yzzhang2008 / mgid-dataset Goto Github PK

To support the development of social meida based multi-modal depression and emotion joint recognition, we collect textual and visual documents from Getty Image, and create a weakly labeled multi-modal depression and emotion dataset, called MGID. We we set a list of keywords with strongly depression (e.g., "depressive patient", "depression", etc.) and emotions ("happy", "scary", etc.) to query Getty Images, and use the labels of these words to weakly label the retrieved multi-modal documents of the first thirty pages. Based on these documents, we construct a new multi-modal Getty Image depression and emotion (MGID) dataset. It contains 2500 depressive and 2500 non-depressive samples. As for emotion, we divide MGID into SIX emotions, i.e., happiness, depressed, neutral, fear, surprise (including positive and negative feelings), anger. The dataset also provides train/dev/test sub-datasets.

All publications reporting on research using this database have to acknowledge this by citing the following article:
To be waited...

*** As for text, the title and textual content has been encapsulated into .zip file. One can access it directionly.
*** As for raw image, since Getty Image is a commercial platform, we will not upload the raw image directionly. Instead, we choose to upload its image URL, and provide the imageDownloader.py to download all raw images in MGID.

MGID is made available for research purposes only. Any commercial use of this data is forbidden.

The user may not distribute the database or parts of it to any third party.

The use of data for illustrative purposes in publications is allowed. Publications include both scientific papers and presentations for scientific/educational purposes.

The author is allowed to change these terms of use at any time.

We open it just for the development of depression detection. The dataset comes without any warranty. In no event shall the provider be held responsible for any loss or damage caused by the use of this data.