186 points by ml_tutor 1 year ago flag hide 11 comments
username1 1 year ago next
Interesting topic! I wonder what kind of ML algorithms are most effective for this task.
username2 1 year ago next
I would think natural language processing models, like LSTM or BERT, would be well-suited for detecting fake reviews.
username3 1 year ago next
Yes, those would be great starting points. But wouldn't we also need labeled data to train the model? How do we go about obtaining that?
username4 1 year ago next
We can either collect fake reviews ourselves (e.g. by creating them or hiring writers) and label them as fake or collect real reviews and hire human annotators to label them as fake or real. The second option is more challenging but also more scalable.
username5 1 year ago prev next
What about features like the number of reviews from the same user or the numerical rating they give? Can we use those to help detect fake reviews?
username6 1 year ago next
Absolutely, those could be used as features for a traditional machine learning model like a decision tree or a random forest. However, for deep learning models, those features might not be necessary since they can learn to extract relevant information from the raw text itself.
username7 1 year ago prev next
I'm curious how well these models would perform in the real world. Wouldn't it be easy for fake review writers to adapt and overcome the model's defenses?
username8 1 year ago next
Yes, that's definitely a concern. One possibility would be to continuously re-train the model on new data and adapt to changing patterns of fake reviews. Another possibility would be to use online learning algorithms that can update the model incrementally as new data comes in.
username9 1 year ago prev next
Or we could target the review writers directly and use honeypots or other deception techniques to catch them in the act. That way we wouldn't need to rely on detecting the reviews themselves.
username10 1 year ago prev next
I wonder whether there's been any research on the ethical implications of using ML algorithms to detect fake reviews. For example, could this be used as a tool for censorship or surveillance?
username11 1 year ago next
Great question. I'm not aware of any specific research on that topic, but it's certainly worth considering. As with any technology, there's a potential for misuse and we need to be mindful of that. But I also think there's a lot of potential for positivi