NEXT uses a cutting-edge similarity search algorithm in the highlights library and for setting the focus for your AI threads. It's designed to help you quickly find the most relevant highlights and context to your phrases. This guide assists you in composing search phrases that give the most accurate results, ensuring you get the most out of NEXT's AI capabilities.
How NEXT AI's Similarity Search Works
NEXT uses vector search technology to compare the semantic meaning of your search to the highlights that live in your library. This process involves converting text into vectors (numerical representations) that represent the meaning of words, phrases, and overall context. We then calculate a "similarity score" between your search and the highlights in your library, returning results with the highest scores and rank them according to similarity.
What Makes for Good and Bad Search Phrases
Getting the best possible results from your searches means that your search phrases need to match the highlights you're looking for. If the meaning of your search is too different from any of the highlights you've collected, you will not get any (good) results.
First, let's dive into what actually makes for a "good" or "bad" search phrase.
Good Search phrases:
Comprehensive and Contextual: Good phrases are descriptive, containing multiple relevant keywords that cover the broad themes or specific details you're interested in.
Focus on Specific Topics or Issues: They directly address the particular aspects or features of interest, making them more likely to match closely with relevant documents.
Bad Search Phrases:
Too Generic or Vague: Lack of specificity can lead to a wide array of results with low relevance.
Overly Complicated Phrases: Complex phrases that don't align well with natural language or common terminology used in the documents can result in poor matches.
What does this mean in practice?
Now that we know the theory, let's dive into a couple of real-life examples. To help you on your way, let's assume we had a few discovery calls to get users' feedback for NEXT, uploaded them to NEXT and want to find back relevant highlights. With that context in mind, here are a couple of concrete example of "good" and "bad" search phrases.
Great Search Phrases:
"Feedback on NEXT's transcription quality, accuracy and speed for text processing improvement needs."
"User interface improvements for tagging highlights in recordings and the recordings library."
"Strategies for getting high quality answers from NEXT's AI chat by writing better prompts and setting a clear focus for the thread"
Bad Search phrases:
"NEXT feedback and features."
"What are the problems with NEXT?"
"User issues and NEXT improvements."
Note: as you can see, the more specific your search is, the better the results will be. Generic searches might work, but you will likely get less or less relevant results.
Top 5 Recommendations to Improve Search Phrases
Be Descriptive But Concise: Include key themes or specific features you're interested in, but avoid overly complicated language.
Use Relevant Keywords: Identify the most important terms related to your query and make sure they are included.
Specify the Context: If your search is about a particular aspect, such as AI reliability or user interface design, clearly mention these topics.
Adapt Based on Results: If your initial query doesn't yield expected results, refine it by adding more specific terms or altering phrasing based on the content of found highlights.
Leverage Synonyms and Related Terms: Broaden your query's reach by including synonyms or related phrases that might be used in the document library.
Important:
If you don't get the expected results, refine your search by focusing on the core issue or feature you're interested in. Use synonyms or related terms to capture the variety of ways an idea can be expressed. Remember, effectiveness in search phrases comes with practice and adjustment based on the responses you receive.