Interactions
Any call from a client (device/browser/app/system) with one of our APIs is called an Interaction.
An Interaction is typically initiated by a User (a person) but it can also be initiated by the browser, or even from code in an autonomous system (without any active end-user intervention at all).
An Interaction is uniquely identified internally with a GUID, which is not exposed through the API output but is attached to the immutable recording of API calls in the Application Logs.
Conversational AI Cloud logs these Interaction types:
- Events
- Article (Q&A) questions
- FAQ clicks
- FAQ Searches
- autoDialogs
- Dialogs
- T-Dialogs
- Feedback
- Link Clicks
- DataRetrievals
Articles (Q&A)
Questions are interactions started by the client containing a textual input.
This input may be typed, spoken, a click, anything that converts a user’s action into an input string requiring an answer from Conversational AI Cloud’s knowledge base.
A question will typically result in a response as an answer to that question.
Requests to the ask endpoint are treated as article interactions unless there is a parameter -- present in the request or set in the Conversational AI Cloud engine - to treat these requests as dialogs or t-dialogs.
FAQ Clicks
FAQs can be shown and retrieved from Conversational AI Cloud’s APIs (see data retrieval below). A click on an FAQ (or another form of interaction to retrieve the answer to an FAQ) will get the response.
FAQ search and FAQ clicks are both defined as separate interaction types in the Interaction Log.
Requests to the faq/{faqId}/output endpoint are logged as FAQClicks.
FAQ Search
With Conversational AI Cloud it is possible to retrieve a list of FAQs through a text-matching algorithm derived from Conversational AI Cloud's NLP, instead of showing a single Answer (as in a article interaction) or retrieving an FAQ through its categorization (one form of DataRetrieval).
Requests to the faqs endpoint that include a searchText are logged as FAQ Searches.
Dialogs
Conversational AI Cloud can give a ‘normal’ answer to a question, or an answer which is a dialog start, i.e., an answer attached to a (set of) follow-up questions.
After the activation of a dialog, the state (step) within the dialog for that client is stored to interpret the next input of the client in the proper context of the dialog flow.
In the Interaction Log, a specific field defines when a dialog was activated.
Dialog Node and Links
A Dialog consists of several Dialog Nodes. Each Node is a specific step in the dialog, (picture the boxes in a flow chart). A Dialog typically starts with a question. A client can respond to that question and Conversational AI Cloud’s reaction to the client’s response is defined in the following Dialog Node (i.e. the following box in the flow chart).
Each Dialog Node holds one or more questions as examples of the expected user’s response to the previous step in the Dialog, as well as one or more answers (given specific context). Dialog
Nodes are connected by defining the Dialog Links to form a Dialog. Each Dialog Link connects two Dialog Nodes to define a next step (like a parent and child) to thus form a flow. In the Interaction Log, Dialog nodes are clearly referenced with their respective ID.
Dialog Option
To guide the client, it’s possible to define pre-set responses as part of the answer that a user can just click instead of typing. These are the Dialog Options. In the Interaction Log, the Dialog Options are part of the Response text. A user clicking each step in a dialog is a separate Interaction like a question, yet within a dialog context. Dialog Options are a feature to make it easy for a client to click instead of type, and are thus likewise an Interaction.
Steps in the Dialogs are clearly defined in the Interaction Log with a separate interaction type. Requests to the ask endpoint are treated as Dialog interactions if they explicitly include a session.dialogPath parameter or where this parameter is active in the session in Conversational AI Cloud because the previous interaction activated a dialog.
Transactional Dialogs
Transactional dialogs (t-dialogs) are like dialogs, but with only one flow of questions and responses, with no branching. Typically, they are used to perform some sort of a transaction against an API. They consist of various slots (similar to nodes) and one end-slot where the transaction is handled.
T-dialogs can be started by articles, events and dialogs.
Like with dialogs, each interaction of a user to follow the flow of such a t-dialog, is part of the t-dialog interactions.
Requests to the "ask endpoint" are treated as t-dialog interactions if they explicitly include a "tdialogstate" parameter.
Feedback
Feedback can be given by users to indicate whether they were happy with the response they received.
Feedback is part of the user journey and Conversational AI Cloud has tools to respond, escalate or continue the conversation (where it currently mostly stops at that point).
Link Click
Any Link click that is part of a response from Conversational AI Cloud is also an interaction and logged as such.
Strictly speaking, logging these clicks is tracking. Yet, since these are such an important part of a customer journey, they are presented as an integral part of the Interaction log.
Link clicks need to be explicitly logged. Any request to the linkclick endpoint is logged as a linkclick interaction.
DataRetrieval
In many user interfaces, additional data are shown alongside an input field to ask a question.
Examples are Classifications (e.g. content category structures), the retrieval of related FAQs or the calls to retrieve content from the knowledge base (outside of the CMS forms, directly calling the data APIs) for external BI or syncing with other systems.
All of this additional data helps to build a rich user experience.
Data retrieval interactions happen alongside other interactions of other types to retrieve FAQs but can also be initiated deliberately. The interaction logs generally do not show these interactions, but they are used to calculate, among other metrics, how often FAQs are shown.
