Conversational Training

i want to train the model based on previous conversation or from some documents that includes conversation

@ravikrcs That is definitely possible. Did you already build a bot with Rasa? If not, please have a look at our documentation Rasa Docs. Let me know if you have a concrete question I can help you with.

i already build a faqbot

its working on my website also through socketio.

Great! So what exactly is your question?

i have answers for the question i.e conversation in file and now i want to use that conversation as training data for intuitive answer. How to get that?

i mean how to get intuitive response from bot. i am using simple Utterance Templates. is there any other way i can do ?

You need to convert the conversations into the training data format Rasa needs, e.g. NLU data, stories, and the corresponding domain file.

We don’t generate any answers. Everything the bot says, needs to be defined in a template.

which one should i use, knowledge base or retrieval action.

Knowledge bases can only fill in some gaps in a template as well. E.g. when the user ask for some banks it would list the names or whatever you define (Knowledge Base Actions). Retrieval actions pick the correct template response for you when the user is asking some chitchat questions (Retrieval Actions). In both cases, we do not generate any answers. So, depending on what you mean by “intuitive response”, retrieval actions could be something you should take a closer look at.

ok, Thanks

could i train answers like questions (nlu.md)

could i train answers like questions (nlu.md)

Can you explain a bit more what exactly you mean?

i have many reports from various industries in excel sheet (link) or in database. Now, i ask bot to give me latest report on specific industry like automobile. Then, how to get most specific answer (report).

If you store the links/documents together with a category and date, I guess you can construct a query to fetch the link/document, when the user is asking something like “show me the latest automobile report”. The simplest way to do it would be to create a new custom action that has a connection to the knowledge base. You need to extract the time constraint (= latest) and the category (= automobile) from the user’s message (via the tracker in your custom actions). That should enable you to construct a query and check the knowledge base for the requested document. Does that help?

Thanks, we are proceeding in same line. if any problem , will update you

suppose i have 100 articles on bmw now i want to retrive the article on bmw with specific topic. like “report on new BMW X5 Protection VR6 which is literally bulletproof”. Now, How can i get that one specific articles?

In the end you need to query your database with a more specific query. For that you need to (1) enrich the data in the database with some metadata that describe the topic of the article and (2) extract the topic from the user message. (2) can be done via entities. E.g. mark the relevant topic in the user message (“report on new [BMW X5 Protection VR6](topic) which is literally bulletproof”) and train your model again. Does that help?

@Tanja what if i have millions of reports.

You definitely should add more trainings data that represents queries on reports. Those training data should include the topic entity, so that your action later on is able to perform the actual query to the database. You might also want to add a lookup table (Training Data Format) that includes possible topics. Does that help?