3 ms·

It would be cool if there were a way to see how "repeated" the content of a given comment is across similar stories so users could make their own judgement on t

by Bjartr 7y ago

It would be cool if there were a way to see how "repeated" the content of a given comment is across similar stories so users could make their own judgement on the value of novel views. Similarly, penalizing comments (display-order wise) for being "mere" repetition would be a neat way of nudging discussion towards novelty without applying the "strong" censorship of deletion.

Retra 7y ago
Novelty has diminishing returns when you don't optimize for something else; there are many more ways to be wrong than there are to be right, after all.
- clairity 7y ago
  it's useful to be reminded of where the argument stands, and to get people unfamiliar with the state of argument up to speed. a quick, factual recitation of basic arguments could be helpful, if boring, to direct the novelty toward the edges of the argument where it can hopefully push toward insightful territory? it should be ok to be wrong (if hard to identify and harder to internalize); the problem seems to be obstinacy (e.g., shouting the same things over and over).
dang 7y ago
My gut feeling is that software techniques are still too crude to capture just the kind of repetition that we'd want to downweight. But it would be interesting if that were wrong.
- spiralx 7y ago
  Sentiment analysis would be possible to run on either or both of articles and individual comments - I've done it before on a forum that had far less activity than HN does and gotten at least some usable data. You could then track individual features (e.g. "Julian Assange", "Uber" or "free speech") over time, I'd assume that both how frequently it appears in comments and how the sentiment towards it changes over time are both interesting if you were thinking of some sort of metrics. Forums are also quite good sources for intent and emotional analysis along with just sentiment, you could probably get to see what topics are (5, Flamebait) in the old Slashdot sense :) https://www.kdnuggets.com/2018/08/emotion-sentiment-analysis-practitioners-guide-nlp-5.html https://www.kdnuggets.com/2018/08/emotion-sentiment-analysis...

by Bjartr 7y ago

Retra 7y ago

Novelty has diminishing returns when you don't optimize for something else; there are many more ways to be wrong than there are to be right, after all.

clairity 7y ago
it's useful to be reminded of where the argument stands, and to get people unfamiliar with the state of argument up to speed. a quick, factual recitation of basic arguments could be helpful, if boring, to direct the novelty toward the edges of the argument where it can hopefully push toward insightful territory? it should be ok to be wrong (if hard to identify and harder to internalize); the problem seems to be obstinacy (e.g., shouting the same things over and over).

dang 7y ago

My gut feeling is that software techniques are still too crude to capture just the kind of repetition that we'd want to downweight. But it would be interesting if that were wrong.

spiralx 7y ago
Sentiment analysis would be possible to run on either or both of articles and individual comments - I've done it before on a forum that had far less activity than HN does and gotten at least some usable data. You could then track individual features (e.g. "Julian Assange", "Uber" or "free speech") over time, I'd assume that both how frequently it appears in comments and how the sentiment towards it changes over time are both interesting if you were thinking of some sort of metrics. Forums are also quite good sources for intent and emotional analysis along with just sentiment, you could probably get to see what topics are (5, Flamebait) in the old Slashdot sense :) https://www.kdnuggets.com/2018/08/emotion-sentiment-analysis-practitioners-guide-nlp-5.html https://www.kdnuggets.com/2018/08/emotion-sentiment-analysis...