Collections

Collections enhance how the agents interact with organizational knowledge. By grouping related sources into topic-based repositories, Collections enable more precise, context-aware information retrieval and intelligent knowledge management.

Feature overview

• Better Information Retrieval: Better information retrieval by organizing knowledge sources into topic-specific repositories, allowing the system to search within relevant content clusters rather than across the entire knowledge base. This significantly reduces search space and improves retrieval speed.

• More accurate answers: By organizing related sources together, collections create focused knowledge domains that reduce noise and interference from unrelated content.

• Contextual understanding: Goes beyond simple keyword matching, enabling the system to deeply comprehend the nuances and intent behind user queries. By analyzing not just the immediate question, but also considering previous user interactions and configurable contextual parameters, the AI can retrieve information that precisely addresses the underlying need.

• Search Types: Determine how the system will process queries across your sources. Choose between Semantic, Full Text, or Hybrid approaches. Your selection will apply whenever the system accesses this collection to retrieve information, ensuring that all queries follow your preferred search methodology.

Each collection is a standalone entity housing knowledge sources relevant to its theme. The search functionality allows you to find information within a specific collection, limiting results to only the sources contained within. Each collection needs to be trained individually, and the system maintains previous versions for reference.

Creating a New Collection

To create a new collection, go to the Collections section and click New Collection.

Search Types

When configuring the collection, you can select the search type that best fits your needs through a dropdown menu. Collections offers three distinct search approaches, each with its own strengths: Semantic, Full-text and Hybrid.

• Semantic Search

Retrieves results based on the meaning of the text rather than exact word matches. This type of search understands the intent behind queries and can find relevant information even when the exact terminology differs.

Example: If a user asks "How do I reset my password?", Semantic Search would also return sources containing phrases like "password restoration procedure" or "how to change forgotten login credentials" because it understands these concepts are semantically related.

• Full Text Search

Focuses on exact word matches within sources. It excels at finding specific terminology, product names, or unique identifiers.

Example: If searching for "Model X500 error code 3021", Full Text Search would be ideal as it would specifically look for sources containing these exact terms, ensuring technical specificity.

• Hybrid Search

Combines both Semantic and Full Text. This approach produces more relevant results, especially in cases where relying on a single approach may not be sufficient.

If you choose "Hybrid," you'll be prompted to input percentage values to determine how much each search type contributes to the final results.

Example: For a query like "smartphone battery draining quickly" (e.g., Semantic: 60%, Full Text: 40%) a Hybrid Search might:

• Use Semantic Search (60%) to understand concepts related to battery conservation and power management

• Use Full Text Search (40%) to ensure specific terms like "smartphone" and "battery" appear in the results

This combination ensures you get sources that both specifically mention the key terms and understand the underlying issue of power consumption, even if they don't use the exact phrase "draining quickly."

When to Use Each Search Type

• Choose Semantic Search when: Your content contains various ways of expressing similar concepts, or when users might ask questions using different terminology than what appears in your sources.

• Choose Full Text Search when: Precise terminology is critical, such as with product codes, specific error messages, or technical documentation where exact wording matters.

• Choose Hybrid Search when: Your knowledge base contains both technical specifications and conceptual information, or when you want to balance finding exact matches with understanding user intent.

Advanced Settings

You can now move on to customize each collection with advanced settings to fine-tune how information is retrieved using the following parameters:

• Top K: Determines how many paragraphs the system should consider when searching for information. With a low K value, the model selects among the most likely options, which makes the results presented more focused and relevant, increasing perceived accuracy.

• Similarity Threshold: Establishes the minimum level of relevance required for search results to be considered relevant. A lower score can lead to more unpredictable results, offering broader insights. Higher scores yield more precise matches.

• Previous user inputs: Customize the search based on the user's previous inputs. The default value is 0, meaning only the current message is considered. Move the cursor to include previous messages. This allows to go back in the conversation and comprehend the context when a previously discussed topic is revisited. For example, if a user asks about "unlock new credit card" and then follows up with "how do I do it?", the system can understand that the follow-up question refers to "unlock new credit card".

Training

Now, you have to make sure that they are now part of the knowledge base. You do that by using the training process. Click on the button Training on the top right corner. Train each collection individually to keep the agent updated with the latest information. Train the affected Collections after these actions to ensure the knowledge base is up-to-date.

You can train using the icon available in the list or via Training section.

The training process of a PDF may take a little longer than training TXT files. If you face any issues with specific files in cases of very large trainings (which may involve too many documents or PDFs close to the 100-page limit), wait a few moments and try training again.

Delete Collection

When deleting collections, the system maintains careful handling of associated content. Questions and sources within the deleted collection remain accessible, ensuring no critical information is lost.