Do you really need a chatbot? Chatbots can be implemented in many different ways depending on what they should do. Take a look at natural language processing (NLP) tasks as sentiment analysis, semantic similarity search or text classification as they may be more suitable for the particular problem you are solving.
The chatbot dilemma
At Peltarion we receive numerous questions regarding chatbots and conversational interfaces. Here we have put together a quick checklist to look if your current need for conversational interaction is pointing more towards a more discrete behind-the-scenes AI-based solution or a chatbot.
Is your challenge either of the following?
- Coping with the increasing demand for customer support or customer service
- Answering frequently asked questions in a more efficient way
- Finding the most human-like bot because your users prefer to talk to people
- Taking action on some cases but not all of them that concern the business operations
If you answered yes to any of the above then it is worth considering the following alternatives, as you might be able to solve your challenge without using a bot.
- Increasing demand for customer support/service: If your customer service department is constantly growing to cope with demand, you can easily automate the most simple and repetitive tasks. One way is to use text classification on incoming requests. This not only reduces the response times to your customers but also will free-up your agent’s time and allow them to focus on the most complex and difficult inquiries.
- Intelligent FAQ answering: Chatbots sometimes are not the answer to overlooking the importance of an effective search. Powerful semantic search, leveraged by the latest natural language processing (NLP) models such as BERT, goes beyond keyword search and finds answers to the query solely based on its meaning. The advantage is that once the model has been created, it can keep improving with minimal effort compared to the effort needed on extending the intends/actions on traditional chatbots frameworks.
- The most human-like bots are humans: If I get the option to interact with a real person or a bot I undoubtedly will go for the person. If your users feel the same way, then empower your user-facing staff not only by automating the simple and repetitive tasks—as described in point 1 above—but also by providing them with the tools to help them retrieve information much faster and redirecting to them only those inquiries that are part of their area of expertise. Methods such as text classification, sentiment analysis to prioritize inquiries and similarity search can be used here. In this way, you are not only improving your first and second line support but also maintaining a very close and personalized relationship with your customer base.
- Focused on some cases but not all: Once the chatbot is out it becomes an extension of your company and your brand. The range of possible inputs becomes endless, hence making sure you have the correct response to all possible intends expressed by your users is paramount. Minimize the possibility to generate chatbot frustration that can cost you a loyal customer by embedding a solution only in the areas of your organization that need it most.
- A hybrid solution: If a conversational interaction solution fits your needs but lacks some functionality chances are that a hybrid solution may be the most appropriate. For example, NLP could help to make sense of all the domain of the dialog space for chatbot configuration, adding multilingual support missing on some frameworks or guide the customer towards a product or service. Backend services can provide added functionality while maintaining the existing user interaction.
02/ More on data science here
Building a Stack Overflow question tagging model with public BigQuery data
In this blog post, I'll show how you can shape one of these public datasets, the Stack Overflow posts dataset, to train a question tagging BERT model.
Connecting the dots in Neural Networks
Image similarity with deep learning explained
How can a neural network deal with a vague concept like similarity? Why do we measure angles with the cosine similarity? Is it that far from classification?
Peltarion on Microsoft's AI Show
Create a text dataset from voice recordings