Skip to content

Tests on Astro web framework

Wassim edited this page Jul 22, 2023 · 2 revisions

Astro tags classification

The goal here is to automatically identify which tags are affected by a given post

Let's consider the following list, each of them is a "tag"
[cms, css/styling, deployment, i18n, images, markdown/mdx/markdoc, routing, SSR, editor setup, preact, react, solid, svelte, tailwind, vue, starlight, typescript, vite]
and the attached files explaining the meaning of each tag.
now I'm going to post a text between triple quotes and will ask you to categorize it to one of the tags from the "tag" list, it is also possible to assign more than one tag and if not any tag applies then assign "other"
"""
The post text which tag is to be identified can go here
"""

Attached file google docs exported and uploaded as .docx Tags

Astro support post sentiment

number of topics per post

The goal here is to first identify if the user has only asked one topic or if there was a divergence to a new topic. It was not easy to find a prompt that does not create a new topic per question and it is tricky to explain the defined intention of the difference between "topic" and "point"

prompt

I would like to identify the number of topics discussed in a thread post, a thread post starts with one question, it could be that the user adds follow up clarification questions, that counts as the same topic, but sometimes, the users starts asking about a topic that is with small deviations or through a jump in reasoning, disconnected from the first question, that should count as a separate topic.

A topic can include multiple points, all points mentioned in the first question count all as only one topic, it could be a big topic.
First start by identifying a title for all points if the first question and post it as the first bullet point, very briefly without explaining any detail, just a new meaningful title.

Then only in case the conversation deviated to new points with no or very far relation to the original question, add all the points of the new question in a new bullet point as a new topic. Do add a new bullet only if there is no existing bullet similar to it, otherwise edit the existing one to cover both topics that are related to each other.

I will list the whole conversation between triple quotes, the user text is preceded by the user name then "OP" on its own line, the bots answer is preceded by kapa.ai on a line then BOT on a line.
"""
The conversation of the support thread post goes here
"""

The result is still fragmented to questions, so adding a targeted one shot example as follows

added prompt

Now here an example of topics that are too many
* Astro Build Error
* Deploying to Vercel
* Pnpm Version Requirements for Vercel
* Issues with Vercel Adapter
* Vercel's Support for Node Built-ins
* Astro's Experimental Code Component Dependency

They should be grouped in similar topics and the result should look as follows
* Astro Vercel adapter build error
* Astro's Experimental Code Component Dependency

added prompt

Brings the response to be multi staged,

  • first a list of fragmented 5 topics
  • then a consolidated list of 3 topics
  • then recognizing the one shot and add another merge in 2 topics

answer state and sentiment

The goal here is first to identify if a post has reached its objective of answering the first question, and then finally to define a positive, negative or neutral

added prompt

Then at the end, if you could add as conclusion :
* If the original question was answered or not. First in a single word "answered" or "not answered", additional brief details can follow
* Then independent from if the question is answered or not, mention if the user is overall got a benefit from the discussion, if it helped or if it rather generated frustration from not being able to get info. First mention only "positive", "negative" or "neutral", any addition brief explanation can follow.

Astro content questions answering

Clone this wiki locally