Repository for the project of the course "Information Retrieval" of the master degree "Data Science and Scientific Computing" @UniTS

Implementation in Python of an Information Retrieval system using the Boolean Model.

The IR system is able to:

• answer boolean queries with AND, OR, and NOT. The system is also able to evaluate complex queries, even with many nested parentheses, like

  "hello OR ((how AND (are OR you) OR I AND (am AND fine) OR I) AND am AND (sleepy OR hungry) AND cold)".

  • Use the and_query function to connect all the words in your query text with ANDs.
  • Use the or_query function to connect all the words in your query text with ORs.
  • Use the not_query function to connect all the words in your query text with NOTs.
  • Use the query function to answer a query with AND, OR and NOT without parentheses.
  • Use the query_with_pars function to answer complex queries with AND, OR or NOT with parentheses, also nested.

• answer phrase queries using a positional index and also answer to queries like “$\texttt{term}_1 /k \texttt{ term}_2$”, with $k$ an integer indicating the maximum number of words that can be between $\texttt{term}_1$ and $\texttt{term}_2$

• answer wildcards queries using a permuterm index

  • Use the trailing_wildcards function to answer to trailing wildcards, like "term*".
  • Use the leading_wildcard function to answer to leading wildcards, like "*term".
  • Use the general_wildcard function to answer to general wildcards, like "C*T".
  • Use the multiple_wildcards function to answer to multiple wildcards, like "*A*T".

• perform normalization

• perform on demand spelling correction, using the edit distance

  (for time reasons keeping as correct the first character and searching only among the terms in the index that start with that character, but changing a parameter allows for a search in the entire index)

I evaluated the IR system on a set of test queries for each functionality, and in addition I checked that the results of the queries where correct using asserts.

I also implemented a way to save and load the entire index from disk, to avoid re-indexing when the program starts. To save the index I used Pickle.

The dataset I used is the "CMU Movie Summary Corpus" dataset, which is a collection of 42,306 movie plot summaries and metadata, available at the following link.