Document search

Every document in Strix has a document vector. These vectors are used in the document search functionality. At search time, the search query is converted into a vector and compared to the document vectors. The fifty closest documents to the query are returned.

These documents are the ones that are semantically close to the given vector query, as shown in the figure below. The current default number of documents that the document search returns is limited to 50, but this number will be a dynamic input instead.

KBLab's KB-SBERT is used to create the vectors and also to perform the document search. This means that the search does not look for exact matches of the query but instead finds documents that are semantically similar to the query based on vector representations.

Users can search for a word, phrase, sentence, or even a whole document. Below are some examples:

Examples

1. Word search
   Query: klimat
   Result: Documents in Swedish party programs and election manifestos that are semantically related to the word "klimat."

2. Phrase search
   Query: klimat politik
   Result: Documents in Swedish party programs and election manifestos that are semantically related to the phrase "klimat politik."

3. Sentence search
   Query: Våra barn kommer att fråga oss vad vi gjorde när vi insåg vidden av klimathot och miljöförstöring
   Result: Documents in Swedish party programs and election manifestos that are semantically similar to the sentence.

4. Document search
   Query: A full document text.
   Result: Documents with content or context that is semantically similar to the provided document.