Post by simranratry20244 on Feb 12, 2024 19:25:04 GMT 10
To see an example of how a chatbot can respond to digressions and the problem that arises if it is not prepared for it, let's recover the one we already designed in the previous post for a florist. This is based on four nodes and one of them, the one that jumps with the #order intention, contains mandatory slots to store information about the product and the type of flower that the user is interested in (Figure 1). chat conversation Figure 1 With this chatbot, the conversation with digressions that was previously proposed develops differently, as shown in Figure 2. The assistant detects the intention in the user's first request and begins to evaluate the slots . Since it does not find the type of product in slot 1, it asks for it, saves that the user requests a bouquet in the variable $product and asks for the type of flowers. However, the client's response is not what the chatbot expected, since the entity is not found in it and, therefore, the assistant asks again for the type of flower to store the information for slot 2. Thus In this way, a loop is entered in which the user does not find the answer they want and the chatbot continues asking infinitely for the information that it needs to store
Almost certainly, this conversation would end badly, as the user would feel frustrated by an inflexible assistant who does not answer their questions. Ultimately, the client would abandon the conversation and the chatbot would not be Colombia Telemarketing Data able to fulfill its function because it was unable to respond to digressions in the dialogue. Manage digressions with Watson Assistant To solve the problem that digressions present for conversational assistants , there are different strategies, more or less complex and possible, depending on the platform used. With Watson Assistant, it is possible to manage this conversational phenomenon by configuring the nodes offered by the platform itself. First of all, we must understand that digressions can be input and output: The input digression occurs at the node to which the dialogue is redirected when the user changes the topic of conversation.
The outbound digression occurs at the node from which the dialog is redirected when the user changes the conversation topic. Giving a practical example, if a user is in a node with slots to confirm the purchase of a product and, suddenly, asks about the business hours, an exit digression occurs in the purchase node and an entry digression in the schedule node. With Watson it is also possible to decide whether, after going to the schedule node and resolving the question, return to the purchase node. Distinguishing these two types of digressions and knowing that there is the option to return to the exit node is important, since it is possible to configure each node according to interest. However, it is necessary to keep in mind that not all nodes allow digressions . For example, in nodes that do not have slots or subnodes (as in #schedule and #address in the florist example), digressions are not possible because, upon detecting that intention and entering that node, a single response to the user and then exit that node. Chat Figure 3 To see this better, Figure 3 shows how the florist chatbot's output and input digressions would work if it were designed with Watson Assistant. As you can see, all nodes accept input digressions by default (if we wanted to change it in some, it would be possible).