🎶 WanderingMelody

Our recommendation system uses a multi-layered rating approach to rank songs:

• Topic Match Score (via SVD)
  Measures how closely a song aligns with the user's input in terms of general topics.

  We use Singular Value Decomposition (SVD), a dimensionality reduction technique, to uncover deeper themes in both user descriptions and song lyrics. SVD breaks down a large matrix (like a bag-of-words model) into a smaller set of latent features—abstract "topics" that capture patterns in word usage. The user's input is vectorized and projected into this reduced space, and we then compute how close this vector is to each song’s topic vector. This gives us a topic match score that reflects the thematic similarity between the user’s input and each song.

• Word Match Score
  Counts how many meaningful words from the user input appear in a song’s lyrics, using lemmatization and stopword filtering.
• Sentiment Match (Age + Weather-based)
  Adjusts the rating based on how emotionally “bright” or “dark” a song’s lyrics are, in relation to the user’s age and optionally, current weather sentiment. Songs are rated higher when their emotional tone aligns positively with these factors.
• Antonym Penalty
  Lowers the rating if a song strongly features antonyms of the user's input words, helping to avoid emotional or thematic mismatches.

Each component contributes to a final combined score. Songs are ranked based on this overall score to ensure the best emotional and thematic fit.

Additional, every song has a popularity score from 0 to 100, based on recent streaming activity. This metric does not significantly affect recommendations, but helps highlight currently trending songs on the platform.