The mobile applications and web services we use have search functions. Most are developed using the basic text search provided by databases or full-text search provided by search engines like Elasticsearch. Full-Text Search is one of the traditional methods mainly used for searching text data, focusing on finding specific keywords, words, phrases, etc., in documents, web pages, databases, and more. It typically involves inputting keywords or short sentences to search text data and finding documents that match the keywords, but it does not consider context or semantic similarity.

However, due to these characteristics, Full-Text Search has limitations. You've probably had the experience of wanting to find something but struggling because you didn't know the exact name. You can describe the features or what it does, but you keep hitting the search bar because you don't know the name of the target. While Full-Text Search offers synonym processing, similar word search, and typo correction features, there are limitations.

For example, let's look at the following five sentences:

• Happens maybe once in a hundred years
• A 0.001% chance
• Zero chance of success
• Not possible even if you die and come back to life
• Getting struck by lightning consecutively

All five sentences indicate "almost no possibility." However, there isn't a single common word included in all the sentences. Even with Full-Text Search that incorporates similar words and synonym processing, it's nearly impossible to search all the above sentences at once. In other words, Full-Text Search relies only on the words or phrases in the text and cannot consider semantic similarity.

Additionally, there is the limitation that the search target is confined to text. For example, when you input "puppy" as text, you cannot find "puppy" in image files that have no description. The images we find by typing a word into Naver or Google are images included in documents like blogs or news articles where that word is mentioned.

As machine learning technology advances, it has become possible to convert various data such as images, videos, audio, and documents (words, paragraphs) into the same form called vector embeddings. By measuring the distances between these vectors, it's now possible to assess the similarity between the original data, leading to a new search technology called vector search.

Vector Embedding - Representing All Data

Vector search overcomes the limitations of Full-Text Search and allows searching based on meaning. If we represent sentences expressing the meaning of "almost no possibility" as vectors, they are located close to each other. Assuming they are expressed in two-dimensional vectors, it would look like this:

Using vector search offers several advantages:

• Consideration of Semantic Similarity: Full-Text Search relies on the exact match of the search term and does not consider context. However, vector search represents data in vector space, considering semantic similarity. This allows you to find not only documents containing the search term but also semantically similar documents even if they don't contain the search term.
• Easy Synonym Handling: For example, in Full-Text Search, to search for documents containing "vehicle" with the search term "car," you need separate synonym processing. In contrast, vector search naturally handles synonyms because it represents the meaning of words as vectors.
• Searching Various Data Types: Full-Text Search applies only to text data and is difficult to apply to unstructured data like images, audio, and video. Vector search can process different data types by converting them into vector embeddings, making it effective for searching and analyzing different kinds of data.
• Handling Complex Data: Modern data is very complex and has multidimensional characteristics. Full-Text Search finds it difficult to handle such multidimensional data, but vector search represents multidimensional data as vectors, easily handling complex data structures.
• Improved User Experience Through Enhanced Search Quality: Since vector search considers semantic similarity, customers can quickly find information that fits their intent even if they don't input exact search terms.

In summary, vector search is a powerful tool that overcomes the limitations of Full-Text Search and provides better search results by considering the meaning and similarity of data. It is effective in handling various data types and complex data structures, helping customers find information more accurately and quickly. It is an essential technology in modern data analysis and search systems.