How will Discover work
Summary
Discover feature leverages heatmap and recordings data to generate insights. If the users manually tagged behavior in recordings or in highlights, that is also used to complete the summary.
Implementation Approach Discover: Recordings
Discover: Recordings feature will be available to get concise summary of recordings. There will be a new icon for Discover with each recording. In case, the user has annotated the recording then the insights will also add information on the basis of that. As of now, users can tag these keywords - Bug, Confusion, Frustration, UI/UX Issue, Conversion
When the user clicks on Discover for any recording, user will see a summary as below. Note that this will be available only for the recordings which the system identifies as having Moderate or High relevance.
Implementation Approach Discover: Heatmap
Hotjar Discover: Heatmap feature will be available to get concise summary of Heatmaps. There will be a new icon for Discover with each Heatmap. This option will be available after we have at least 500 views of a page. And this will keep getting updated after every 100 new views or a design refresh. This is to ensure that we have sufficient data to make any conclusions from heatmaps.
The summary data here will show which elements on a page attract users’ attention(move map data), rage clicks and top clicks on the page
This summary also includes data from Highlights, Rage Clicks and general analysis of Heatmap
User Flow
Assist icon will be present on the left side on the dashboard.
Clicking on the same will open a ChatGPT like interface. Users can select either Support or FAQ context and then ask the queries.
Flow for Assist
Here’s the user flow of how this will be used
Implementation Approach
Assist will be developed on top of GPT.
What is GPT:
GPT stands for Generative Pre-trained Transformer. It is a neural network machine learning model which is trained using data on the internet to generate any type of text. This sophisticated neural network is used to train large language models (LLMs) to simulate human communication. GPT-4 is OpenAI’s new technology that provides the most advanced system, producing much safer and more useful responses.
We can use OpenAI’s APIs to built this feature.
- Get OpenAI Keys
- Prepare FAQ data in a format that can be used to train GPT
{ "role": "user", "content": "How to Install Your Hotjar Tracking Code?" }, { "role": "assistant", "content": "The Hotjar Tracking Code is a JavaScript snippet that is unique to each site you add to Hotjar. You'll find the Tracking Code on your Sites & Organizations page by clicking on Install tracking code. To begin collecting data, you'll first need to install this Tracking Code on your live site. To get started with this, choose from one of the following four installation methods: Installation when your website uses Google Tag Manager Installation when your website uses WordPress Installation using other platform-specific instructions Manually adding the code to your website" }
{"prompt": "<prompt text>", "completion": "<desired generated text>"} {"prompt": "<prompt text>", "completion": "<desired generated text>"} {"prompt": "<prompt text>", "completion": "<desired generated text>"} ...
- Uploaded created dataset on the OpenAI server.
- Train the GPT-4 model(text-davinci-002) for the custom dataset uploaded
- Evaluate the model
- Repeat the process until we get the answers as needed
- Use the OpenAPI Chat Completion API to answer queries from customers
