In our Explained series, we demystify the intricate technologies that power our AI, Aplysia. In this article, we explore behaviour analysis, its role within the HiJiffy solution, and how it is leveraged for maximum impact.
Understanding behaviour analysis means focusing on how people act and interact. By analysing data, we spot patterns and identify unusual behaviours. Here’s how we apply this in our solution:
- Tone Classification: We categorise the user’s communication style to understand their tone better.
- FAQs Insights: We gain valuable insights into frequently asked questions to improve response strategies.
Tone Classification
Let’s use Instagram as an example here. On Instagram Stories, most replies are emojis or brief comments, but requests and inquiries are common, too. By identifying the purpose of each message, we can respond effectively.
Understanding a sentence involves more than just its structure; context, tone, and cultural norms are crucial. We use tone classification to interpret these complexities, ensuring clear communication.
In general, sentences can be divided into four categories:
- Declarative: Making a statement or expressing an opinion. They provide information and end with a period.
- “I booked a room for three nights.”
- “The view from my window is breathtaking.”
- “The staff was very nice.”
- Interrogative: Asking a question and usually ending with a question mark. In English, they often begin with question words such as who, what, where, when, why, and how, or with auxiliary verbs like do, does, did, is, are, and can.
- “What time is check-out?”
- “Do you have any available rooms for tonight?”
- “Is there a gym in the hotel?”
- Imperative: Giving a command, making a request, or offering an invitation. They usually end with a period but can also end with an exclamation mark for emphasis. The subject is often implied to be “you.”
- “Please send extra towels to my room.”
- “Turn down the air conditioning.”
- “Arrange a taxi for 8 AM.”
- Exclamatory: Expressing strong emotion or excitement and typically ending with an exclamation mark.
- “This hotel is amazing!”
- “I can’t believe how comfortable the beds are!”
- “What a fantastic experience!”
Our goal is to identify requests and questions to be forwarded to Aplysia. To achieve this, we focus on imperative, declarative (including exclamatory), and interrogative sentence types.
To achieve this, we employ vector embeddings, which are high-dimensional representations of text that capture the semantic meaning and context of words and phrases. These embeddings allow us to perform nuanced intent classification by analysing the underlying meaning of sentences rather than just their surface structure.
Our sophisticated in-house embedding model is designed to classify each new sentence by comparing it to known sentence types, ensuring that it is categorised into the closest matching type. This process aids in understanding the user’s intent more accurately and efficiently.
Figure: Vector embeddings. Sentences that are similar end up closer together in vector space, making it easy to compare them with new ones.
FAQs Insights
Every chatbot response offers users a chance to provide feedback with a thumbs up or down. Users provide feedback by selecting either 👍 or 👎.
Negative feedback highlights where improvements are needed. It’s categorised into:
- “No information”: Questions that the chatbot was unable to answer due to a lack of information on its knowledge base.
- “User feedback”: Descriptions of issues written by users.
Feedback insights help hoteliers quickly identify the most urgent information to address, preventing similar complaints from arising. By focusing on the topics that frequently lead to negative feedback, staff can make improvements that benefit both the hotel and its guests.
Each question comes with related topics used to generate the answer, pinpointing where information might be missing. We’ve listed these questions and their topics and included a direct link to help you improve the related FAQ.
Learn more about our AI
If you are interested in learning more about various technologies used in Aplysia, explore a section of our website dedicated to our artificial intelligence, follow HiJiffy on LinkedIn and subscribe to our newsletter in the footer.
More articles in the Explained series
Sources
This article is based on technical contributions by Eduardo Machado and Vanda Azevedo from HiJiffy’s AI Team.
