LIBS 894: INFORMATION RETRIEVAL

SYSTEM
Individual Assignment
Topic: Three Classic Models in
Information Retrieval System
Submitted by: Rebecca
Course Code: LIBS 894
Lecturer: Aminu Musa
Date: April 2025
Boolean Model

The Boolean model is one of the earliest and most fundamental models in information

retrieval (IR), developed based on the principles of Boolean algebra. In this model,

documents are represented as sets of terms, and queries are formulated using logical

operators: AND, OR, and NOT. It treats document relevance as binary—either a

document matches a query (relevant) or it does not (non-relevant) (Baeza-Yates &

Ribeiro-Neto, 2011).

Core Principles:

- AND: A document must contain all the specified terms.

- OR: A document must contain at least one of the specified terms.

- NOT: A document must not contain the specified term.

Example:

Assume a digital library contains documents on different topics. A user searches for

“education AND technology”. The Boolean model will retrieve only documents that

contain both terms, excluding any that contain only one. If the query was “education OR

technology”, it would retrieve all documents that contain either or both terms. If the user

writes “education AND NOT technology”, it returns documents that contain the word

"education" but not "technology."

Advantages:

- Simplicity: The model is easy to understand and implement.

- Efficiency: Suitable for databases with structured data and well-defined vocabularies.
- Exact matching: Allows users to precisely control the search scope using logical

operators.

Limitations:

- Lack of ranking: All matching documents are treated equally; there’s no notion of

relevance scoring.

- Rigid matching: If a document uses synonyms or alternate phrasing, it might be missed

unless explicitly included.

- User complexity: Users must understand how to construct Boolean queries effectively,

which may be challenging for novices.

Real-World Application:

Boolean retrieval is still widely used in legal databases (e.g., LexisNexis), library catalog

systems, and search interfaces in professional databases like PubMed or Scopus, where

precision and exact filtering are important (Croft, Metzler, & Strohman, 2015).

Vector Space Model

The Vector Space Model (VSM) represents documents and queries as vectors in a multi-

dimensional space where each dimension corresponds to a distinct term. Unlike the

Boolean model, it provides a graded notion of relevance by measuring the cosine

similarity between the query vector and each document vector (Salton, Wong, & Yang,

1975).

Core Principles:

- Each document and query is represented as a vector of term weights.

Frequency) are used to reflect the importance of a term within a document and across the

collection.

- Cosine similarity is calculated as:

cosine(θ) = (D · Q) / (||D|| × ||Q||)

Example:

Suppose a user enters the query "e-learning platform". Two documents are:

- D1: "E-learning has transformed education using online platforms."

- D2: "Healthcare platforms improve patient outcomes."

After preprocessing (removal of stopwords, stemming, etc.), term frequencies are

computed, and the cosine similarity is calculated between the query vector and each

document vector. D1 would likely score higher due to higher semantic overlap with the

query terms.

Advantages:

- Relevance ranking: Documents are ranked based on similarity scores, improving

retrieval quality.

- Partial matching: Even if a document does not contain all query terms, it can still be

retrieved based on similarity.

- Scalability: Effective for large-scale systems and adaptable to machine learning

enhancements.

Limitations:

- High dimensionality: Representing each term as a dimension can result in large, sparse
vectors.

- Term independence assumption: It ignores the relationships or dependencies between

terms.

- Semantic gaps: Synonyms or related concepts may not be captured unless additional

processing (e.g., Latent Semantic Indexing) is used.

Real-World Application:

The vector space model is foundational in modern search engines like Google and Bing

and is widely used in text mining, document classification, and recommender systems

(Manning, Raghavan, & Schütze, 2008).

Probabilistic Model

The Probabilistic model of information retrieval assumes that given a user query, there

exists a probability that a document is relevant. Documents are ranked according to this

estimated probability of relevance. This approach views information retrieval as a

problem of inference under uncertainty (Robertson & Sparck Jones, 1976).

Core Principles:

- Each document has a probability of being relevant to a given query.

- The system estimates this probability using term frequency and document statistics.

- The most common implementation is the Binary Independence Model (BIM).

The probability that a document is relevant (R) given a query (Q) is estimated as:

P(R|D, Q) ∝ ∏ (P(t|R) / P(t|¬R)), for each term t in Q

- P(t|R): Probability term t appears in relevant documents.

- P(t|¬R): Probability term t appears in non-relevant documents.

Example:

If a user searches for “remote work policies,” the system examines the frequency of these

terms in previously judged relevant vs. non-relevant documents and calculates the

likelihood that new documents containing similar patterns are relevant.

Advantages:

- Theoretical foundation: Based on Bayesian probability theory.

- Relevance feedback: Improves retrieval through user feedback on document relevance.

- Ranking by likelihood: More intuitive ranking of documents by probability of

relevance.

Limitations:

- Initial assumption: Requires estimation of relevance probabilities, which may not be

available at first.

- Independence assumption: Assumes terms are conditionally independent, which is not

always realistic.

- Complexity: More computationally intensive than Boolean or vector space models.

Real-World Application:

Probabilistic models underpin modern ranking systems in web search engines and are

foundational in algorithms like BM25 used in Elasticsearch, Solr, and other full-text

search libraries (Robertson & Zaragoza, 2009).

Baeza-Yates, R., & Ribeiro-Neto, B. (2011). *Modern information retrieval: The

concepts and technology behind search* (2nd ed.). Addison-Wesley.

Croft, W. B., Metzler, D., & Strohman, T. (2015). *Search engines: Information retrieval

in practice* (2nd ed.). Pearson.

Manning, C. D., Raghavan, P., & Schütze, H. (2008). *Introduction to information

retrieval*. Cambridge University Press.

Robertson, S. E., & Sparck Jones, K. (1976). Relevance weighting of search terms.

*Journal of the American Society for Information Science*, 27(3), 129–146.

https://doi.org/10.1002/asi.4630270302

Robertson, S., & Zaragoza, H. (2009). The probabilistic relevance framework: BM25 and

beyond. *Foundations and Trends in Information Retrieval*, 3(4), 333–389.

https://doi.org/10.1561/1500000019

Salton, G., Wong, A., & Yang, C. S. (1975). A vector space model for automatic

indexing. *Communications of the ACM*, 18(11), 613–620.

https://doi.org/10.1145/361219.361220