Markov_CN_node

Backend creating random Chinese sentences based on chat history / 根据消息记录生成随机句子

---

Project maintained by cleoold Hosted on GitHub Pages — Theme by mattgraham

construction site!

Markov-Chinese

Generates random text based on the input sentences. Simple, one process, no database.

Intro

The program will start a server that only does two things:

• receive sentences to build/update models and
• output new random Chinese sentences using the Markov chain algorithm, upon calling.

It can be served as a minimalist backend of a Markov chat bot just by using the http request. Each model must have a name (string) that can be differentiated among others and the generated text will be generated using the model with this name.

Usage

Type npm run start in the project root folder to run the app. By default it runs on 127.0.0.1:6758, however this can be changed in config.ts.

The app manages multiple models. Let us assume that the name is erika.

Insert sentences / modify table

Send the following POST request to the endpoint /sentence:

POST /sentence HTTP/1.1
Host: 127.0.0.1:6758
Content-Type: application/json

{"id":"erika","message":"你好，我是马尔可夫机器人"}
// OR
{"id":"erika","message":["你好，我是马尔可夫机器人", "哈喽", "楼上是群龙王"]}

The posted json must be in a json format:

• field id: the name of the model, it cannot contain any of /\?%*:|"<>.
• field message: the message passed to the model. This field can be an array of strings, indicating that all of the content will be added.

If everything works, a folder models/erika will be created which contains the history. conversation.txt records all the messages sent, while the json files contain the actual Markov model. Also a status code 200 is returned. Otherwise codes 400 or 500 are returned, and information is logged to the console.

Generate sentence

After repeating the above steps many times, one can send a GET request to the endpoint /sentence to retrieve a sentence:

GET /sentence?id=erika&wc=20 HTTP/1.1
Host: 127.0.0.1:6758

• parameter id: specify the name of the model
• parameter wc: maximum word count of the sentence. 10 will be used if this option is not present.

If the model does not exist, a status code of 404 is returned, otherwise the result json is returned and it looks like

{
    "status": 200,
    "message": "楼上是马尔可夫机器人"
}

plus a status code of 200.

Yes, that’s all.

Scripts

• npm run start

Starts server at 127.0.0.1:6758.

• npm run dict

This command will regenerate all the dictionaries based on conversations.txt. It is good to run this script every time after a breaking update.

Thanks to

Thanks to nodejieba for providing such a wonderful word cut library.

TODO list

• Complete API readme docs.
• Improve English sentences (probably necessary).