Configuration files

The different programs (with the exception of EWC3) receive, as an argument, an XML file containing the different sets of parameters to use with each algorithm. In this page, we describe how to configure an algorithm.

Algorithms

An algorithm is defined using the algorithm tag, and is composed of two fields: the name field, which just contains the identifier of the algorithm, and the parameters of the algorithm (if any). A partial example is shown next:

<algorithms>
  <algorithm>
    <name>QLJM</name>
    <params>
      ...
    </params>
   </algorithm>
</algorithms>

Name

The name allows the program to know which algorithm you want to configure. Depending on the algorithm, there are some fixed values in the code. We show a relation of them here:

Algorithm	Basic	Without term discrimination	Without length normalization
Random	Random
Popularity	Popularity
Most Common Neighbors	MCN
Jaccard	Jaccard
Adamic-Adar	Adamic
Cosine similarity	Cosine
BIR	BIR
BM25	BM25	BM25 No Term Discrimination	BM25 No Length Normalization
Extreme BM25	EBM25	EBM25 No Term Discrimination	EBM25 No Length Normalization
Pivoted normalization VSM	Pivoted VSM	Pivoted VSM No Term Discrimination	Pivoted VSM No Length Normalization
Query Likelihood Dirichlet	QLD	QLD No Term Discrimination	QLD No Length Normalization
Query Likelihood Jelinek-Mercer	QLJM	QLJM No Term Discrimination	QLJM No Length Normalization
Query Likelihood Laplace	QLL
PL2	PL2
DLH	DLH
DPH	DPH
DFRee	DFRee
DFReeKLIM	DFReeKLIM

Parameters

The parameters, represented with the params tag, indicate the parameter combinations which have to be executed in the program. The parameters element must contain the different parameters, represented by the param tag.

Each individual parameter is defined by three fields: its name (field name), its type (field type), and the set of values it can take (field values). Next, we can see an example:

<params>
  <param>
    <name>lambda</name>
    <type>Double</type>
    <values>
      ...
    </values>
  </param>
  <param>
    ...
  </param>
</params>

The name of the parameter depends on the algorithm. We include in the repository two files named conf\fullgrid-dir.xml and conf\fullgrid-undir.xml which contain a configuration example for each algorithm (for directed/undirected graphs), so the names of the parameters are easy to obtain.

The type determines which variable type will be used to define the parameters. We provide the following types:

Type	Description
Integer	The values will be taken as int values
Long	The values will be taken as long values
Double	The values will be taken as double values
Boolean	The values will be taken as boolean values. If the set value is "true", it will be taken as the true boolean value. Otherwise, it will be taken as false.
String	The values will be taken as they are (as strings)
LinkOrientation	The values represent the selection of edges for a user. It can take three values: IN for the incoming neighborhood, OUT for the outgoing neighborhood and UND for the union of both.

Finally, for the values, there are two possible ways to describe them: indicating the individual values, as it is shown in the following example:

<values>
  <value>0.1</value>
  <value>0.2</value>
  ...
  <value>1.0</value>
</values>

or using ranges, as it is shown next:

<values>
  <range>
    <start>0.1</start>
    <end>1.0</end>
    <step>0.1</step>
  </range>
</values>

where start indicates the first value, end the last one (both included), and step indicates the difference between the values in the range.

Both ranges and individual values can be combined at will, even in the same parameter. For example:

<values>
  <value>1.0</value>
  <range>
    <start>0.1</start>
    <end>0.999</end>
    <step>0.1</step>
  </range>
</values>

Configuration parameters

Next, we include a table including the different parameter configurations we have selected for the experiments in our paper. We include these configurations in the conf folder. The notation followed is the same as the one indicated in the previous formulas. In addition, similarly to [1], we choose, for the directed graphs, different orientations for the target and candidate users' neighborhoods (the incoming, outgoing or undireted neighborhood of the users). In BM25 and EBM25 we also select a different neighborhood for the length.

Algorithm (and variants)	Parameters
BM25
EBM25
QLD
QLJM
QLL
PL2
Pivoted normalization