Using AI to moderate

Discussion in 'NoFap Technical Support and Feedback' started by The Pleasure Delusion, Jun 10, 2020.

  1. I have studied Deep Learning and Machine Learning. With the coming of tensorflow.js, it has become very easy to integrate AI into websites.
    As a member of this profession, it is kind of my responsibility to share the wonders of my field.
    What I have in mind is that the human moderates who delete posts should save the text of those posts in an excel file. Over time they will have collected a huge dataset of spam that hits this website. Once the data is collected, a classification algorithm like Multinational Naive Bayes needs to be trained. The resulting model can then be uploaded using tensorflow.js and it will moderate posts for you.
    It is very easy.
    I've also been thinking of using data science to gain insight on patterns. Perhaps it is too soon to explain that. But just imagine, figuring out some unknown insight into addiction. The bad industry uses it. Those fighting it should use it too.
  2. dudealone

    dudealone Fapstronaut

    Wow, very impressive post. You said right that the bad industry uses it, so why shouldn't we. Did you learn ML/DL as a part of the course curriculum or online? Please tell me the resources.

  3. I wouldn't have a problem with that as long as there's an opt-out option for people who don't want their data processed in this way. Sometimes deletion has to mean deletion, or the concept becomes incoherent.
    dudealone likes this.
  4. Caligula

    Caligula Fapstronaut

    @The Pleasure Delusion You can't hope for more than a few hundred data points on this small site. Is that enough for Naive Bayes? I'm skeptical.
  5. Depending on the kind of spam that hit's this site, we can utilize public datasets. ML works fine with less data. SVM does a great job. It is deep learning that is data hungry.
    If spam isn't that frequent an issue and moderates spend more time shuffling content around, this too can be solved with classification. How effectively? Depends on data.
  6. I learned both in University and through a specialization on Coursera.
    If you are interested, Andrew Ng teaches it so nicely a high school student familiar with programming and math till partial derivatives could do it.
    Do the Machine Learning one first. You can audit it. If you like the field you can then move to deep learning. The investment will benefit you whether you pursue it as a career or not.
    dudealone likes this.
  7. Not to worry. Data laws already exist. These are valid concerns.

    Most of us don't know but we already relinquish quite a lot of data to sites like Facebook and they experiment on us or a portion of us. You can google Facebook experiments.
    Chris_Cactusblossom likes this.
  8. I'm well aware of this and make minimal use of Facebook for this reason. I also avoid Google where possible, etc.
  9. I aggree but I would probably not use naive bayes but transform the text with BERT and then input into a neural network.

    Also there is a pretty high change that a database of deleted comments already exists. So maybe you should ask Alexander Rhodes.

    Also you can data mine It is a website, so you can scrape everything. For starters you could write this webscraper. Then you could look for the top posts. Or count word occurences. Or mine association rules. (This is a bit harder).

    To mine association rules one needs to build a data base of nouns for all posts. So each entry consists of a tuple of nouns in a given post like

    (pmo, nofap, pied, de, ...)


    (pe, fetishes, porn addiction, ...)

    and then you could mine rules. A rule would be something like

    pmo, pied --> fetishes

    So this means that many posts which mention pmo and pied also mention fetishes.
    Last edited: Jun 22, 2020

Share This Page