Jawaharlal Nehru Technological University Hyderabad MCA 2022 INFORMATION RETRIEVAL SYSTEMS Previous Question Paper (September/October - 2022 - R20)

Information Retrieval (IR) is the process of searching for, identifying, and retrieving relevant information from a collection of unstructured data, typically text, to satisfy a user's information need. Unlike database retrieval, which deals with structured data and exact matches, IR focuses on relevance and ranked results based on queries.

The Vector Space Model (VSM) represents documents and queries as vectors in a multi-dimensional space, where each dimension corresponds to a unique term in the collection. The similarity between a document and a query is calculated using the cosine of the angle between their respective vectors.

Key steps in the VSM process:

• Term Weighting: Each term in a document is assigned a weight, typically using TF-IDF (Term Frequency-Inverse Document Frequency), which balances the frequency of a term within a document against its rarity across the entire corpus.
• Vector Construction: A document D is represented as a vector V(D) = (w₁, w₂, ..., w_n), where w_i is the weight of term i.
• Similarity Calculation: The similarity between a query vector Q and document vector D is computed as:

Example: Consider a collection with terms {apple, banana, fruit}.
Document 1 (D1): "apple fruit" -> Vector [1, 0, 1]
Document 2 (D2): "banana fruit" -> Vector [0, 1, 1]
Query (Q): "apple" -> Vector [1, 0, 0]

Calculating similarity for Q and D1:
Dot product = (1*1) + (0*0) + (0*1) = 1
Magnitudes = |Q|=1, |D1|=sqrt(1²+0²+1²) = sqrt(2)
Similarity = 1 / sqrt(2) ≈ 0.707

Calculating similarity for Q and D2:
Dot product = (1*0) + (0*1) + (0*1) = 0
Similarity = 0

The system ranks D1 higher than D2 because it shares the term 'apple' with the query.

The Probabilistic Retrieval Strategy is based on the Probability Ranking Principle, which aims to rank documents by the probability of their relevance to a user query. It assumes that the relevance of a document is independent of other documents and that there exists a subset of documents that are relevant to the query. The model typically uses the Binary Independence Retrieval (BIR) approach, where terms are binary (present or absent) and the probability of relevance is estimated using the odds of a term appearing in relevant versus non-relevant documents.

To calculate the similarity, we use the RSV (Retrieval Status Value) formula: RSV = Σ log((P(ti|R) * (1 - P(ti|NR))) / (P(ti|NR) * (1 - P(ti|R)))). Assuming a simplified model where P(ti|R) = 0.5 and P(ti|NR) = 0.01 (or using term frequency as a weight proxy), we calculate the presence of query terms {gold, silver, truck} in each document:

• D1: {gold, truck} (shipment of gold damaged in a fire - 'truck' is absent, 'gold' is present). Note: 'truck' is not in D1. Terms: {gold}.
• D2: {silver, truck} (delivery of silver arrived in a silver truck). Terms: {silver, truck}.
• D3: {gold, truck} (shipment of gold arrived in a truck). Terms: {gold, truck}.

Ranking based on the number of matching query terms:

D2 and D3 are tied for the first rank as they both contain two query terms, while D1 is ranked third.