Lexical search matches and ranks documents by the terms or attributes specified in a query. It is a practical starting point for a search engine: an explicit query can retrieve matching indexed content without a user profile or a model that understands the domain. Matching is only the first step; ranking determines which matches appear first. section 7.4.3 section 1.2.3 section 3.0
An inverted index maps each term in document fields to a postings list of documents containing it. Query terms provide fast lookups; Boolean operations can combine matching document sets, while positional matching can account for term sequences. One way to picture the process is as sparse query and document vectors with a dimension for each indexed term: present terms have a value of 1, absent terms 0. A relevance score ranks the matches; BM25 is one lexical ranking approach. section 2.1.3 section 2.3 section 7.1
A keyword match is not necessarily the right answer
For a reviews search, `bbq near charlotte` returned just one matching review: it contained all three words, but described a festival rather than a barbecue restaurant. Removing `near` produced more results, yet still missed relevant places whose reviews did not say `charlotte`. The example exposes two separate problems: a word can make a query too restrictive, and matching words need not capture intent. section 7.2
Listing 4.3 Running a search on the product catalog
def product_search_request(query, param_overrides={}):
request = {"query": query,
"query_fields": ["name", "manufacturer", "long_description"],
"return_fields": ["upc", "name", "manufacturer",
"short_description", "score"],
"limit": 5,
"order_by": [("score", "desc"), ("upc", "asc")]}
return request | param_overrides
query = "ipod"
request = product_search_request(query)
response = products_collection.search(**request)
display_product_search(query, response["docs"])
This Python search request selects product fields to query, returns a score, and sorts five matches by that score. It provides a small starting point for testing a lexical baseline. section 4.1.2
Exact terms are useful when the query names content directly, but they cannot by themselves establish meaning. A query for `soda` cannot retrieve related drinks if that term is absent from the index; `What is DEET?` can rank unrelated questions highly because their wording overlaps. Improve a baseline by testing real queries, tuning fields and analysis, or adding domain-aware query interpretation and expansion. Keep keyword matching as a fallback when richer interpretation cannot resolve a query. section 2.3 section 13.4.1 section 1.3 section 2.5.2
