JAWAHARLAL NEHRU TECHNOLOGICAL UNIVERSITY HYDERABAD B.Tech IV INFORMATION RETRIEVAL SYSTEMS Previous Question Paper (October/November - 2020 - R15)

The Binary Independence Model (BIM) is a fundamental probabilistic retrieval strategy that estimates the probability that a document is relevant to a query based on the presence or absence of terms. It assumes that terms are distributed independently within the relevant and non-relevant document sets.

The model calculates the probability of relevance for a document d given a query q using the Odds ratio. The ranking is based on the Retrieval Status Value (RSV), defined as:

Where P(t|R) is the probability that term t is present in a relevant document, and P(t|NR) is the probability that term t is present in a non-relevant document. Documents are ranked in descending order of their RSV, which effectively measures the weight of evidence provided by each query term.

Parameter estimation in probabilistic models involves determining the values for P(t|R) and P(t|NR). Since relevance information is often unavailable during initial retrieval, estimation techniques are required.

• Initial Estimation: If no relevance data exists, P(t|R) is often set to a constant (e.g., 0.5), assuming all terms are equally likely to be in relevant documents. P(t|NR) is estimated as df / N, where df is the document frequency and N is the total number of documents.
• Relevance Feedback: Once a user provides feedback on initial results, the parameters are updated using the known relevant (R) and non-relevant (NR) sets. The estimate becomes P(t|R) = (r + 0.5) / (R + 1) and P(t|NR) = (df - r + 0.5) / (N - R + 1), where r is the number of relevant documents containing the term.
• Smoothing: The addition of 0.5 (the constant 0.5 is used to avoid zero probabilities) acts as a smoothing factor, ensuring that the model remains robust even when certain terms do not appear in the feedback set.