README

A document search engine written from scratch in Python. Based on concepts from Stanford's Introduction to Information Retrieval Book.

Uses the TREC8Adhoc part of the TIPSTER collection for index building and evaluation. You'll have to obtain TREC8Adhoc.tar.bz2 from this collection (Disk 4 & 5) to reproduce the reported results.

To evaluate the results with trec_eval you will have to download the TREC-8 ad hoc qrels.

Features:

• Inverted index construction methods: Simple, Single-pass in-memory indexing (SPIMI )and Map Reduce
• Document similarity metrics: TF-IDF, BM25, BM25VA (PDF Link) and TF-IDF Cosine Distance
• Performance evaluation on TREC and result reporting in qrel format.

An early Rust port is located at ir-search-engine-rust.

Prerequisites

This project requires at least python 3.6.

Install Dependencies

pip install -r requirements.txt

Index Creation

Run

Run python cmd_index.py --help for instructions.

Example:

To create an index and document stats using the SPIMI method, run: python cmd_index.py --document_folder=./data/TREC8all/Adhoc/ --index_file=spimi.index --stats_file=spimi.stats spimi

Output

The script creates two output files:

• index_file: Inverted index
• stats_file: Document stats collected during index creation (document lengths and term counts)

Index Format

The index file format is text-based.

The first line states the number of unique documents in the index.

Each consecutive line represents a term including related data:

<TERM> <DOCUMENT_FREQUENCY> <POSTINGS>

• TERM - The term itself
• DOCUMENT_FREQUENCY - Document Frequency (Number of documents the term appears in)
• POSTINGS - A comma-separated list of documents the term appears in along with the term frequency separated by pipe in the given document: <DOCUMENT_ID>|<TERM_FREQUENCY>,<DOCUMENT_ID>|<TERM_FREQUENCY>,...
• TERM_FREQUENCY - Number of times the term appears in the corresponding document

Evaluation

Run

Run python cmd_search.py --help for instructions.

Example:

To evaluate a topic list on a previously created SPIMI index using a bm25 ranking, run: python cmd_search.py --output_file=out.txt --run_name=dev --topics_file=./data/TREC8all/topicsTREC8Adhoc.txt --index_file=spimi.index --stats_file=spimi.stats bm25

Output

The script creates an output file which can be used with trec_eval, like: trec_eval -q -m map -c ./data/TREC8all/qrels.trec8.adhoc.parts1-5 ./out.txt