The network to analyze can be either directed or undirected. In order to= indicate that some edges are more important, an attribute for edge weight = can be used. Hence the network can be weighted or un-weighted. It produces = two measures for each node in the network, Authority and Hub score. It was = originally developed for ranking web pages on the basis of these two measur= es. But it can be used as effectively on more general problems involving ne= tworks of different kinds like citation and other scientometric, bibliometr= ic networks.

=20The algorithm requires 3 inputs, the *file* storing the informati=
on about the network, *number of iterations* the Authority and Hub v=
alues will be calculated (by default it is 20) and the column name that rep=
resents the edge weight. The iterations facilitate convergence of the autho=
rity and hub scores. But empirically it is found that authority and hub sco=
res tend to converge after 20 iterations, hence the default value. To indic=
ate appropriate relative importance of edges, edge weight attribute can be =
used. The name of the edge weight column can be selected from the drop down=
box. The column type should be either *int, real or float*. To indi=
cate un-weighted network the default option of "Unweighted" can be selected=
.

In the first read-in, the submitted file is validated against the ** (n)** in the network is passed on. This is used to initialize=
the matrices that will be used for the HITS computation. An Adjacency matr=
ix

*In the second read-in, the adjacency matrix is populated with either the=
default edge weight (1.0) or the value from the file. When the network is =
directed, the matrix element corresponding to [ALGDOC:source no=
de, target node] is populated. In case of an undirected netwo=
rk, the matrix elements corresponding to [ALGDOC:node1, node2]<=
/em> and [ALGDOC:node2, node1] are popul=
ated with the same edge weight.*

Next transpose of the adjacency matrix is created. Now the Authority and= Hub matrices are updated in the following manner,

=20- =20
- Authority Matrix =3D (Transpose of Adjacency Matrix) x (Hub Matrix)=20
- Hub Matrix =3D (Adjacency Matrix) x (Authority Matrix) =20
- Normalize Authority and Hub Matrices by dividing each authority value b= y the sum of all authority values, and dividing each hub value by the sum o= f all hub values. This is done to facilitate convergence. =20

This algorithm is run for *number of iterations* provided by the =
user.

Then the output file is created in the **NWB** format. In a=
ddition to the original file, the node section will have 2 extra attributes=
, *authority_score and hub_score* of type float. As the name suggest=
s, values from the Authority and Hub matrices corresponding to each node wi=
ll be inserted.

- =20
- Source Code: Link =20

The original HITS algorithm authored by Jon Kleinbe= rg was implemented, integrated and documented by Chintan Tank. Also tha= nks to Micah Linnemeier and Russell Duhon for providing guidance during imp= lementation. For the description I acknowledge Wikipedia.

=20- =20
- Kleinberg, Jon. Authoritative sources in a hyperlinked environment. In = Journal of ACM, pages 604-632, September, 1999. Link= =20
- Chakrabarti, Soumen., Dom, Byron., Ravi Kumar, S., Raghavan, Prabhakar.= , Rajagopalan, Sridhar., Tomkins, Andrew., Gibson, David., Kleinberg, Jon. = Mining the Web's Link Structure. In Computer, pages 60-67, August, 1999. Link =20
- Bharat, Krishna., Henzinger, Monika R. Improved algorithms for topic di= stillation in a hyperlinked environment. At Proceedings of the 21st annual = international ACM SIGIR conference on Research and development in informati= on retrieval, pages 104-111, 1998. Link =20
- Dean, Jeffrey., Henzinger, Monika. Finding Related Pages in the World W= ide Web. At Conference, pages 389-401, 1999. Link =20

The license could not be verified: License Certificate has expired!=20
Generate a Free license now.