Should you use ChatGPT or DialogFlow in a Shopping Centre / Mall?
Shopping Centre / Mall Chatbots can use ChatGPT or DialogFlow to deliver a better user experience when a Shopper engages with your Chatbot. But there is a slight difference between the two technologies.
When developing a Chatbot for our clients, we typically opt for a hybrid model of Chatbot. We offer the shopper a choice of menu items to click on, like a menu on a website. This can lead them down a structured path to find details on shops, promotions, events and so on. But often they are impatient and simply ask the Chatbot a question. When this happens, the Chatbot needs to call on some Artificial Intelligence to figure out what the shopper is asking and how to reply.
At this point, the Chatbot can call on ChatGPT or DialogFlow – or both.
But first a quick discussion on what is ChatGPT and what is DialogFlow.
ChatGPT is a highly advanced language model developed by OpenAI, capable of understanding and generating human-like text. It is an abbreviation for “Chat Generative Pretrained Transformer”. And the operative word here is “pre-trained”. It has been trained on a large corpus of text data.
The training process involved predicting the next word in a sequence of text, given the previous words, using an attention mechanism. This approach allowed the model to learn the context-dependent relationships between words and generate more coherent and semantically meaningful text. In short, it is ready to start chatting with shoppers. But the answers will not always be quite what they are expecting. For this reason, ChatGPT can be fine-tuned.
During fine-tuning, the model is further trained on a smaller, task-specific dataset. The goal of fine-tuning is to make the pre-trained model perform better on the specific task by learning task-specific patterns and features from the smaller dataset. In other words, we can upload a set of questions and associated answers into the model and it will use them to reply more accurately to your shoppers.
But ChatGPT doesn’t keep a history of previous chats to use for training purposes. It simply generates text based on what your shoppers ask – but it doesn’t retain those questions, which is important for making the model more intelligent.
DialogFlow, on the other hand, is a conversational platform developed by Google that provides a platform to build conversational interfaces, such as chatbots, voice apps, and messaging apps. It uses natural language processing (NLP) to understand user input and can respond in a human-like manner.
DialogFlow is the complete opposite of ChatGPT because when it starts it doesn’t know anything. Which isn’t altogether a bad thing. In this way, it offers 100% control of how it replies to shoppers. DialogFlow also keeps a history of what questions it is asked which can be used for training.
Typically when deploying a Chatbot integrated with DialogFlow, we add a small set of phrases to give the Chatbot at least some personality. So when a Shopper says ‘hello’ or “good morning’ it can reply with a personalised greeting.
Each industry is different but in the Shopping Centre / Mall vertical users ask pretty much the same type of thing and it is these learnings that we can add on top of the initial small set of phrases, to hit the ground running with a Chatbot fit for purpose. Using user utterances to train the Chatbot will make it more intelligent and offer Shoppers a better experience.
ChatGPT can be used to make DialogFlow more intelligent. By running user utterances through ChatGPT to source pre-trained replies, and adding the replies to DialogFlow the training phase can be accelerated.
Another useful deployment of ChatGPT is to stand in when DialogFlow doesn’t know the answer to a user utterance. It does offer a second shot at finding an answer but it does present the risk of engaging in off-topic conversation with Shoppers which detracts from its primary purpose.
And not least of all is a question of cost. DialogFlow offers a free and paid version depending on the features required and the usage. Using the free version has served us very well until now.
ChatGPT offers the GPT-3 model which can be accessed through their API. They offer a small credit based on the usage of their API after which it is a paid service. Words and parts of words are broken down into tokens and you are charged based on the number of tokens used.
CentreChat.com is a Content Management System integrated into the popular Chatbot Development platform, ManyChat.com. It helps Shopping Centre / Mall marketers to update content on the fly which appears in Facebook Messenger, WhatsApp, Instagram and Telegram. www.centrechat.com