Talk to the City Extraction Steps

https://gyazo.com/9a06a82b7d3bdb2c7dc9542d6e09ad7b

This is an interesting point that differs significantly from the analytical methods used prior to the advent of LLM.

Extract 0~n elements using LLM once for raw data obtained by various means

Progress in LLM is that you can now describe what elements you want extracted in natural language rather than in a program.

Conventional methods would be, for example, "divide into words and count the number of words".

The "words" were defined in the morphological analysis engine and were not easily controllable.

This has a similar effect to changing open-ended questions to [multiple choice questions

If you strongly constrain them with prompts, you can make them choose from only the options you have chosen, just like a traditional multiple choice question.

but in many cases, the feature of "no need to make choices in advance" is useful for [exploratory analysis

If you use open-ended questions instead of multiple-choice questions at the stage of creating the questionnaire, it is often because you do not have enough knowledge about the distribution of the target data to design the options in advance.

When free-text responses are gathered, people will find it hard to read them all in a certain amount of time.

Tolerance is different for everyone, some people have a hard time with 100 cases, some can tolerate a few hundred, but 1000 cases is hard for most people.

When things get hard, people try to compress them by adding structure.

Just as it is difficult to create multiple choice questions in advance.

TTTC will assist in this process

The "classification" produced by TTTC will not have a perfect score to start with, but it will help humans design better classification criteria

---