rumors line bot
2024
Line bot that checks if a message contains internet rumor.
This is a one of the sub-project of 真的假的。
This state diagram describes how the LINE bot talks to users:
Developing rumors-line-bot requires you to finish the following settings.
After cloning this repository & cd into project directory, then install the dependencies.
$ git clone --recursive [email protected]:cofacts/rumors-line-bot.git # --recursive for the submodules
$ cd rumors-line-bot
Please follow all the steps in LINE official tutorial.
Create .env file from .env.sample template, at least fill in:
API_URL=https://dev-api.cofacts.tw/graphql
LINE_CHANNEL_SECRET=
LINE_CHANNEL_TOKEN=
LINE_LOGIN_CHANNEL_ID=
LIFF_URL=
Other customizable env vars are:
REDIS_URL: If not given, redis://127.0.0.1:6379 is used.PORT: Which port the line bot server will listen at.GTM_ID: Google Tag Manager ID. For the events and variables we push to dataLayer, see "Google Tag Manager" section below.DEBUG_LIFF: Disables external browser check in LIFF. Useful when debugging LIFF in external browser. Don't enable this on production.RUMORS_LINE_BOT_URL: Server public url which is used to generate tutorial image urls and auth callback url of LINE Notify.You will need Node.JS 16+ to proceed.
$ npm i
Spin up peripherals like Redis and MongoDB using:
$ docker-compose up -d
Then spin up the application, including chatbot server and webpack-dev-server for LIFF, using:
$ npm run dev
The server will be started on localhost:5001 (or the PORT you specified in your .env file.)
If you wish to stop the peripherals, run docker-compose stop.
Just run npm test. It will automatically spin up the aforementioned docker and run unit tests.
We recommend using ngrok to create a public address that directs the traffic from LINE server to your local machine. With ngrok in your path, just
$ ngrok http 5001
ngrok will give you a public URL. Use this to set the webhook URL of your Channel (See the section "Channel Console" in LINE official tutorial).
We recommend using ngrok configuration file to setup a tunnel with a fixed subdomain. In this way the public URL can be fixed (means no repeatitive copy-pasting to LINE Channel settings!) as long as the subdomain is not occupied by others.
Inside LINE Developers console in your Message API channel, under Messaging API > Webhook settings set the Webhook URL to ${ngrok_url}/callback and turn on Use webhook. Click verify to confirm it is successfully connected to your local machine.
We are using LIFF to collect user's reason when submitting article & negative feedbacks.
If you don't need to develop LIFF, you can directly use LIFF_URL provided in .env.sample, which links to staging LIFF site.
If you want to modify LIFF, you may need to follow these steps:
To create LIFF apps, please follow instructions under official document, which involves
chat_message.write in scope (for LIFF to send messages)
After acquiring LIFF URL, place it in .env as LIFF_URL.Endpoint URL to start with your chabbot endpoint, and add /liff/index.html as postfix.To develop LIFF, after npm run dev, it is accessible under /liff/index.html of dev server (http://localhost:5001) or production chatbot server.
In development mode, it spins a webpack-dev-server on localhost: (default to 8080),
and /liff of chatbot server proxies all requests to the webpack-dev-server.
A tip to develop LIFF in browser is:
https:///liff/index.html?p=&... in desktop browser.liff.logout() manually in JS console to trigger a re-login.liff.init() would still work in desktop browser, so that the app renders, enabling us to debug web layouts on desktop.
liff.sendMessages() would not work, though.
liff.closeWindow() will not work either if your browser window has gone through login redirects.
The LINE bot server starts a GraphQL server that stiches Cofacts GraphQL API and API specific to the LINE chatbot.
Whenever Cofacts API updates, use npm run cofactsapi to fetch the latest Cofacts API schema.
During development, use the following command to start a storybook on your local machine:
npm run storybook # Then visit http://localhost:6006
You can also visit https://cofacts.github.io/rumors-line-bot for pre-built storybook on master branch.
On production, LIFF files are compiled to /liff directory and served as static files by the chatbot server.
If you get 400 bad request in LIFF, please search for liff.init function call in compiled JS binary and see
if LIFF ID is consistent with your LIFF URL, which should be the path without leading https://liff.line.me/.
The LIFF ID is set using Webpack Define plugin during build, thus swapping LIFF URL env variable without rebuilding the LIFF binaries will cause 400 bad request.
We use ttag to support build-time i18n for the chatbot.
Please refer to ttag documentation for annotating strings to translate.
To extract annotated strings to translation files, use:
$ npm run i18n:extract
The translation files are located under i18n/, in Gettext PO format.
en_US.po: Since the language used in code is already English, this empty translation file exists to simplify settings.zh_TW.po: Traditional Chinese translation.ja.po: Japanese translation.You can replace this with any language you want to support, by leveraging Gettext msginit command.
You will need to change i18n:extract and i18n:validate script in package.json to reflect the locale change.
By default, the chatbot will be built under en_US locale.
On Heroku, please set LOCALE to one of en_US, zh_TW or any other language code that exists under i18n/ directory.
If you want to build using docker instead, you may need to modify Dockerfile to include the desired LOCALE.
Prerequisites :
To use push message :
in .env file, sets NOTIFY_METHOD=PUSH_MESSAGE
To use LINE Notify :
Callback Url : RUMORS_LINE_BOT_URL/authcallback/line_notify.env file, sets
LINE_NOTIFY_CLIENT_ID=
LINE_NOTIFY_CLIENT_SECRET=
NOTIFY_METHOD=LINE_NOTIFY
RUMORS_LINE_BOT_URL=
LINE_FRIEND_URL=https://line.me/R/ti/p/
You can set up a setting page entry point(LIFF_URL?p=setting) in account manager -> rich menu
$ npm run notify
$ node build/scripts/scanRepliesAndNotify.js
rumors-line-bot uses Google cloud services that is authenticated and authorized using Google Cloud service accounts and Application Default Credentials.
Please create a service account under the project, download its key and use GOOGLE_APPLICATION_CREDENTIALS env var to
provide the path to your downloaded service account key. See documentation for detail.
We use Dialogflow to detect if user is trying to chit-chat. If user input matches any of the Dialogflow intents, we can directly return predefined responses in that intent.
To use Dialogflow, please do the following setup:
dialogflow.sessions.detectIntent permission.DAILOGFLOW_LANGUAGE : Empty to agent's default language, or you can specify a language.DAILOGFLOW_ENV : Default to draft agent, or you can create different versions.Create a custom (user scope) dimemsion for Message Source, and a custom (hit scope) metrix for Group Members Count. Both of them default index is 1. If the indexes GA created are not 1, find cd1 and cm1 in the code and change them to cd$theIndexGACreated and cm$theIndexGACreated respectively.
Use npm run typecheck to check types; use npm run typegen to generate type from GraphQL schema.
Prepare .env file (which should be identical to your deployment environment) and run docker build . to generate docker image.
.env will be copied over to the builder image to generate LIFF static file with the env.
When building image, you can just include the "Build-time variables" (denoted in .env.sample) in .env to ensure that no server credentials are leaked in the built client code.
Since built docker images will encode public URLs into statically built files, these build-time variables when we run the image as a container. Therefore, each separate deployment environment will require a separate build of the image.
You can test the built image locally using the docker-compose.yml; just uncomment the line bot section and provide the built image name.
For production, please see rumors-deploy for sample docker-coompose.yml that runs such image.
We push variables and events in Google Tag Manager's dataLayer when the user interacts with LIFF.
You can prepare the following setup in .env file:
GTM_ID: Google Tag Manager Container ID (GTM-XXXXXXX)The application will fire the following custom events in GTM dataLayer:
dataLoaded - when data is loaded in article, comment or feedback LIFF.routeChangeComplete - when LIFF is loaded or changes path.feedbackVote - when the user submits a feedback.
chooseArticle - when the user chooses an article in Articles LIFF.Also, it will push the following custom variable to dataLayer;
pagePath - Set when routeChangeComplete event fires. The page path from LIFF's router.userId - Set after LIFF gets ID token and decodes LINE user ID inside.articleId and replyId: set on Article, Comment and Feedback onMount() lifecycle is called. Or when chooseArticle event is fired.doc - Set when dataLoaded event fires. The loaded content itself in object (article in Article LIFF, comment in Comment LIFF and feedback in feedback LIFF).Sent event format: Event category / Event action / Event label
We use dimension Message Source (Custom Dimemsion1) to classify different event sources
user for 1 on 1 messagesroom | group for group messagesUserInput / MessageType / UserInput / ArticleSearch / ArticleFoundArticle / Search / UserInput / ArticleSearch / ArticleNotFoundUserInput / ArticleSearch / ArticleFoundButNoHitUserInput / IsForwarded / Yes | NoUserInput / IsCooccurrence / Yes | NoUserInput / ChatWithBot / Article / Selected / Reply / Search / Article / NoReply / Reply / Selected / Reply / Type / UserInput / Feedback-Vote / / /feedback/yes or /feedback/no is also sent.Article / Create / YesArticle / Create / NoArticle / ProvidingReason / /reason is also sent./articles is sentutm_source=rumors-line-bot&utm_medium=richmenuutm_source=rumors-line-bot&utm_medium=pushLIFF / ChooseArticle / utm_source, utm_medium also applies./setting is sentutm_source=rumors-line-bot&utm_medium=reply-request&utm_source=rumors-line-bot&utm_medium=tutorialTutorial / Step / ON_BOARDINGTutorial / Step / RICH_MENUTutorial / Step / Group / Join / 1 (Event category / Event action / Event value)Group Members Count (Custom Metric1) to record group members count when chatbot joined.Group / Leave / -1 (Event category / Event action / Event value)Note:
- We set ga event value 1 as join, -1 as leave. To know total groups count chatbot currently joined, you can directly see the total event value (Details see Implicit Count).
- To know a group is currently joined or leaved, you should find the last
JoinorLeaveaction of theClient Id.- Also, you should find the last
Joinaction of theClient Idto get a more accurateGroup Members Count.Group Members Countis only recorded when chatbot joined group, to know the exact count, you should directly get it from line messaging-api.
UserInput / ArticleSearch / ArticleFoundArticle / Search / Article / Selected / Reply / Selected / UserInput / Intro /ContentProxy / Forward / LICENSE defines the license agreement for the source code in this repository.
LEGAL.md is the user agreement for Cofacts website users.
