Predicting User Intents and Musical Attributes from Music Discovery Conversations

Here is the implementation code and data for the paper titled "Predicting User Intents and Musical Attributes from Music Discovery Conversations" by Daeyong Kwon, SeungHeon Doh, Juhan Nam, 2024

Figure 1: Examples of user intents and musical attributes classifcation

Setting

The packages and version information required for the implementation are stored in the requirements.txt file.

Implementation

Sparse representation, Word Embedding (Word2Vec), DistilBERT_Probing, DistilBERT_Finetune, and Llama are each implemented in separate .py files. The functions.py file should be placed in the same directory to import functions.

Each .py file can be executed by running python filename.py, and the resulting .csv files will be saved in the "./results" directory.

For the concatenated setting, you can use concat_history function in functions.py.

Open Source Material

• Models
• Dataset
• Huggingface Demo