Search#
This page provides an overview of Squirro search functions and features.
Typically, search features and settings are adjusted by the Project Creator.
Overview#
Search is a core functionality of the Squirro platform that quickly retrieves and accesses relevant information across vast amounts of data, often dispersed across various systems, platforms, and formats. Manually browsing individual documents to look for specific details becomes problematic even if the information is well structured. As the volume and complexity of data continue to grow, standard search methods can become increasingly ineffective, leading to decreased productivity. By leveraging advanced technologies such as natural language processing (NLP), machine learning (ML), large language models (LLMs), and conversational AI, Squirro enables employees to quickly find the information they need, improving productivity.
Search Experience#
Not every journey is the same when interacting with a search system. Nevertheless, there are key steps in a typical user journey, from initiating a query to interpreting the results or improving the data sources.
Starting a search#
Every search begins with a few words, phrases, keywords, or questions. The more consideration you put into the words you use and how you position them, the better the system can understand your intent. Ensure the content you input is linked to the type of information you seek. Your first query does not need to be perfect, it is just the starting point of your search journey. Begin with a quick, simple query to get an initial set of relevant results, then refine and adjust it as you go. With each iteration, the system refines and improves the results.
Interpreting search results#
The top results are typically the most relevant, but they still may not be the exact ones you seek. As you scan through the result titles, extracts, and metadata, assess if you are heading in the right direction. If the list of content you get seems off, consider drastically changing your approach by using fewer keywords and more descriptive sentences, or vice versa. If you feel you are on the right track, fine-tune your query by incorporating specific keywords or phrases found in the documents that closely match your needs.
Enhancing search queries#
A first way of enhancing search queries is optimizing how employees interact with the search functionality. Creating an effective query involves more than using the right keywords. Visit the UI Search Features page to learn more about the various search techniques and the Document Relevancy page to understand how the relevancy ranking of documents works.
Additionally, project administrators can fine-tune the query term matching strategy to specify how user search terms are interpreted and translated, or customize the relevancy scoring for the project. For situations where project data contains similar content but differs in structure, see the How To Use Best-Bets Labels to Map Query Terms page.
Understanding Search#
Squirro offers out-of-the-box Cognitive Search capabilities, an advanced technology that leverages artificial intelligence and goes beyond the traditional keyword-based search. SquirroGPT combines the technologies powering Cognitive Search with a large language model (LLM), creating a retrieval augmented generation (RAG) application, and represents a significant advancement in enterprise search.
Cognitive Search#
Squirro Cognitive Search offers a traditional approach to search, powered by Squirro Insight Engine, to uncover insights in complex datasets by entering keywords or phrases into a search bar. See the Squirro Cognitive Search page for more information.
Chat with Data#
Powered by Squirro Insight Engine, Chat with Data uses a dialogue-based approach to interact with datasets. Start a discussion with the system by asking questions, writing a statement, or inputting queries in a natural and conversational tone. SquirroGPT mimics a human conversation to provide answers or ask clarifying questions, while focusing on improving user interaction through dialogue.
The Squirro prompt generation layer interprets and restates the question, aiming to condense it into a format optimized for retrieving the most relevant context pieces from various sources. The data processing layer semantically searches through the information the user has access to, retrieving contextually relevant content for the LLM to answer questions accurately. Next, if the conversation continues, the system adjusts the subsequent query by integrating the context from the previous question, thus maintaining a coherent and contextually enriched dialogue. See the SquirroGPT page for more information.
Squirro is developing AI Guardrails and Privacy layers that function as strategic frameworks guiding systems within defined boundaries, ensuring that all outputs are accurate, compliant, and aligned with the organization’s values. They mitigate potential risks and negative consequences such as errors, bias, ethical issues, and misinformation, building trust in AI-assisted decisions.
Common Search Challenges#
Irrelevant results#
Often, you can improve the results by adjusting the query formulation. See the Query Syntax page for more information. Adding more context or follow-up questions, rephrasing the initial prompt, or just telling the system that its answers are irrelevant, in the case of conversational search, helps SquirroGPT refine its understanding of your intent.
Too many results#
For large datasets, getting very specific information is difficult and might be a sign that the data enrichment layer needs additional fine-tuning or that you need a dedicated project with a more specialized categorization of your data sources. See the Labels and Communities pages for more information or contact your Project Creator.
Inconsistent results#
When search results vary in relevance or accuracy, the issue might come from ambiguous queries, incomplete metadata, or inadequate data tagging within the datasets. See the Query Syntax and Known Entity Extraction pages for more information.
Unfiltered results#
When you encounter insufficient options to refine or narrow down search results, visit the Labels and Communities pages to ensure the data enrichment layer works properly.
Outdated results#
Make sure the data ingestion and processing layers are fully functional. See the Data Loading page for more information.
Advanced Configuration#
You can fine-tune your query term matching strategy to specify how user search terms are interpreted and translated.
Field boostings, default term matching configuration, and multi-word rescoring are examples of ways to achieve this.
To learn more about tuneable parameters, see How To Handle User Query Terms Correctly.
Relevancy#
One of the best ways to create exceptional user search experiences is to customize relevancy scoring for your project.
You can adjust the default Relevancy Scoring Algorithm via domain-specific Scoring Profiles or by modifying the Search Term Matching Strategy.
For more information on relevancy, see Document Relevancy.
For a step-by-step guide to adjusting relevancy, see How to Use Scoring Profiles to Customize Document Relevancy Scoring.
For situations where project data contains similar content, but differs in structure, see How To Use Best-Bets Labels to Map Query Terms.
Query Processing#
Query processing improves a user’s search experience by providing more relevant search results.
Query-processing workflows are modified by search engineers to fine-tune the search experience for users.
To learn more, see Query Processing.
Recommendations#
Squirro can make item, entity, and label recommendations based on various types of inputs.
See Recommendations for further details.
Synonyms#
To learn how project creators can use synonyms to improve search results, see Synonyms.