AAAI 2024 Workshop W24: ML4CHM 2024

ConversationMoC: Encoding Conversational Dynamics using Multiplex Network for Identifying Moment of Change in Mood and Mental Health Classification

This work introduces a unique conversation-level dataset and investigates the impact of conversational context in detecting Moments of Change (MoC) in individual emotions and classifying Mental Health (MH) topics in discourse. In this study, we differentiate between analyzing individual posts and studying entire conversations, using sequential and graph-based models to encode the complex conversation dynamics. Further, we incorporate emotion and sentiment dynamics with social interactions using a graph multiplex model driven by Graph Convolution Networks (GCN). Comparative evaluations consistently highlight the enhanced performance of the multiplex network, especially when combining reply, emotion, and sentiment network layers. This underscores the importance of understanding the intricate interplay between social interactions, emotional expressions, and sentiment patterns in conversations, especially within online mental health discussions.

This work was supported by the Natural Environment Research Council (NE/S015604/1), the Economic and Social Research Council (ES/V011278/1) and the Engineering and Physical Sciences Research Council (EP/V00784X/1). The authors acknowledge the use of the IRIDIS High Performance Computing Facility, and associated support services at the University of Southampton, in the completion of this work.

Paper

Loitongbam Singh, Stuart E. Middleton, Tayyaba Azim, Elena Nichele, Pinyi Lyu, Santiago De Ossorno Garcia. ConversationMoC: Encoding Conversational Dynamics using Multiplex Network for Identifying Moment of Change in Mood and Mental Health Classification.

@inproceedings{singh-et-al_2024,
    title = "ConversationMoC: Encoding Conversational Dynamics using Multiplex Network for Identifying Moment of Change in Mood and Mental Health Classification",
    author = "Singh, Loitongbam Gyanendro and
      Middleton, Stuart E. and Azim, Tayyaba  and Nichele, Elena and Lyu, Pinyi and  Garcia, Santiago De Ossorno",
    booktitle = "Proceedings of the Machine Learning for Cognitive and Mental Health Workshop (ML4CMH)@AAAI 2024",
    month = Feb,
    year = "2024",
    address = "Vancouver, Canada",
}

Proposed pipeline

Multiplex network encoding pipeline:

Data Set:

Dataset can be downloaded from Zendo.

Software:

• © Copyright University of Southampton, 2023, Highfield, University Road, Southampton SO17 1BJ.
• Created By : Gyanendro Loitongbam
• Created Date : 2023/06/20

Installation Requirements Under Ubuntu 20.04LTS

• The experiments were run on Dell Precision 5820 Tower Workstation with Nvidia Quadro RTX 6000 24 GB GPU using Nvidia CUDA Toolkit 11.7 and Ubunti 20.04 LTS.
• Install the following pre-requisite Python3.8 libraries:

pip install transformers==4.20.1
pip install tensorflow==2.9.1
pip install keras==2.9.0
pip install networkx
pip install stellargraph==1.2.1
pip install sklearn==1.1.1
pip install sentence-transformers==2.2.0
pip install gensim==4.0.1

Pretrained Models

• Download fastText embedding vectors

Preparing Data set

For a conversation idx prepare the dataset to train in the following formats:

Post embedding and Training model

# Post embedding [avg. fastText word emb + SBERT + Taskspecific-scores]
python Post_embedding.py --source_dir work_dir --wv_model wiki-news-300d-1M.vec

# Training model 
python Training-all-model-multitask-kfold-v3.py --source_dir work_dir --out_dir model-kfold-v3-out --shuffle_seed 42 --epochs 40 --batch_size 64

## Training with Focal Loss Function version
python Training-all-model-multitask-kfold-v3-FLF.py --source_dir work_dir --out_dir model-kfold-v3-out --shuffle_seed 42 --epochs 40 --batch_size 64