Search for Similar Items with a Research Template

Overview

Improve efficiency and specificity in your research tasks with Athena Intelligence's built-in Research Templates. These templates offer structured workflows for streamlined data exploration and analysis.

Setting Up a Research Template

You can use built-in Research Templates to improve efficiency and specificity of research tasks. In this example we'll be using a Research Template "Find Similar Items":

Define Subject and Sample Item

Scroll to the bottom of the research template to find the boilerplate structure for Task 1. This structure will help you organize your task. Fill out the following sections in the template:

Sample item details: Provide comprehensive details about the sample item within the <sample_item> tags.
Output format: If you want to change the default output format, modify it within the <output_format> tags to suit your needs.
Preferred sources: If you have specific sources for your search, like academic websites or news outlets, mention them within the <preferred_sources> tags.

Scroll to the bottom of the template to find boilerplate structure for Task 1. Fill it out.

Introduce the research subject and sample item
Provide the sample item details in <sample_item> tags
If you want to change the default output format for items Athena will find, you can change the format inside <output_format>
If you have preferred sources for search like websites publishing academic papers or news outlets, specify them with <preferred_sources> tags

Set Clear Research Subjects

Clearly define the research subjects and sample items. Detail specifics to help Athena tailor searches accurately.

Additionally, you can edit the task template structure to provide step-by-step instructions. This could include:

Analyzing the sample item
Searching for potential similar items using provided search strategies
Comparing each potential item to the sample to assess similarity
Compiling the five most similar items
Formatting and finalizing the output

Change the Output Format structure if you want a different breakdown of each item.

Start Research Task

After you have defined the template, start the research task by clicking on Chat in the top right corner. A new chat with Athena will open. If you have a side-by-side view, Athena can access the contents of the opened report, and you can ask it to carry out Task 1.

We have our first results!

For workspaces with custom default models, Side Chat might not come eqiupped Search & Browse.

To execute the research workflow:

go to Conversations,
choose GPT4 Turbo model,
turn on Search & Browse tool,
attach the Research Template report and send a message.

Two of the items are relevant.

Improve Search Iterations

To improve the next search iteration, we'll:

Create Task 1 Results section and add relevant results there
Create Task 1 Additional Instructions section and add task-specific instructions: don't show studies, don't show items present in Task 1 Results.

As you can see, Athena listened to additional instructions and narrowed the search down to programs and policies this time, not studies.

Note that Large Learning Models are stochastic by nature, so output might vary from time to time, especially when dynamic search is involved.

You can iterate on the specific instructions after each iteration to help Athena zero in on the desired outcome.

Depending on the research workflow, you might want to copy the original template and start a new one to change the search template structure and algorithm more substantially.

Bonus: Chain Tasks

After you finished with one task, it might make sense to kick off a second one in the same document. For our use case, once we got a list of relevant traffic congestion programs, we might want to look into congestion studies. We already have a couple of example results, since Athena provided some instead of congestion programs and policies/

We'll add Task 2 structure below Task 1, and ask Athena to carry it out:

Having two tasks in a single research document might be beneficial when topics are connected.

Note that after a significant amount of research, Athena's performance may decline due to context overload. It's generally advisable to switch to a new research report and side conversation after 2-3 tasks or 20+ iterations of searches