Hello everyone, my name is Benjamin, and John from the David Allen Co. suggested I post here. I'm a math student and computer science teaching fellow at Harvard, and an avid GTD'er! I even listen to the GTD podcast. For a summer course in natural language processing, I'm writing a short study about using AI techniques that, given an English-language task such as "mow the lawn", will label it with a user's context, goal, and other metadata.
For this program to work it needs training data, i.e. lots and lots of examples of correctly labeled tasks. In fact with this type of AI, the biggest factor in accuracy is the size of the training dataset! My own system has a record of 'only' a few thousand tasks. I'm looking for folks who would volunteer to add tasks to the dataset.
The only information the software needs in order to train are the text of to-do items (i.e. "shave the goat") and their labels (context: at home, project: goat hair maintenance, ...). The training data does not need reference material, phone numbers, email addresses, etc.
If you'd be interested to be a part of this study and can contribute tasks in CSV or Excel format, that would be much appreciated! If the prediction algorithm works, I'd be keen to roll it into a GTD app that predicts your classifications as you type. If it sweetens the pie, I'd be happy to give contributors free copies/web access to the software!
I understand people may have personal information in their systems and privacy is important. So if you are more comfortable sending only a portion of your task list that would still help! In fact, less personal tasks would probably train the AI better anyway.
Thanks for bearing with this long post. I figure it may raise more questions than it answers. If you are interested in participating and/or have any questions please feel free to email me at beb259@g.harvard.edu, or post here and I'll respond. My project is due in two weeks so I will respond very fast to any and all questions!
For this program to work it needs training data, i.e. lots and lots of examples of correctly labeled tasks. In fact with this type of AI, the biggest factor in accuracy is the size of the training dataset! My own system has a record of 'only' a few thousand tasks. I'm looking for folks who would volunteer to add tasks to the dataset.
The only information the software needs in order to train are the text of to-do items (i.e. "shave the goat") and their labels (context: at home, project: goat hair maintenance, ...). The training data does not need reference material, phone numbers, email addresses, etc.
If you'd be interested to be a part of this study and can contribute tasks in CSV or Excel format, that would be much appreciated! If the prediction algorithm works, I'd be keen to roll it into a GTD app that predicts your classifications as you type. If it sweetens the pie, I'd be happy to give contributors free copies/web access to the software!
I understand people may have personal information in their systems and privacy is important. So if you are more comfortable sending only a portion of your task list that would still help! In fact, less personal tasks would probably train the AI better anyway.
Thanks for bearing with this long post. I figure it may raise more questions than it answers. If you are interested in participating and/or have any questions please feel free to email me at beb259@g.harvard.edu, or post here and I'll respond. My project is due in two weeks so I will respond very fast to any and all questions!