Describing relationships provides latitude for interpretation and innovation.
Specifying relationships also sets boundaries and helps prioritize elements (around the *centers*) inside the project.
Focusing on the relationships between elements, rather than rigidly defining their forms from the outset, leads to more adaptable, creative, and ultimately more successful designs.
Below is an example of the Spotify app and how it is today versus how it could be tomorrow should there have been a more generative approach to design.
**Approach 1: Form-Focused/Prescriptive (Current Spotify-like System)**
1. **Collect User Data:** Track explicit "likes," saved songs, followed artists, and completed playlists. Also track implicit data like play counts, skip rates, and listening time.
2. **Genre Classification:** Categorize songs into predefined genres and sub-genres using metadata and audio analysis.
3. **Similarity Calculation:** Calculate song similarity based on genre, tempo, key, and other audio features.
4. **Recommendation Generation:**
- **Collaborative Filtering:** Recommend songs liked by users with similar tastes (based on likes and listening history).
- **Content-Based Filtering:** Recommend songs similar to those the user has liked or listened to frequently.
5. **Playlist Creation:** Generate personalized playlists based on genre, mood, or activity (e.g., "Daily Mix," "Discover Weekly").
6. **Presentation:** Display recommendations as lists of songs or albums with album art, ordered by relevance or popularity.
**Key Characteristics of Approach 1:**
- **Static Preferences:** Relies on fixed signals like "likes" and past listening history.
- **Genre-Centric:** Heavily emphasizes genre classifications, which can be limiting.
- **One-Size-Fits-All Algorithms:** Uses general-purpose algorithms that don't fully capture individual nuances.
- **Limited Adaptability:** Struggles to adapt to evolving tastes and moods.
**Approach 2: Generative/Relationship-Focused (Proposed System with Dynamic Preferences)**
1. **Collect User Data:** In addition to the data collected in Approach 1, track:
- **Dynamic Preference Signals:** Implement "more/less" buttons or sliders with decay factors to capture changes in desired play frequency.
- **Contextual Data:** Gather information about the user's context (time of day, location, activity) to understand the situational relevance of music.
- **Emotional Response Data (Optional):** Explore using biofeedback (via, e.g., an Apple Watch) or other methods to infer emotional responses to music.
2. **Relationship Modeling:**
- **Dynamic Preference Modeling:** Model how user preferences change over time based on the "more/less" signals and decay.
- **Contextual Relationship Modeling:** Model the relationship between context and music preference.
- **Emotional Relationship Modeling (Optional):** Model the relationship between music features and emotional responses.
3. **Recommendation Generation:**
- **Dynamic Preference-Based Recommendations:** Prioritize songs that align with the user's current desired direction of change in listening habits.
- **Context-Aware Recommendations:** Recommend music that is appropriate for the user's current context.
- **Mood-Based Recommendations (Optional):** Recommend music that evokes desired emotional responses.
4. **Adaptive Playlist Creation:** Generate playlists that adapt to the user's changing preferences and context. The playlists themselves could even have "more/less" controls.
5. **Interactive Presentation:** Present recommendations more interactively and dynamically, perhaps using visualizations to represent mood, energy, or other relevant factors.
**Key Characteristics of Approach 2:**
- **Dynamic Preferences:** Uses dynamic signals to capture evolving tastes and moods.
- **Relationship-Centric:** Focuses on the relationships between music, user context, and desired listening experiences.
- **Personalized Algorithms:** Uses more sophisticated algorithms that adapt to individual user behavior.
- **High Adaptability:** Easily adapts to changing tastes, moods, and contexts.
Shifting from static "likes" to dynamic "more/less" signals with decay is a perfect example of "Specify Relationships Over Forms." It shifts the focus from the *form* of a fixed preference to the *relationship* between current and desired future listening behavior, leading to a more adaptive and personalized music recommendation experience. This directly addresses the UX problem of algorithmic ruts by capturing the dynamic nature of human taste and creating life (emergence and adjacent possible niches) in design.
For more on this kind of writing: