Text based depression analysis

Source code for the paper Text-based Depression Detection: What Triggers An Alert

Prequisites

The required python packages can be found in requirements.txt.

torch==1.2.0
kaldi_io==0.9.1
bert_serving_server==1.9.6
pytorch_ignite==0.2.0
numpy==1.16.4
librosa==0.7.0
tabulate==0.8.3
mistletoe==0.7.2
scipy==1.3.0
tqdm==4.32.2
pandas==0.24.2
fire==0.1.3
imbalanced_learn==0.5.0
allennlp==0.8.5
gensim==3.8.0
ignite==1.1.0
imblearn==0.0
nltk==3.4.5
plotnine==0.6.0
scikit_learn==0.21.3
PyYAML==5.1.2

Pretrained models

Pretrained models from the paper are available at The release page.

Dataset

The dataset is freely available for research purposes and can be found here.

Preprocessing

The preprocessing feature extraction scripts can be found in data_preprocess/.

We exclusively use kaldi as our data processing pipeline ( except for extraction ).

Word2Vec features require manual training. For this purpose, one should create some all_transcripts.csv file with the same format as all other *TRANSCRIPTS.csv files.

The easies way to do that is just use bash or pandas (python) to merge all files into one, removing all headers except the first one e.g.,

cd labels_processed;
python -c "from glob import glob as g; import pandas as pd; pd.concat([pd.read_csv(f,sep='\t') for f in g('*TRANS*.csv')]).to_csv('all_transcripts.csv', index=False, sep='\t')"

Then just run the extract_text_gensim.py script, which extracts 100dim features into the features/text folder.

ELMo features can be rather directly downloaded and pretrained given one installed allennlp. However, BERT embeddings require the download of a pretrained BERT. In order to have easy access to BERT embeddings we use bert serving client. Here we prepared a simple script start_bert_server.sh in order to start the serving client. We recommend the use of screen or tmux.

In order to extract the features, prepare a folder named labels and softlink the train_split_Depression_AVEC2017.csv and dev files into it.

Further prepare the transcripts for feature extraction by directly copying all transcripts into a dir called labels_processed. In other cases pass --transcriptdir OUTDIRNAME to any of the extraction scripts.

Running the code

The main script of this repo is run.py.

The code is centered around the config files placed at config/. Each parameter in these files can be modified for each run using google-fire e.g., if one opts to run a different model, just pass --model GRU.

run.py the following options ( ran as python run.py OPTION):

• train: Trains a model given a config file (default is config/text_lstm_deep.yaml).
• evaluate: Evaluates a given trained model directory. Just pass the result of train to it.
• evaluates: Same as evaluate but runs multiple passed directories e.g., passed as glob (experiment/*/*/*), and returns an outputfile as well as a table report of the results. Useful for multiple runs with different seeds.

Notes

Due to the limited amount of training data, reproducing the results might be problematic. In our work we specifically pointed out that the proposed results are chosen as the peak over multiple runs.