ΕΕΛΛΑΚ - Λίστες Ταχυδρομείου

Re: Clarification on Mood Classification Approach

Dear Srilekha,

Thank you for the update and for sharing your approach. Using a local model
for mood detection makes sense, especially given the API constraints. The
*"facebook/bart-large-mnli"* model seems like a solid choice for emotion
classification.

For song recommendation, *RecBole, RecCobeats, and LightFM* each have their
strengths. If scalability and flexibility are key, *RecBole* might be a
strong candidate due to its broad support for different recommendation
algorithms. However, if we prioritize music-specific embeddings,
*RecCobeats* could be worth further exploration. We would advice to make a
decision now and maybe refine it when you make more progress in the actual
implementation.

Thank you!

Best,

Giannis Prokopiou & Thanos Aidinis

Στις Σάβ 29 Μαρ 2025 στις 5:56 π.μ., ο/η Srilekha <kslekha02 [ at ] gmail [ dot ] com>
έγραψε:

> Dear *Giannis Prokopiou *and *Thanos Aidinis*,
>
> I hope you both are doing well. I apologize for mailing you multiple
> times, but I wanted to share my thoughts on our current approach and seek
> your guidance. I’ve been working on the mood detection model and opted for
> a local approach instead of using an API. Considering the rate limits and
> potential costs associated with APIs—especially since we’re already using
> one and might need another for song recommendation—I felt that managing
> everything locally would be more efficient.
>
>   Currently, I'm using a model based on the *"**facebook/bart-large-mnli*
> *"* pipeline for mood detection. By mood, I mean identifying emotions
> like *sad*, *happy*, *angry*, etc from the user’s input. As for intent
> detection, we aim to classify whether the input is a *mood expression*, *direct
> song request*, or *mixed*. For instance, if a user says, *"I’m feeling
> low, play something upbeat,"* it would be classified as *mixed*, while *"Play
> something by The Weeknd"* would be a *direct request*.
>
>   For the song recommendation model, I’m currently exploring options like *RecBole,
> Reccobeats* and *LightFM*. Since scalability and data availability are
> key considerations, I would greatly value your input on which of these
> models—or any other alternative—you think would be more suitable for our
> use case. I’m confident that I can do more research on this, but I’ve
> already spent significant time on this, and I don’t want to waste any more
> if it’s not the right path altogether.
>
> Your thoughts and suggestions would really help us make informed
> decisions. Thank you once again for your guidance and support!
>
> Best regards,
> Sai Sri Lekha
> kslekha02 [ at ] gmail [ dot ] com
> ----
> Λαμβάνετε αυτό το μήνυμα απο την λίστα: Λίστα αλληλογραφίας και συζητήσεων
> που απευθύνεται σε φοιτητές developers \& mentors έργων του Google Summer
> of Code - A discussion list for student developers and mentors of Google
> Summer of Code projects.,
> https://lists.ellak.gr/gsoc-developers/listinfo.html
> Μπορείτε να απεγγραφείτε από τη λίστα στέλνοντας κενό μήνυμα ηλ.
> ταχυδρομείου στη διεύθυνση <gsoc-developers+unsubscribe [ at ] ellak [ dot ] gr>.
>
----
Λαμβάνετε αυτό το μήνυμα απο την λίστα: Λίστα αλληλογραφίας και συζητήσεων που απευθύνεται σε φοιτητές developers \& mentors έργων του Google Summer of Code - A discussion list for student developers and mentors of Google Summer of Code projects.,
https://lists.ellak.gr/gsoc-developers/listinfo.html
Μπορείτε να απεγγραφείτε από τη λίστα στέλνοντας κενό μήνυμα ηλ. ταχυδρομείου στη διεύθυνση <gsoc-developers+unsubscribe [ at ] ellak [ dot ] gr>.

πλοήγηση μηνυμάτων