# a Dynamic Bayesian Network Click Model for web search ranking

One of the important techniques in web-search ranking is Click modelling, which aims to estimate the probability of click to benefit a wide range of search–related application. Click modeling is a challenging task as it affected by different factors including position bias and hidden variables. Four hypotheses are common in the click modeling which are examination hypothesis, cascade hypothesis, rationality hypothesis and positional rationality hypothesis. Based on these hypotheses many models have proposed to understand user click behaviour, but, the most common models in click modeling are cascade model and examination model. In this paper, the author proposed Dynamic Bayesian Network (DBN) model to compute the value of the perceived relevance ai and user satisfaction si assuming that users scan the links top to bottom and if they satisfy with a document they will not examine the next document. This model is similar to hidden Markov model in that there is a conditional dependency between the probability of examining a URL and the probability of examining the next URL, but it differs in that the probability of examining the next URL depends on the observation and hidden variables of the current documents.

# Another models :

## Examination Hypothesis:

This hypothesis assumes that if a URL clicked by a user that means it’s both examined and relevant to the query . In another word , given a query q , position i and URL u the probability of examining a URL is  : [math]P(C=1|u, p)=P(C=1|u,E=1)P(E=1|p)[/math]

Many models have proposed based on the examination hypothesis including the Clicks Over Expected Clicks (COEC) model , the Examination model, and the Logistic model.

## The cascade model:

This model is based on the assumption that the user scans the results top to bottom, and the click depends on the relevance of all the results of a query. This model simply aggregate the click and skip of each documents in the results. [math]P(C_i=1)=r_i\prod_{j=1}^i-1(1-r_j)[/math] where [math]r_i[/math] the probability of click in position i and [math](1-r_j)[/math] is the probability of a document being skipped