customers tended to plan extra high-involvement targets for the long run (corresponding to studying and gaining new information\/views), whereas their immediate listening intent was centered round low-involvement targets (corresponding to being entertained). This emphasizes the significance of understanding the person\u2019s underlying motives and aspirations for listening to a specific podcast along with specializing in previous consumption habits.\u00a0<\/p>\n![\"Exploring](\"https:\/\/storage.googleapis.com\/research-production\/1\/2023\/03\/RS045_Exploring-Goal-oriented-Podcast-Recommendations_finalnologo.png\")
Goal-setting in recommender techniques<\/h2>\n
Methodology<\/h2>\n
![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2023\/03\/image3-700x143.png\")
Podcast targets survey<\/h2>\n
GoalPods overview<\/h2>\n
![\"\"](\"https:\/\/storage.googleapis.com\/research-production\/1\/2023\/03\/image2-1-700x405.png\")
<\/figure>\n\n
\n
\n
Evaluation<\/h2>\n
Findings<\/h2>\n
What does this imply for future recommender techniques?<\/h2>\n