Conversation log 2021-01-28

• Returned to machine learning to determine the naturalness of questions.
  • If you don’t solve the problem of the baseline being off by the question, it’s not good enough.
    • Observed distribution of keyword scores

What would you like to see happen in this conversation? I have temporarily turned off naturalness determination of questions by machine learning and would like to test it now that I have it back on. What type of judgment is that judgment? Binary classification as natural or not natural for the immediately preceding document/question pair, ignoring context What type of context is that context? I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice. What type of question is that? 🙁

• I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice.

• Is there anything else you can tell us about this question? Like, if you don’t ask this first, it’s not natural to ask this out of the blue.

• I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice.

• Where is this question? 🙁

• I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice.

• What does this question seem to be? 🙁

• I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice.

• Where is this question? 🙁

• I haven’t looked at the relationship between the multiple series of questions or anything like that. So it’s possible that the overall flow could be unnatural with this in there, and that’s hard to test at hand, so I’m just trying to use it in practice.

• What type of linkage is this linkage? For example, “Is there anything else?” would be funny if it came out of nowhere.

• I’ve turned back on machine learning to determine the naturalness of questions, which I had temporarily turned off, and I’d like to test that.

• Is there anything else about this decision? Nothing in particular.

• Binary classification as natural or not natural for the immediately preceding document/question pair, ignoring context

• Is there anything else about this context? This time there were many test cases for the flow behavior, so I was able to notice once the flow was broken, and I was able to fix it and confirm that it was back to normal. What type of behavior is this behavior? Now “Is there anything else?” is asked first. What type of destination is the destination? What kind of X is that X?” before asking. Is there anything else you can tell us about beyond that? If you just cut out the question and ask if it’s natural, the question is, “What kind of X?” is more unnatural than “What other kind of X?” is more unnatural than “Is there anything else? What is the type of type? NGKW Type How do you know that question? 🙁

• If you just cut out the question and ask if it’s natural, the question is, “What kind of X?” is more unnatural than “What other kind of X?” is more unnatural than “What else?

• What kind of X ka is this X ka? NGKW X.

• “What kind of X is that X?” before asking.

• Where is this destination? NGKW destination

• “What kind of X is that X?” before asking.

• What kind of X is this X? NGKW X

• Binary classification as natural or not natural for the immediately preceding document/question pair, ignoring context

• Where is this context? Flow of questions repeated

• Since there were many test cases for flow behavior this time, I was able to notice once the flow was broken and was able to fix it and confirm that it was back to normal.

• Is there anything else you can tell us about this behavior? In the end, I decided not to make a naturalness judgment on the two basic questions because I could not reconcile the comfortable order of the two questions with the fact that their naturalness as individual questions did not match the order of the two questions. How will you know the decision? The user doesn’t know how it’s determined internally. What type of user is this user? Users of this chat Is there anything else you can tell us about that user? I want to be asked a good question. What is it that you value? It would be good if the naturalness of questions could be judged to reduce the number of cases where clearly unnatural questions are asked to users, which impairs their comfort. Where is that user located? 🙁

• It would be good if the naturalness of questions could be judged to reduce the number of cases where clearly unnatural questions are asked to the user, which impairs the user’s comfort.

• What is the relationship between this question and the user? Information is drawn from the user by being asked questions. Is the user the same as the question? Is it different? No, it is not.

• It would be good if the naturalness of questions could be judged to reduce the number of cases where clearly unnatural questions are asked to the user, which impairs the user’s comfort.

• What happens to the user when this question is asked? A good question leads the user to a flow state by providing a series of moderately sized tasks. Asking bad questions confuses the user and interrupts the flow. What happens to the question when that user 🙁

• It would be good if the naturalness of questions could be judged to reduce the number of cases where clearly unnatural questions are asked to the user, which impairs the user’s comfort.

• What is the connection between this question and the user? 🙁

• In the end, I decided not to make a naturalness judgment on the two basic questions because I could not reconcile the comfortable order regarding the two basic questions with the fact that their naturalness as individual questions did not match that order.

• What is the relationship between this decision and the question? If the judgment is well done, the quality of the questions will improve. If it’s not done well, it’s worse. Is that judgment the same as the question? Is it different? No, it is not.

• If the judgment is well done, the quality of questions will improve. If it’s not done well, it will get worse.

• What happens to the question when this judgment is made? If the baseline for judging is off, you’re going to lose the diversity of questions by selecting only those questions that tend to be natural as a base. What happens to the judgment when you ask that question? 🙁

• If the baseline for judging is off, you’re going to lose the diversity of questions by selecting only questions that tend to be natural as a base.

• What is the relationship between this decision and the user? (At any rate, I was concerned that it would slow down the response time, but that’s not really harmful. The issue of the baseline being off depending on the question needs to be resolved, imho) 🙁.

---

This page is auto-translated from /nishio/会話ログ2021-01-28 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.