Implementation of Semantic Search Based on Vector Database for Personal Documents

Finding personal documents on our own devices can be a frustrating task. Especially when we forget the exact label of our document, we will only rely on our memory to remember the context of the document and then guess where the document is located. Windows solves this problem by searching for keywords and metadata of a document, whereas MacOS offers a Spotlight feature to read the file content. Both options have limitations that require users to use specific and simple phrases or keywords without any additional semantic content, such as meaning that can be derived from the document. To overcome these challenges, we offer a novel document search system that captures the document's meaning and matches each feature in the document with each query feature. We also compare our system with existing personal document search engines such as Mac OS Spotlight and Windows Search. We find that our approach results in slower running time but high accuracy.