An e ective graphic interface is a key tool to improve the fruition of the results retrieved by an Information Retrieval (IR) system. In this work, we describe a two-dimensional interface that represents the documents ranked on a Cartesian space and allows the user to interact with the documents in order to improve the results of the search engine. Results are classi ed and ranked according to the best separating line of the two classes of documents: relevant and non relevant documents. Mathematical tools such as least squares distances are used to train the supervised algorithm that nds the separating and ranking lines.
Visualization is the process of transforming data,
information, and knowledge into graphic presentations to support
tasks such as data analysis and information exploration. The
de nition of a spatial structure for information visualization
is challenging because data in an information space may be
multi-faceted, relationships of data are interwoven and are
complicated. Moreover, the de nition of such a space means
a complex process of extracting displayable attributes from
objects, organizing the information, projecting objects onto
This is an extended abstract of [
The introduction of visualization environments may add
cognitive processes to the user who needs to understand and
learn the characteristics of the new environment and
interact with them to get the best from the system. In fact,
the aim of visualization environments, as external
representation of the world of interest, is to reduce the amount of
cognitive e ort required to solve informationally equivalent
problems [
In this paper, we present the design and implementation a
tool for the visualization of Nave Bayes (NB) probabilistic
models for information access components that represents
digital objects on the two-dimensional space [
The model which upholds the visualization tool de nes a direct relationship between the probability of an object given a category of interest and a point on a two-dimensional space. In this light, it is possible to graph entire collections of objects on a Cartesian plane, and to design algorithms that categorize and retrieve documents directly on this twodimensional representation. This tool demonstrates to be a valid visualization tool also for understanding the relationships between categories of objects.
The design of the two-dimensional visualization tool follows two main requirements: for end-user, the interface should give the opportunity to de ne the query with simple or advanced options, and to express judgements for the documents retrieved which will be used to re-rank documents; for researchers, the interface should display the decisions taken by the search engine in terms of separating line and explain how the relevance feedback given by the user a ects the list of ranked documents.
The interface o ers the possibility to write free text queries, as any other search engine, or load prede ned queries; prede ned queries are used for research purposes and recreates the environment of evaluation tasks organized by campaign such as TREC1 or CLEF2.
The interface associate each document of the collection to a point in the two-dimensional space according to a probabilistic algorithm: the abscissa re ects how much the document is relevant to the query, the ordinate re ects how much the document is not relevant to the query. The pair of numbers gives an indication of the fraction of relevance for that particular document given the query, this pair is plotted on a frame and the relative position of this point with respect to the other documents in the collection determines its position in the list of ranked documents.
In the two-dimensional representation of documents, the
equation of the ranking or the classi cation function has to
be written in such a way that each coordinate of a document
is the sum of two addends: a variable component P(djci),
the probability of a document d given a category of interest
ci, and a constant component P(ci), the prior of the category
of interest ci [
X{i(zd) log (P(djci)) + log (P(ci)) > log (P(djci)) + log (P(ci)) | } | } Yi{(zd) When the inequality holds, the document is considered an element of category ci. If ci and ci are considered respectively the set of relevant documents and the set of non relevant documents, we can divide the collection of documents in these two sets; if we are only interested in the ranking of documents, we can compute the list of retrieved documents by combining the two components into one relevance weight.
Documents can be classi ed or ranked di erently
according to the Focused Angular Region algorithm which
computes the best separating (or ranking) line by means of
regression techniques and least squares orthogonal, and
vertical, distances. Information about the categories of
documents are collected during the interaction of the user with
the interface; in particular, the relevance judgements that
the user expresses for the documents are used to re-compute
the probabilities and train the algorithm (details of this
supervised algorithm are given in [
This visualization tool was tested on standard benchmark
collections and a demonstration was presented at [
During the interaction with the system, new questions and new research ideas were collected about advances types of interaction: changing the estimated probability of terms directly; smoothing parameters in order to see how the clouds of points move in the space and how the performance changes accordingly; drawing the clouds of points incrementally, highlighting the contribution of each term to understand which terms better discriminate the two sets of points.
In Figure 1, a screen-shot of the main window of the visualization tool si shown. The example shows the interface used by researchers. The di erent separating lines are calculated for a blind relevance feedback of 10 documents: the category of relevant documents in blue, the category of non relevant documents in red, the best separating line in purple. The list of retrieved documents is presented on the right. The user can choose to select a document, read it, and judge it as relevant or non relevant. This information is stored and used to train the supervised algorithm when the user selects the \update search" box. 4.