Design and Evolution of the DaJia App Search System

The DaJia app search system is introduced, prompting readers to consider their own design ideas before comparing them with the presented solution.

Why Implement Search

Search reduces user onboarding cost and shortens the distance to order placement, improving conversion rates.

Typical Search Implementation

Search requirements include handling large data volumes, supporting text and multi‑field queries, and delivering fast responses.

Relational databases like MySQL can handle large volumes but struggle with unstructured text search, multi‑field indexing, and performance for fast responses.

Lucene‑based engines (Elasticsearch, Solr) excel because they support unstructured storage, rich text matching, and high‑performance multi‑field queries using an inverted index.

Inverted Index

An inverted index maps each term to the list of document IDs containing that term.

docid

age

gender

1

22

male

2

22

female

3

18

male

Separate inverted indexes are built for each field (e.g., age, gender), each consisting of a dictionary and posting list.

age

docid list

22

1,2

18

3

gender

docid list

male

1,3

female

2

Common Search Process

Typical steps include selecting the top‑N relevant documents, applying coarse ranking (e.g., by rating), performing fine‑grained re‑ranking for business‑specific rules, and finally returning the results.

DaJia App Search Iterations

Phase 1 – List Era

Search displays a list of services matching the query, with basic content matching, multi‑criteria sorting, and special handling for promoted or top‑ranked services.

Phase 2 – Self‑Operated Direct Booking Era

Prioritizes self‑operated and direct‑booking services, provides query suggestions, and recommends related services when no exact match is found, using multiple term dictionaries.

Phase 3 – Semi‑Automatic Era

Automates term‑dictionary maintenance by mining user search logs, reducing manual effort and the chance of errors, while also decreasing the rate of zero‑result queries.

Future Development

Plans include continuously improving accuracy by expanding the term dictionary, enlarging the product catalog (e.g., adding household‑helper data), and enhancing query parsing with part‑of‑speech analysis to avoid irrelevant results.

End :)