Semantic Search
Semantic search is a method that lets users search data based on the meaning of their query. Users can use any form of natural human language and still get the same results. This differs from classic search, which relies on exact keyword matching.
For instance, if you’ve got a collection of documents about Quentin Tarantino, Maradona, and Elon Musk, the user doesn’t need to type the full term “Quentin Tarantino” to find documents about him. Instead, they can enter phrases like “a movie director” or “who made Pulp Fiction” and still reach the same results. This method is best to use when you need to provide a more intuitive and user-friendly search experience, handle queries with ambiguity, or when you want to provide more relevant search results based on the context of the query.
Syntax
You can perform semantic searches on virtually any text (string) field of any object. Here’s how:
Usage
Example 1 - Google Docs
Example 2 - Discord messages
Example 3 - Any text block
Arguments
| Name | Type | Default | Description |
|---|---|---|---|
| concepts | string[] | Specify what concepts to look into | |
| moveTo | object | Move your search closer to another vector described by keywords | |
| moveAwayFrom | object | Move your search further away to another vector described by keywords | |
| targets | object | Subsearch list | |
| targetVectors | string[] | Target vectors | |
| distance | float | The required degree of similarity between an object’s characteristics and the provided filter values | |
| certainty | float | Normalized distance between the result item and the search vector |
Semantic Search: Continued
In this section we’re going to explore more sophisticated functionalities to enhance the search capabilities. Let’s dive in, folks!
We have already explained vector space in the previous doc. nearVector provides direct access to the vector space, allowing for precise navigation and customized search experiences. Here, we’ll explore advanced techniques and use cases to fully leverage this powerful feature.
Advanced Usage
Fine-Tuning with moveTo and moveAwayFrom
moveTo and moveAwayFrom are very powerful advanced functionalities that enable you to steer your search query towards or away from specific concepts. You can probably sense how powerful this concept is which gives you precise control over your search results.
Example: Balancing Concepts
For example, let’s’ consider you’re searching for articles related to web development, but you want a balanced view, considering both frontend and backend perspectives. The example given below imitates the scenario.
Utilizing Custom Vectors
If you have access to custom words or document vectors, nearVector allows you to integrate them directly into your search queries.
Example: Search near a Vector
Optimizing Search with Boosting and Filtering
Let us now have a look at how you can enhance your search results by boosting certain properties and applying filters to hone in on the most relevant data.
Property Boosting
Some of the properties in the data are more relevant than others. You can boost the relevance of specific properties in your data to ensure they have a greater impact on the search results.
Example: Boosting Titles
Conditional Filtering
You can also apply filters to your semantic search queries. It helps to narrow down results based on specific conditions.
Example: Filtering by Tags