Ranking (information retrieval)

Ranking Algorithms
– PageRank: Originated in the 1940s, developed by Wassily Leontief in economics.
– Endorsement Ranking: Developed by Charles H Hubbell in 1965, based on the importance of endorsers.
– Journal Ranking: Developed by Gabriel Pinski and Francis Narin, based on citations from important journals.
– HITS Algorithm: Developed by Jon Kleinberg, treating web pages as hubs and authorities.

Ranking Models
– Boolean Model: Fetches complete matches, does not rank documents.
– Vector Space Model: Addresses partial matches, assigns weights to index items, calculates similarity scores using cosine similarity.
– Probabilistic Model: Uses probability theory, ranks documents based on decreasing probability of relevance.

Evaluation Measures
– Precision: Measures the proportion of top-ranked results that are relevant.
– Recall: Measures the completeness of the information retrieval process.
– F1 Score: Combines precision and recall into a harmonic mean.
– Precision-Recall Curves: Plotted to evaluate ranked retrieval results.

HITS Algorithm
– HITS uses Link Analysis for analyzing page relevance.
– Works on small sets of subgraph.
– Query dependent.
– Subgraphs are ranked according to weights in hubs and authorities.
– Pages with the highest ranks are fetched and displayed.

Learning to Rank: Application of Machine Learning
– Learning to rank is an application of machine learning.
– Used for solving the ranking problem.
– Machine learning techniques are applied to rank items.
– Widely used in information retrieval.
– Learning to rank improves the accuracy of search results.