PMS Full-Text Search Engine for .NET Core

Full-Text Search Engine with no external dependencies written in C# for .NET Core.

The aim of this project is to showcase algorithms, data structures and techniques that are used to create full-text search engines.

Getting Started

On Windows:

1. Download and build code. Use the following commands:

   dotnet restore
   dotnet build

2. Open folder with binaries: bin\Debug\netcoreapp2.0

3. Start the following command. Replace DATA_PATH with a path to Datasets folder

   run_test.bat DATA_PATH

4. If everything goes well the following messages are printed:

   Log from index construction:

   dotnet Protsyk.PMS.FullText.ConsoleUtil.dll index --input "F:\Sources\FullTextSearch\Datasets"
   
   PMS Full-Text Search (c) Petro Protsyk 2017-2018
   F:\Sources\FullTextSearch\Datasets\Simple\TestFile001.txt
   F:\Sources\FullTextSearch\Datasets\Simple\TestFile002.txt
   F:\Sources\FullTextSearch\Datasets\Simple\TestFile003.txt
   Indexed documents: 3, time: 00:00:00.1010004

   Dump of the index (for each term in the dictionary - the list of all occurrences):

   dotnet Protsyk.PMS.FullText.ConsoleUtil.dll print
   
   PMS Full-Text Search (c) Petro Protsyk 2017-2018
   2017 -> [1,1,9]
   algorithms -> [1,1,19]
   and -> [1,1,20]
   apple -> [3,1,1]
   banana -> [3,1,2]
   build -> [1,1,25]
   c -> [1,1,16]
   data -> [1,1,21]
   demonstrate -> [1,1,18]
   ...

   Search with query WORD(pms):

   dotnet Protsyk.PMS.FullText.ConsoleUtil.dll search --query "WORD(pms)"
   
   {filename:"TestFile001.txt", size:"180", created:"2018-04-02T10:09:41.4208444+02:00"}
   {[1,1,1]}
   
   {filename:"TestFile002.txt", size:"29", created:"2018-04-02T10:09:41.4248447+02:00"}
   {[2,1,1]}
   
   Documents found: 2, matches: 2, time: 00:00:00.0564721

   Lookup in the dictionary using a pattern i.e. all terms matching pattern:

   dotnet Protsyk.PMS.FullText.ConsoleUtil.dll lookup --pattern "WILD(pet*)"
   petro-mariya-sophie
   Terms found: 1, time: 00:00:00.0704173
   
   dotnet Protsyk.PMS.FullText.ConsoleUtil.dll lookup --pattern "EDIT(projct, 1)"
   project
   Terms found: 1, time: 00:00:00.0847931

Query Language

• WORD(apple) - single word
• WILD(app*) - wildcard pattern
• EDIT(apple, 1) - Levenshtein (edit distance, fuzzy search)

Conjunction operators

• OR - boolean or
• AND - boolean and
• SEQ - sequence of words, phrase

Examples of queries:

• AND(WORD(apple), OR(WILD(a*), EDIT(apple, 1)))
• SEQ(WORD(hello), WORD(world))

Data Structures

• Dictionary of the persistent index is implemented using either:
  • Ternary Search Tree.
  • Finite State Transducer (FST).
• Key-value storage for document metadata is based on persistent B-Tree implementation: B-Tree.
• Packed Integers.
• Classic data structures:
  • Heap
  • Linked List
  • LRU Cache

Algorithms

• Approximate term matching is based on Levenshtein automaton.
• Query Language parser is implemented using recursive descent parser technique.
• A method for encoding/decoding occurrences uses:
  • Group VarInt encoding.
  • Delta encoding.
  • VarInt encoding.
• Classic algorithms:
  • Binary search
  • Heap sort
  • Huffman Encoding

References

• Introduction to Information Retrieval

  

• Compilers: Principles, Techniques, and Tools

  

• Information Retrieval: Implementing and Evaluating Search Engines

  

Links

• Project Website

Publishing

MacOS

dotnet publish -c Release --self-contained -r osx.10.13-x64

Windows

dotnet publish -c Release --self-contained -r win10-x64