Unstructured data is free-form information that does not follow a fixed schema or fit neatly into table columns. It commonly centers on text, while also carrying dates, numbers, or facts within documents such as emails, social posts, photos, and other content. section 1.2.1 section 7.1.1
A search application treats unstructured data differently from exact-value fields. Rather than merely deciding whether a document matches, full-text search aims to return the documents most relevant to a user’s words and assigns scores based on how well they match. section 8.9 section 10.0
Searching a quotation
Given the quote below, searches for “imagination,” “knowledge,” or combinations such as “knowledge important” can return it. Default search behavior may miss related wording such as “passion,” misspellings such as “knowlege,” and abbreviations; analyzers can later be adjusted for synonyms, stemming, spelling mistakes, and similar variations. section 7.1.1
"quote": "Imagination is more important than knowledge"
Use this short text document to see why a full-text query can match its words yet fail to recognize alternate wording without additional analysis. section 7.1.1
For Elasticsearch search design, unstructured data usually implies these choices:
- Use full-text queries rather than term-level queries when searching text fields expected to receive relevance scores. section 8.9
- Build requests with Query DSL when constructing simple or advanced search criteria; it supports leaf and compound queries. section 8.9
- Plan for real user input, including synonyms, acronyms, emojis, and lingo, instead of assuming every query repeats document wording exactly. section 7.1.1
- Combine a transactional database with a search engine when transactional storage and search or analytics have distinct roles. section 1.2.3
Search engines are designed for unstructured text analysis, high query loads, and large amounts of text data. They are not a universal replacement for every storage system: large binary content such as videos or images may call for a dedicated binary store, and the right choice depends on the use case. section 1.2.3 section 1.4.5
